init

.gitignore

@@ -0,0 +1,8 @@
+
+.DS_Store
+gen.bat
+gen_ref.bat
+train.bat
+__pycache__
+*/**/__pycache__
+logs

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2022 Rinon Gal, Yuval Alaluf, Yuval Atzmon, Or Patashnik and contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -0,0 +1,124 @@
+# "Dreambooth" on Stable Diffusion
+
+## Notes by Joe Penna
+### **WARNINGS!**
+- Unfreezing the model takes a lot of juice.
+  - ~~You're gonna need an A6000 / A40 / A100 (or similar top-of-the-line thousands-of-dollars GPU).~~
+  - You can now run this on a GPU with 24GB of VRAM (e.g. 3090). Training will be slower, and you'll need to be sure this is the *only* program running.
+  - If, like myself, you don't happen to own one of those, I'm including a Jupyter notebook here to help you run it on a rented cloud computing platform. 
+  - It's currently tailored to [runpod.io](https://runpod.io?ref=n8yfwyum), but can work on vast.ai / etc.
+  
+- This implementation does not fully implement Google's ideas on how to preserve the latent space.
+
+  - Most images that are similar to what you're training will be shifted towards that.
+  - e.g. If you're training a person, all people will look like you. If you're trianing an object, anything in that class will look like your object.
+
+- There doesn't seem to be an easy way to train two subjects consecutively. You will end up with an 11-12GB.
+  - I'm currently testing ways of compressing that down to ~2GB.
+  
+- You might have better luck training with `sd-v1-4-full-ema.ckpt`
+  - However, it's huge and it's annoying.
+  
+# RunPod Instructions
+- Sign up for RunPod. Feel free to use my [referral link here](https://runpod.io?ref=n8yfwyum), so that I don't have to pay for it (but you do).
+- Click **Deploy** on either `SECURE CLOUD` or `COMMUNITY CLOUD`
+- Click `Select` on a GPU with at least 35 GB of VRAM (e.g. A100, A40, A6000, etc)
+- Select a template > `Runpod / Stable Diffusion`
+- Click `Connect` and choose `Jupyter Lab`
+- Make a new notebook (it's just like Google Colab) and run the code below
+```python
+!git clone https://github.com/JoePenna/Dreambooth-Stable-Diffusion/
+```
+- With the file navigator on the left, `/workspace/Dreambooth-Stable-Diffusion/dreambooth_runpod_joepenna.ipynb` -- follow the instructions in there.
+
+# Original Readme from XavierXiao
+
+This is an implementtaion of Google's [Dreambooth](https://arxiv.org/abs/2208.12242) with [Stable Diffusion](https://github.com/CompVis/stable-diffusion). The original Dreambooth is based on [Imagen](https://imagen.research.google/) text-to-image model. However, neither the model nor the pre-trained weights of Imagen is available. To enable people to fine-tune a text-to-image model with a few examples, I implemented the idea of Dreambooth on Stable diffusion.
+
+This code repository is based on that of [Textual Inversion](https://github.com/rinongal/textual_inversion). Note that Textual Inversion only optimizes word ebedding, while dreambooth fine-tunes the whole diffusion model.
+
+The implementation makes minimum changes over the official codebase of Textual Inversion. In fact, due to lazyness, some components in Textual Inversion, such as the embedding manager, are not deleted, although they will never be used here.
+## Update
+**9/20/2022**: I just found a way to reduce the GPU memory a bit. Remember that this code is based on Textual Inversion, and TI's code base has [this line](https://github.com/rinongal/textual_inversion/blob/main/ldm/modules/diffusionmodules/util.py#L112), which disable gradient checkpointing in a hard-code way. This is because in TI, the Unet is not optimized. However, in Dreambooth we optimize the Unet, so we can turn on the gradient checkpoint pointing trick, as in the original SD repo [here](https://github.com/CompVis/stable-diffusion/blob/main/ldm/modules/diffusionmodules/util.py#L112). The gradient checkpoint is default to be True in [config](https://github.com/XavierXiao/Dreambooth-Stable-Diffusion/blob/main/configs/stable-diffusion/v1-finetune_unfrozen.yaml#L47). I have updated the codes.
+## Usage
+
+### Preparation
+First set-up the ```ldm``` enviroment following the instruction from textual inversion repo, or the original Stable Diffusion repo.
+
+To fine-tune a stable diffusion model, you need to obtain the pre-trained stable diffusion models following their [instructions](https://github.com/CompVis/stable-diffusion#stable-diffusion-v1). Weights can be downloaded on [HuggingFace](https://huggingface.co/CompVis). You can decide which version of checkpoint to use, but I use ```sd-v1-4-full-ema.ckpt```.
+
+We also need to create a set of images for regularization, as the fine-tuning algorithm of Dreambooth requires that. Details of the algorithm can be found in the paper. Note that in the original paper, the regularization images seem to be generated on-the-fly. However, here I generated a set of regularization images before the training. The text prompt for generating regularization images can be ```photo of a <class>```, where ```<class>``` is a word that describes the class of your object, such as ```dog```. The command is
+
+```
+python scripts/stable_txt2img.py --ddim_eta 0.0 --n_samples 8 --n_iter 1 --scale 10.0 --ddim_steps 50  --ckpt /path/to/original/stable-diffusion/sd-v1-4-full-ema.ckpt --prompt "a photo of a <class>" 
+```
+
+I generate 8 images for regularization, but more regularization images may lead to stronger regularization and better editability. After that, save the generated images (separately, one image per ```.png``` file) at ```/root/to/regularization/images```.
+
+**Updates on 9/9**
+We should definitely use more images for regularization. Please try 100 or 200, to better align with the original paper. To acomodate this, I shorten the "repeat" of reg dataset in the [config file](https://github.com/XavierXiao/Dreambooth-Stable-Diffusion/blob/main/configs/stable-diffusion/v1-finetune_unfrozen.yaml#L96).
+
+For some cases, if the generated regularization images are highly unrealistic (happens when you want to generate "man" or "woman"), you can find a diverse set of images (of man/woman) online, and use them as regularization images.
+
+### Training
+Training can be done by running the following command
+
+```
+python main.py --base configs/stable-diffusion/v1-finetune_unfrozen.yaml 
+                -t 
+                --actual_resume /path/to/original/stable-diffusion/sd-v1-4-full-ema.ckpt  
+                -n <job name> 
+                --gpus 0, 
+                --data_root /root/to/training/images 
+                --reg_data_root /root/to/regularization/images 
+                --class_word <xxx>
+```
+
+Detailed configuration can be found in ```configs/stable-diffusion/v1-finetune_unfrozen.yaml```. In particular, the default learning rate is ```1.0e-6``` as I found the ```1.0e-5``` in the Dreambooth paper leads to poor editability. The parameter ```reg_weight``` corresponds to the weight of regularization in the Dreambooth paper, and the default is set to ```1.0```.
+
+Dreambooth requires a placeholder word ```[V]```, called identifier, as in the paper. This identifier needs to be a relatively rare tokens in the vocabulary. The original paper approaches this by using a rare word in T5-XXL tokenizer. For simplicity, here I just use a random word ```sks``` and hard coded it.. If you want to change that, simply make a change in [this file](https://github.com/XavierXiao/Dreambooth-Stable-Diffusion/blob/main/ldm/data/personalized.py#L10).
+
+Training will be run for 800 steps, and two checkpoints will be saved at ```./logs/<job_name>/checkpoints```, one at 500 steps and one at final step. Typically the one at 500 steps works well enough. I train the model use two A6000 GPUs and it takes ~15 mins.
+
+### Generation
+After training, personalized samples can be obtained by running the command
+
+```
+python scripts/stable_txt2img.py --ddim_eta 0.0 
+                                 --n_samples 8 
+                                 --n_iter 1 
+                                 --scale 10.0 
+                                 --ddim_steps 100  
+                                 --ckpt /path/to/saved/checkpoint/from/training
+                                 --prompt "photo of a sks <class>" 
+```
+
+In particular, ```sks``` is the identifier, which should be replaced by your choice if you happen to change the identifier, and ```<class>``` is the class word ```--class_word``` for training.
+
+## Results
+Here I show some qualitative results. The training images are obtained from the [issue](https://github.com/rinongal/textual_inversion/issues/8) in the Textual Inversion repository, and they are 3 images of a large trash container. Regularization images are generated by prompt ```photo of a container```. Regularization images are shown here:
+
+![](assets/a-container-0038.jpg)
+
+After training, generated images with prompt ```photo of a sks container```:
+
+![](assets/photo-of-a-sks-container-0018.jpg)
+
+Generated images with prompt ```photo of a sks container on the beach```:
+
+![](assets/photo-of-a-sks-container-on-the-beach-0017.jpg)
+
+Generated images with prompt ```photo of a sks container on the moon```:
+
+![](assets/photo-of-a-sks-container-on-the-moon-0016.jpg)
+
+Some not-so-perfect but still interesting results:
+
+Generated images with prompt ```photo of a red sks container```:
+
+![](assets/a-red-sks-container-0021.jpg)
+
+Generated images with prompt ```a dog on top of sks container```:
+
+![](assets/a-dog-on-top-of-sks-container-0023.jpg)
+

configs/autoencoder/autoencoder_kl_16x16x16.yaml

@@ -0,0 +1,54 @@
+model:
+  base_learning_rate: 4.5e-6
+  target: ldm.models.autoencoder.AutoencoderKL
+  params:
+    monitor: "val/rec_loss"
+    embed_dim: 16
+    lossconfig:
+      target: ldm.modules.losses.LPIPSWithDiscriminator
+      params:
+        disc_start: 50001
+        kl_weight: 0.000001
+        disc_weight: 0.5
+
+    ddconfig:
+      double_z: True
+      z_channels: 16
+      resolution: 256
+      in_channels: 3
+      out_ch: 3
+      ch: 128
+      ch_mult: [ 1,1,2,2,4]  # num_down = len(ch_mult)-1
+      num_res_blocks: 2
+      attn_resolutions: [16]
+      dropout: 0.0
+
+
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 12
+    wrap: True
+    train:
+      target: ldm.data.imagenet.ImageNetSRTrain
+      params:
+        size: 256
+        degradation: pil_nearest
+    validation:
+      target: ldm.data.imagenet.ImageNetSRValidation
+      params:
+        size: 256
+        degradation: pil_nearest
+
+lightning:
+  callbacks:
+    image_logger:
+      target: main.ImageLogger
+      params:
+        batch_frequency: 1000
+        max_images: 8
+        increase_log_steps: True
+
+  trainer:
+    benchmark: True
+    accumulate_grad_batches: 2

configs/autoencoder/autoencoder_kl_32x32x4.yaml

@@ -0,0 +1,53 @@
+model:
+  base_learning_rate: 4.5e-6
+  target: ldm.models.autoencoder.AutoencoderKL
+  params:
+    monitor: "val/rec_loss"
+    embed_dim: 4
+    lossconfig:
+      target: ldm.modules.losses.LPIPSWithDiscriminator
+      params:
+        disc_start: 50001
+        kl_weight: 0.000001
+        disc_weight: 0.5
+
+    ddconfig:
+      double_z: True
+      z_channels: 4
+      resolution: 256
+      in_channels: 3
+      out_ch: 3
+      ch: 128
+      ch_mult: [ 1,2,4,4 ]  # num_down = len(ch_mult)-1
+      num_res_blocks: 2
+      attn_resolutions: [ ]
+      dropout: 0.0
+
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 12
+    wrap: True
+    train:
+      target: ldm.data.imagenet.ImageNetSRTrain
+      params:
+        size: 256
+        degradation: pil_nearest
+    validation:
+      target: ldm.data.imagenet.ImageNetSRValidation
+      params:
+        size: 256
+        degradation: pil_nearest
+
+lightning:
+  callbacks:
+    image_logger:
+      target: main.ImageLogger
+      params:
+        batch_frequency: 1000
+        max_images: 8
+        increase_log_steps: True
+
+  trainer:
+    benchmark: True
+    accumulate_grad_batches: 2

configs/autoencoder/autoencoder_kl_64x64x3.yaml

@@ -0,0 +1,54 @@
+model:
+  base_learning_rate: 4.5e-6
+  target: ldm.models.autoencoder.AutoencoderKL
+  params:
+    monitor: "val/rec_loss"
+    embed_dim: 3
+    lossconfig:
+      target: ldm.modules.losses.LPIPSWithDiscriminator
+      params:
+        disc_start: 50001
+        kl_weight: 0.000001
+        disc_weight: 0.5
+
+    ddconfig:
+      double_z: True
+      z_channels: 3
+      resolution: 256
+      in_channels: 3
+      out_ch: 3
+      ch: 128
+      ch_mult: [ 1,2,4 ]  # num_down = len(ch_mult)-1
+      num_res_blocks: 2
+      attn_resolutions: [ ]
+      dropout: 0.0
+
+
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 12
+    wrap: True
+    train:
+      target: ldm.data.imagenet.ImageNetSRTrain
+      params:
+        size: 256
+        degradation: pil_nearest
+    validation:
+      target: ldm.data.imagenet.ImageNetSRValidation
+      params:
+        size: 256
+        degradation: pil_nearest
+
+lightning:
+  callbacks:
+    image_logger:
+      target: main.ImageLogger
+      params:
+        batch_frequency: 1000
+        max_images: 8
+        increase_log_steps: True
+
+  trainer:
+    benchmark: True
+    accumulate_grad_batches: 2

configs/autoencoder/autoencoder_kl_8x8x64.yaml

@@ -0,0 +1,53 @@
+model:
+  base_learning_rate: 4.5e-6
+  target: ldm.models.autoencoder.AutoencoderKL
+  params:
+    monitor: "val/rec_loss"
+    embed_dim: 64
+    lossconfig:
+      target: ldm.modules.losses.LPIPSWithDiscriminator
+      params:
+        disc_start: 50001
+        kl_weight: 0.000001
+        disc_weight: 0.5
+
+    ddconfig:
+      double_z: True
+      z_channels: 64
+      resolution: 256
+      in_channels: 3
+      out_ch: 3
+      ch: 128
+      ch_mult: [ 1,1,2,2,4,4]  # num_down = len(ch_mult)-1
+      num_res_blocks: 2
+      attn_resolutions: [16,8]
+      dropout: 0.0
+
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 12
+    wrap: True
+    train:
+      target: ldm.data.imagenet.ImageNetSRTrain
+      params:
+        size: 256
+        degradation: pil_nearest
+    validation:
+      target: ldm.data.imagenet.ImageNetSRValidation
+      params:
+        size: 256
+        degradation: pil_nearest
+
+lightning:
+  callbacks:
+    image_logger:
+      target: main.ImageLogger
+      params:
+        batch_frequency: 1000
+        max_images: 8
+        increase_log_steps: True
+
+  trainer:
+    benchmark: True
+    accumulate_grad_batches: 2

configs/latent-diffusion/celebahq-ldm-vq-4.yaml

@@ -0,0 +1,86 @@
+model:
+  base_learning_rate: 2.0e-06
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0195
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    image_size: 64
+    channels: 3
+    monitor: val/loss_simple_ema
+
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 64
+        in_channels: 3
+        out_channels: 3
+        model_channels: 224
+        attention_resolutions:
+        # note: this isn\t actually the resolution but
+        # the downsampling factor, i.e. this corresnponds to
+        # attention on spatial resolution 8,16,32, as the
+        # spatial reolution of the latents is 64 for f4
+        - 8
+        - 4
+        - 2
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 3
+        - 4
+        num_head_channels: 32
+    first_stage_config:
+      target: ldm.models.autoencoder.VQModelInterface
+      params:
+        embed_dim: 3
+        n_embed: 8192
+        ckpt_path: models/first_stage_models/vq-f4/model.ckpt
+        ddconfig:
+          double_z: false
+          z_channels: 3
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+    cond_stage_config: __is_unconditional__
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 48
+    num_workers: 5
+    wrap: false
+    train:
+      target: taming.data.faceshq.CelebAHQTrain
+      params:
+        size: 256
+    validation:
+      target: taming.data.faceshq.CelebAHQValidation
+      params:
+        size: 256
+
+
+lightning:
+  callbacks:
+    image_logger:
+      target: main.ImageLogger
+      params:
+        batch_frequency: 5000
+        max_images: 8
+        increase_log_steps: False
+
+  trainer:
+    benchmark: True

configs/latent-diffusion/cin-ldm-vq-f8.yaml

@@ -0,0 +1,98 @@
+model:
+  base_learning_rate: 1.0e-06
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0195
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    cond_stage_key: class_label
+    image_size: 32
+    channels: 4
+    cond_stage_trainable: true
+    conditioning_key: crossattn
+    monitor: val/loss_simple_ema
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32
+        in_channels: 4
+        out_channels: 4
+        model_channels: 256
+        attention_resolutions:
+        #note: this isn\t actually the resolution but
+        # the downsampling factor, i.e. this corresnponds to
+        # attention on spatial resolution 8,16,32, as the
+        # spatial reolution of the latents is 32 for f8
+        - 4
+        - 2
+        - 1
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 4
+        num_head_channels: 32
+        use_spatial_transformer: true
+        transformer_depth: 1
+        context_dim: 512
+    first_stage_config:
+      target: ldm.models.autoencoder.VQModelInterface
+      params:
+        embed_dim: 4
+        n_embed: 16384
+        ckpt_path: configs/first_stage_models/vq-f8/model.yaml
+        ddconfig:
+          double_z: false
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 2
+          - 4
+          num_res_blocks: 2
+          attn_resolutions:
+          - 32
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.ClassEmbedder
+      params:
+        embed_dim: 512
+        key: class_label
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 64
+    num_workers: 12
+    wrap: false
+    train:
+      target: ldm.data.imagenet.ImageNetTrain
+      params:
+        config:
+          size: 256
+    validation:
+      target: ldm.data.imagenet.ImageNetValidation
+      params:
+        config:
+          size: 256
+
+
+lightning:
+  callbacks:
+    image_logger:
+      target: main.ImageLogger
+      params:
+        batch_frequency: 5000
+        max_images: 8
+        increase_log_steps: False
+
+  trainer:
+    benchmark: True

configs/latent-diffusion/cin256-v2.yaml

@@ -0,0 +1,68 @@
+model:
+  base_learning_rate: 0.0001
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0195
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    cond_stage_key: class_label
+    image_size: 64
+    channels: 3
+    cond_stage_trainable: true
+    conditioning_key: crossattn
+    monitor: val/loss
+    use_ema: False
+    
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 64
+        in_channels: 3
+        out_channels: 3
+        model_channels: 192
+        attention_resolutions:
+        - 8
+        - 4
+        - 2
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 3
+        - 5
+        num_heads: 1
+        use_spatial_transformer: true
+        transformer_depth: 1
+        context_dim: 512
+    
+    first_stage_config:
+      target: ldm.models.autoencoder.VQModelInterface
+      params:
+        embed_dim: 3
+        n_embed: 8192
+        ddconfig:
+          double_z: false
+          z_channels: 3
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+    
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.ClassEmbedder
+      params:
+        n_classes: 1001
+        embed_dim: 512
+        key: class_label

configs/latent-diffusion/ffhq-ldm-vq-4.yaml

@@ -0,0 +1,85 @@
+model:
+  base_learning_rate: 2.0e-06
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0195
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    image_size: 64
+    channels: 3
+    monitor: val/loss_simple_ema
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 64
+        in_channels: 3
+        out_channels: 3
+        model_channels: 224
+        attention_resolutions:
+        # note: this isn\t actually the resolution but
+        # the downsampling factor, i.e. this corresnponds to
+        # attention on spatial resolution 8,16,32, as the
+        # spatial reolution of the latents is 64 for f4
+        - 8
+        - 4
+        - 2
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 3
+        - 4
+        num_head_channels: 32
+    first_stage_config:
+      target: ldm.models.autoencoder.VQModelInterface
+      params:
+        embed_dim: 3
+        n_embed: 8192
+        ckpt_path: configs/first_stage_models/vq-f4/model.yaml
+        ddconfig:
+          double_z: false
+          z_channels: 3
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+    cond_stage_config: __is_unconditional__
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 42
+    num_workers: 5
+    wrap: false
+    train:
+      target: taming.data.faceshq.FFHQTrain
+      params:
+        size: 256
+    validation:
+      target: taming.data.faceshq.FFHQValidation
+      params:
+        size: 256
+
+
+lightning:
+  callbacks:
+    image_logger:
+      target: main.ImageLogger
+      params:
+        batch_frequency: 5000
+        max_images: 8
+        increase_log_steps: False
+
+  trainer:
+    benchmark: True

configs/latent-diffusion/lsun_bedrooms-ldm-vq-4.yaml

@@ -0,0 +1,85 @@
+model:
+  base_learning_rate: 2.0e-06
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0195
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    image_size: 64
+    channels: 3
+    monitor: val/loss_simple_ema
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 64
+        in_channels: 3
+        out_channels: 3
+        model_channels: 224
+        attention_resolutions:
+        # note: this isn\t actually the resolution but
+        # the downsampling factor, i.e. this corresnponds to
+        # attention on spatial resolution 8,16,32, as the
+        # spatial reolution of the latents is 64 for f4
+        - 8
+        - 4
+        - 2
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 3
+        - 4
+        num_head_channels: 32
+    first_stage_config:
+      target: ldm.models.autoencoder.VQModelInterface
+      params:
+        ckpt_path: configs/first_stage_models/vq-f4/model.yaml
+        embed_dim: 3
+        n_embed: 8192
+        ddconfig:
+          double_z: false
+          z_channels: 3
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+    cond_stage_config: __is_unconditional__
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 48
+    num_workers: 5
+    wrap: false
+    train:
+      target: ldm.data.lsun.LSUNBedroomsTrain
+      params:
+        size: 256
+    validation:
+      target: ldm.data.lsun.LSUNBedroomsValidation
+      params:
+        size: 256
+
+
+lightning:
+  callbacks:
+    image_logger:
+      target: main.ImageLogger
+      params:
+        batch_frequency: 5000
+        max_images: 8
+        increase_log_steps: False
+
+  trainer:
+    benchmark: True

configs/latent-diffusion/lsun_churches-ldm-kl-8.yaml

@@ -0,0 +1,91 @@
+model:
+  base_learning_rate: 5.0e-5   # set to target_lr by starting main.py with '--scale_lr False'
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0155
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    loss_type: l1
+    first_stage_key: "image"
+    cond_stage_key: "image"
+    image_size: 32
+    channels: 4
+    cond_stage_trainable: False
+    concat_mode: False
+    scale_by_std: True
+    monitor: 'val/loss_simple_ema'
+
+    scheduler_config: # 10000 warmup steps
+      target: ldm.lr_scheduler.LambdaLinearScheduler
+      params:
+        warm_up_steps: [10000]
+        cycle_lengths: [10000000000000]
+        f_start: [1.e-6]
+        f_max: [1.]
+        f_min: [ 1.]
+
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32
+        in_channels: 4
+        out_channels: 4
+        model_channels: 192
+        attention_resolutions: [ 1, 2, 4, 8 ]   # 32, 16, 8, 4
+        num_res_blocks: 2
+        channel_mult: [ 1,2,2,4,4 ]  # 32, 16, 8, 4, 2
+        num_heads: 8
+        use_scale_shift_norm: True
+        resblock_updown: True
+
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: "val/rec_loss"
+        ckpt_path: "models/first_stage_models/kl-f8/model.ckpt"
+        ddconfig:
+          double_z: True
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult: [ 1,2,4,4 ]  # num_down = len(ch_mult)-1
+          num_res_blocks: 2
+          attn_resolutions: [ ]
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config: "__is_unconditional__"
+
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 96
+    num_workers: 5
+    wrap: False
+    train:
+      target: ldm.data.lsun.LSUNChurchesTrain
+      params:
+        size: 256
+    validation:
+      target: ldm.data.lsun.LSUNChurchesValidation
+      params:
+        size: 256
+
+lightning:
+  callbacks:
+    image_logger:
+      target: main.ImageLogger
+      params:
+        batch_frequency: 5000
+        max_images: 8
+        increase_log_steps: False
+
+
+  trainer:
+    benchmark: True

configs/latent-diffusion/txt2img-1p4B-eval.yaml

@@ -0,0 +1,71 @@
+model:
+  base_learning_rate: 5.0e-05
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.00085
+    linear_end: 0.012
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    cond_stage_key: caption
+    image_size: 32
+    channels: 4
+    cond_stage_trainable: true
+    conditioning_key: crossattn
+    monitor: val/loss_simple_ema
+    scale_factor: 0.18215
+    use_ema: False
+
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32
+        in_channels: 4
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions:
+        - 4
+        - 2
+        - 1
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 4
+        - 4
+        num_heads: 8
+        use_spatial_transformer: true
+        transformer_depth: 1
+        context_dim: 1280
+        use_checkpoint: true
+        legacy: False
+
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.BERTEmbedder
+      params:
+        n_embed: 1280
+        n_layer: 32

configs/latent-diffusion/txt2img-1p4B-eval_with_tokens.yaml

@@ -0,0 +1,77 @@
+model:
+  base_learning_rate: 5.0e-05
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.00085
+    linear_end: 0.012
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    cond_stage_key: caption
+    image_size: 32
+    channels: 4
+    cond_stage_trainable: true
+    conditioning_key: crossattn
+    monitor: val/loss_simple_ema
+    scale_factor: 0.18215
+    use_ema: False
+
+    personalization_config:
+      target: ldm.modules.embedding_manager.EmbeddingManager
+      params:
+        placeholder_strings: ["*"]
+        initializer_words: []
+
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32
+        in_channels: 4
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions:
+        - 4
+        - 2
+        - 1
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 4
+        - 4
+        num_heads: 8
+        use_spatial_transformer: true
+        transformer_depth: 1
+        context_dim: 1280
+        use_checkpoint: true
+        legacy: False
+
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.BERTEmbedder
+      params:
+        n_embed: 1280
+        n_layer: 32

configs/latent-diffusion/txt2img-1p4B-finetune.yaml

@@ -0,0 +1,119 @@
+model:
+  base_learning_rate: 5.0e-3
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.00085
+    linear_end: 0.012
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    cond_stage_key: caption
+    image_size: 32
+    channels: 4
+    cond_stage_trainable: true
+    conditioning_key: crossattn
+    monitor: val/loss_simple_ema
+    scale_factor: 0.18215
+    use_ema: False
+    embedding_reg_weight: 0.0
+
+    personalization_config:
+      target: ldm.modules.embedding_manager.EmbeddingManager
+      params:
+        placeholder_strings: ["*"]
+        initializer_words: ["sculpture"]
+        per_image_tokens: false
+        num_vectors_per_token: 1
+        progressive_words: False
+
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32
+        in_channels: 4
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions:
+        - 4
+        - 2
+        - 1
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 4
+        - 4
+        num_heads: 8
+        use_spatial_transformer: true
+        transformer_depth: 1
+        context_dim: 1280
+        use_checkpoint: true
+        legacy: False
+
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.BERTEmbedder
+      params:
+        n_embed: 1280
+        n_layer: 32
+
+
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 4
+    num_workers: 2
+    wrap: false
+    train:
+      target: ldm.data.personalized.PersonalizedBase
+      params:
+        size: 256
+        set: train
+        per_image_tokens: false
+        repeats: 100
+    validation:
+      target: ldm.data.personalized.PersonalizedBase
+      params:
+        size: 256
+        set: val
+        per_image_tokens: false
+        repeats: 10
+
+lightning:
+  modelcheckpoint:
+    params:
+      every_n_train_steps: 500
+  callbacks:
+    image_logger:
+      target: main.ImageLogger
+      params:
+        batch_frequency: 500
+        max_images: 8
+        increase_log_steps: False
+
+  trainer:
+    benchmark: True
+    max_steps: 6100

configs/latent-diffusion/txt2img-1p4B-finetune_style.yaml

@@ -0,0 +1,117 @@
+model:
+  base_learning_rate: 5.0e-3
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.00085
+    linear_end: 0.012
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    cond_stage_key: caption
+    image_size: 32
+    channels: 4
+    cond_stage_trainable: true
+    conditioning_key: crossattn
+    monitor: val/loss_simple_ema
+    scale_factor: 0.18215
+    use_ema: False
+    embedding_reg_weight: 0.0
+
+    personalization_config:
+      target: ldm.modules.embedding_manager.EmbeddingManager
+      params:
+        placeholder_strings: ["*"]
+        initializer_words: ["painting"]
+        per_image_tokens: false
+        num_vectors_per_token: 1
+
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32
+        in_channels: 4
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions:
+        - 4
+        - 2
+        - 1
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 4
+        - 4
+        num_heads: 8
+        use_spatial_transformer: true
+        transformer_depth: 1
+        context_dim: 1280
+        use_checkpoint: true
+        legacy: False
+
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.BERTEmbedder
+      params:
+        n_embed: 1280
+        n_layer: 32
+
+
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 4
+    num_workers: 4
+    wrap: false
+    train:
+      target: ldm.data.personalized_style.PersonalizedBase
+      params:
+        size: 256
+        set: train
+        per_image_tokens: false
+        repeats: 100
+    validation:
+      target: ldm.data.personalized_style.PersonalizedBase
+      params:
+        size: 256
+        set: val
+        per_image_tokens: false
+        repeats: 10
+
+lightning:
+  modelcheckpoint:
+    params:
+      every_n_train_steps: 500
+  callbacks:
+    image_logger:
+      target: main.ImageLogger
+      params:
+        batch_frequency: 500
+        max_images: 8
+        increase_log_steps: False
+
+  trainer:
+    benchmark: True

configs/stable-diffusion/v1-finetune.yaml

@@ -0,0 +1,110 @@
+model:
+  base_learning_rate: 5.0e-03
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.00085
+    linear_end: 0.0120
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    cond_stage_key: caption
+    image_size: 64
+    channels: 4
+    cond_stage_trainable: true   # Note: different from the one we trained before
+    conditioning_key: crossattn
+    monitor: val/loss_simple_ema
+    scale_factor: 0.18215
+    use_ema: False
+    embedding_reg_weight: 0.0
+    unfreeze_model: False
+    model_lr: 0.0
+
+    personalization_config:
+      target: ldm.modules.embedding_manager.EmbeddingManager
+      params:
+        placeholder_strings: ["*"]
+        initializer_words: ["sculpture"]
+        per_image_tokens: false
+        num_vectors_per_token: 1
+        progressive_words: False
+
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32 # unused
+        in_channels: 4
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions: [ 4, 2, 1 ]
+        num_res_blocks: 2
+        channel_mult: [ 1, 2, 4, 4 ]
+        num_heads: 8
+        use_spatial_transformer: True
+        transformer_depth: 1
+        context_dim: 768
+        use_checkpoint: True
+        legacy: False
+
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 512
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.FrozenCLIPEmbedder
+
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 2
+    num_workers: 2
+    wrap: false
+    train:
+      target: ldm.data.personalized.PersonalizedBase
+      params:
+        size: 512
+        set: train
+        per_image_tokens: false
+        repeats: 100
+    validation:
+      target: ldm.data.personalized.PersonalizedBase
+      params:
+        size: 512
+        set: val
+        per_image_tokens: false
+        repeats: 10
+
+lightning:
+  modelcheckpoint:
+    params:
+      every_n_train_steps: 500
+  callbacks:
+    image_logger:
+      target: main.ImageLogger
+      params:
+        batch_frequency: 500
+        max_images: 8
+        increase_log_steps: False
+
+  trainer:
+    benchmark: True
+    max_steps: 6100

configs/stable-diffusion/v1-finetune_unfrozen.yaml

@@ -0,0 +1,120 @@
+model:
+  base_learning_rate: 1.0e-06
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    reg_weight: 1.0
+    linear_start: 0.00085
+    linear_end: 0.0120
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    cond_stage_key: caption
+    image_size: 64
+    channels: 4
+    cond_stage_trainable: true   # Note: different from the one we trained before
+    conditioning_key: crossattn
+    monitor: val/loss_simple_ema
+    scale_factor: 0.18215
+    use_ema: False
+    embedding_reg_weight: 0.0
+    unfreeze_model: True
+    model_lr: 1.0e-6
+
+    personalization_config:
+      target: ldm.modules.embedding_manager.EmbeddingManager
+      params:
+        placeholder_strings: ["*"]
+        initializer_words: ["sculpture"]
+        per_image_tokens: false
+        num_vectors_per_token: 1
+        progressive_words: False
+
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32 # unused
+        in_channels: 4
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions: [ 4, 2, 1 ]
+        num_res_blocks: 2
+        channel_mult: [ 1, 2, 4, 4 ]
+        num_heads: 8
+        use_spatial_transformer: True
+        transformer_depth: 1
+        context_dim: 768
+        use_checkpoint: True
+        legacy: False
+
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 512
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.FrozenCLIPEmbedder
+
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 1
+    num_workers: 1
+    wrap: false
+    train:
+      target: ldm.data.personalized.PersonalizedBase
+      params:
+        size: 512
+        set: train
+        per_image_tokens: false
+        repeats: 100
+    reg:
+      target: ldm.data.personalized.PersonalizedBase
+      params:
+        size: 512
+        set: train
+        reg: true
+        per_image_tokens: false
+        repeats: 10
+        
+    validation:
+      target: ldm.data.personalized.PersonalizedBase
+      params:
+        size: 512
+        set: val
+        per_image_tokens: false
+        repeats: 10
+
+lightning:
+  modelcheckpoint:
+    params:
+      every_n_train_steps: 500
+  callbacks:
+    image_logger:
+      target: main.ImageLogger
+      params:
+        batch_frequency: 500
+        max_images: 8
+        increase_log_steps: False
+
+  trainer:
+    benchmark: True
+    max_steps: 3000

configs/stable-diffusion/v1-inference.yaml

@@ -0,0 +1,70 @@
+model:
+  base_learning_rate: 1.0e-04
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.00085
+    linear_end: 0.0120
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: "jpg"
+    cond_stage_key: "txt"
+    image_size: 64
+    channels: 4
+    cond_stage_trainable: false   # Note: different from the one we trained before
+    conditioning_key: crossattn
+    monitor: val/loss_simple_ema
+    scale_factor: 0.18215
+    use_ema: False
+
+    personalization_config:
+      target: ldm.modules.embedding_manager.EmbeddingManager
+      params:
+        placeholder_strings: ["*"]
+        initializer_words: ["sculpture"]
+        per_image_tokens: false
+        num_vectors_per_token: 1
+        progressive_words: False
+        
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32 # unused
+        in_channels: 4
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions: [ 4, 2, 1 ]
+        num_res_blocks: 2
+        channel_mult: [ 1, 2, 4, 4 ]
+        num_heads: 8
+        use_spatial_transformer: True
+        transformer_depth: 1
+        context_dim: 768
+        use_checkpoint: True
+        legacy: False
+
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.FrozenCLIPEmbedder

dreambooth_runpod_joepenna.ipynb

@@ -0,0 +1,236 @@
+{
+  "metadata": {
+    "colab": {
+      "collapsed_sections": [],
+      "provenance": []
+    },
+    "kernelspec": {
+      "name": "python",
+      "display_name": "Python (Pyodide)",
+      "language": "python"
+    },
+    "language_info": {
+      "codemirror_mode": {
+        "name": "python",
+        "version": 3
+      },
+      "file_extension": ".py",
+      "mimetype": "text/x-python",
+      "name": "python",
+      "nbconvert_exporter": "python",
+      "pygments_lexer": "ipython3",
+      "version": "3.8"
+    },
+    "vscode": {
+      "interpreter": {
+        "hash": "b0fa6594d8f4cbf19f97940f81e996739fb7646882a419484c72d19e05852a7e"
+      }
+    }
+  },
+  "nbformat_minor": 5,
+  "nbformat": 4,
+  "cells": [
+    {
+      "cell_type": "markdown",
+      "source": "# Dreambooth\n### Notebook implementation by Joe Penna (@MysteryGuitarM on Twitter) - Improvements by David Bielejeski\nhttps://github.com/JoePenna/Dreambooth-Stable-Diffusion\n\n### If on runpod / vast.ai / etc, spin up an A6000 or A100 pod using a Stable Diffusion template with Jupyter pre-installed.",
+      "metadata": {
+        "id": "aa2c1ada"
+      },
+      "id": "aa2c1ada"
+    },
+    {
+      "cell_type": "markdown",
+      "source": "## Build Environment",
+      "metadata": {
+        "id": "7b971cc0"
+      },
+      "id": "7b971cc0"
+    },
+    {
+      "cell_type": "code",
+      "source": "#BUILD ENV\n!pip install omegaconf\n!pip install einops\n!pip install pytorch-lightning==1.6.5\n!pip install test-tube\n!pip install transformers\n!pip install kornia\n!pip install -e git+https://github.com/CompVis/taming-transformers.git@master#egg=taming-transformers\n!pip install -e git+https://github.com/openai/CLIP.git@main#egg=clip\n!pip install setuptools==59.5.0\n!pip install pillow==9.0.1\n!pip install torchmetrics==0.6.0\n!pip install -e .\n!pip install protobuf==3.20.1\n!pip install gdown\n!pip install pydrive\n!pip install -qq diffusers[\"training\"]==0.3.0 transformers ftfy\n!pip install -qq \"ipywidgets>=7,<8\"\n!pip install huggingface_hub\n!pip install ipywidgets",
+      "metadata": {
+        "id": "9e1bc458-091b-42f4-a125-c3f0df20f29d",
+        "scrolled": true
+      },
+      "execution_count": null,
+      "outputs": [],
+      "id": "9e1bc458-091b-42f4-a125-c3f0df20f29d"
+    },
+    {
+      "cell_type": "code",
+      "source": "## Move the sd-v1-4.ckpt to the root of this directory as \"model.ckpt\"\n#actual_locations_of_model_blob = !readlink -f {downloaded_model_path}\n#!cp {actual_locations_of_model_blob[-1]} model.ckpt\n!apt-get update ; apt-get install wget\n!wget 'https://prodesk.home.thijn.ovh/sd-v1-4.ckpt'\n!cp sd-v1-4.ckpt model.ckpt",
+      "metadata": {
+        "collapsed": false,
+        "jupyter": {
+          "outputs_hidden": false
+        }
+      },
+      "execution_count": null,
+      "outputs": [],
+      "id": "ddf7a43d"
+    },
+    {
+      "cell_type": "markdown",
+      "source": "## Download pre-generated regularization images\n\nWe've created the following image sets\n\n* man_euler - provided by Niko Pueringer (Corridor Digital) - euler @ 40 steps, CFG 7.5\n* man_unsplash - pictures from various photographers\n* person_ddim\n* woman_ddim - provided by David Bielejeski - ddim @ 50 steps, CFG 10.0\n\n`person_ddim` is recommended",
+      "metadata": {
+        "id": "mxPL2O0OLvBW"
+      },
+      "id": "mxPL2O0OLvBW"
+    },
+    {
+      "cell_type": "code",
+      "source": "# Grab the existing regularization images\n# Choose the dataset that best represents what you are trying to do and matches what you used for your token\n# man_euler, man_unsplash, person_ddim, woman_ddim\ndataset=\"man_euler\"\n!rm -rf Stable-Diffusion-Regularization-Images-{dataset}\n!git clone https://github.com/djbielejeski/Stable-Diffusion-Regularization-Images-{dataset}.git\n\n!mkdir -p outputs/txt2img-samples/samples/{dataset}\n!mv -v Stable-Diffusion-Regularization-Images-{dataset}/{dataset}/*.* outputs/txt2img-samples/samples/{dataset}",
+      "metadata": {
+        "id": "e7EydXCjOV1v"
+      },
+      "execution_count": null,
+      "outputs": [],
+      "id": "e7EydXCjOV1v"
+    },
+    {
+      "cell_type": "markdown",
+      "source": "# Upload your training images\nUpload 10-20 images of someone to\n\n```\n/workspace/Dreambooth-Stable-Diffusion/training_samples\n```\n\nWARNING: Be sure to upload an *even* amount of images, otherwise the training inexplicably stops at 1500 steps.\n\n*   2-3 full body\n*   3-5 upper body \n*   5-12 close-up on face",
+      "metadata": {
+        "id": "zshrC_JuMXmM"
+      },
+      "id": "zshrC_JuMXmM"
+    },
+    {
+      "cell_type": "code",
+      "source": "#@markdown Add here the URLs to the images of the concept you are adding\nurls = [\n\"https://prodesk.home.thijn.ovh/gijs/IMG_20220926_121625.png\",\n\"https://prodesk.home.thijn.ovh/gijs/IMG_20220926_121630.png\",\n\"https://prodesk.home.thijn.ovh/gijs/IMG_20220926_121632.png\",\n\"https://prodesk.home.thijn.ovh/gijs/IMG_20220926_121636.png\",\n\"https://prodesk.home.thijn.ovh/gijs/IMG_20220926_121637.png\",\n\"https://prodesk.home.thijn.ovh/gijs/IMG_20220926_121640.png\",\n\"https://prodesk.home.thijn.ovh/gijs/IMG_20220926_121642.png\",\n\"https://prodesk.home.thijn.ovh/gijs/IMG_20220926_121644.png\",\n\"https://prodesk.home.thijn.ovh/gijs/IMG_20220926_121647.png\",\n\"https://prodesk.home.thijn.ovh/gijs/IMG_20220926_121649.png\",\n\"https://prodesk.home.thijn.ovh/gijs/IMG_20220926_121653.png\",\n\"https://prodesk.home.thijn.ovh/gijs/IMG_20220926_121656.png\",\n\"https://prodesk.home.thijn.ovh/gijs/IMG_20220926_121659.png\",\n\"https://prodesk.home.thijn.ovh/gijs/IMG_20220926_121702.png\"\n ## You can add additional images here\n]",
+      "metadata": {
+        "collapsed": false,
+        "jupyter": {
+          "outputs_hidden": false
+        }
+      },
+      "execution_count": null,
+      "outputs": [],
+      "id": "60e37ee0"
+    },
+    {
+      "cell_type": "code",
+      "source": "#@title Download and check the images you have just added\nimport os\nimport requests\nfrom io import BytesIO\nfrom PIL import Image\n\n\ndef image_grid(imgs, rows, cols):\n assert len(imgs) == rows*cols\n\n w, h = imgs[0].size\n grid = Image.new('RGB', size=(cols*w, rows*h))\n grid_w, grid_h = grid.size\n\n for i, img in enumerate(imgs):\n  grid.paste(img, box=(i%cols*w, i//cols*h))\n return grid\n\ndef download_image(url):\n try:\n  response = requests.get(url)\n except:\n  return None\n return Image.open(BytesIO(response.content)).convert(\"RGB\")\n\nimages = list(filter(None,[download_image(url) for url in urls]))\nsave_path = \"./training_samples\"\nif not os.path.exists(save_path):\n os.mkdir(save_path)\n[image.save(f\"{save_path}/{i}.png\", format=\"png\") for i, image in enumerate(images)]\nimage_grid(images, 1, len(images))",
+      "metadata": {
+        "collapsed": false,
+        "jupyter": {
+          "outputs_hidden": false
+        }
+      },
+      "execution_count": null,
+      "outputs": [],
+      "id": "df314e2e"
+    },
+    {
+      "cell_type": "markdown",
+      "source": "## Training\n\nIf training a person or subject, keep an eye on your project's `logs/{folder}/images/train/samples_scaled_gs-00xxxx` generations.\n\nIf training a style, keep an eye on your project's `logs/{folder}/images/train/samples_gs-00xxxx` generations.",
+      "metadata": {
+        "id": "ad4e50df"
+      },
+      "id": "ad4e50df"
+    },
+    {
+      "cell_type": "code",
+      "source": "# START THE TRAINING\nproject_name = \"gijsbert\"\nbatch_size = 1000\nclass_word = \"man\"  # << match this word to the class word from regularization images above\nreg_data_root = \"/workspace/Dreambooth-Stable-Diffusion/outputs/txt2img-samples/samples/\" + dataset\n\n!rm -rf training_samples/.ipynb_checkpoints\n!python \"main.py\" \\\n --base configs/stable-diffusion/v1-finetune_unfrozen.yaml \\\n -t \\\n --actual_resume \"model.ckpt\" \\\n --reg_data_root {reg_data_root} \\\n -n {project_name} \\\n --gpus 0, \\\n --data_root \"/workspace/Dreambooth-Stable-Diffusion/training_samples\" \\\n --batch_size {batch_size} \\\n --class_word class_word",
+      "metadata": {
+        "id": "6fa5dd66-2ca0-4819-907e-802e25583ae6",
+        "tags": []
+      },
+      "execution_count": null,
+      "outputs": [],
+      "id": "6fa5dd66-2ca0-4819-907e-802e25583ae6"
+    },
+    {
+      "cell_type": "markdown",
+      "source": "## Pruning (12GB to 2GB)\nWe are working on having this happen automatically (TODO: PR's welcome)",
+      "metadata": {},
+      "id": "dc49d0bd"
+    },
+    {
+      "cell_type": "code",
+      "source": "directory_paths = !ls -d logs/*",
+      "metadata": {
+        "collapsed": false,
+        "jupyter": {
+          "outputs_hidden": false
+        }
+      },
+      "execution_count": null,
+      "outputs": [],
+      "id": "27cea333"
+    },
+    {
+      "cell_type": "code",
+      "source": "# This version should automatically prune around 10GB from the ckpt file\nlast_checkpoint_file = directory_paths[-1] + \"/checkpoints/last.ckpt\"\n!python \"prune_ckpt.py\" --ckpt {last_checkpoint_file}",
+      "metadata": {
+        "collapsed": false,
+        "jupyter": {
+          "outputs_hidden": false
+        }
+      },
+      "execution_count": null,
+      "outputs": [],
+      "id": "965b4654"
+    },
+    {
+      "cell_type": "code",
+      "source": "last_checkpoint_file_pruned = directory_paths[-1] + \"/checkpoints/last-pruned.ckpt\"\ntraining_samples = !ls training_samples\ndate_string = !date +\"%Y-%m-%dT%H-%M-%S\"\nfile_name = date_string[-1] + \"_\" + project_name + \"_\" + str(len(training_samples)) + \"_training_images_\" +  str(batch_size) + \"_batch_size_\" + class_word + \"_class_word.ckpt\"\n!mkdir -p trained_models\n!mv {last_checkpoint_file_pruned} trained_models/{file_name}",
+      "metadata": {
+        "collapsed": false,
+        "jupyter": {
+          "outputs_hidden": false
+        }
+      },
+      "execution_count": null,
+      "outputs": [],
+      "id": "b7a8cec3"
+    },
+    {
+      "cell_type": "code",
+      "source": "# Download your trained model file from `trained_models` and use in your favorite Stable Diffusion repo!",
+      "metadata": {},
+      "execution_count": null,
+      "outputs": [],
+      "id": "ff1a46d9"
+    },
+    {
+      "cell_type": "markdown",
+      "source": "## Generate Images With Your Trained Model!",
+      "metadata": {},
+      "id": "d28d0139"
+    },
+    {
+      "cell_type": "code",
+      "source": "!echo python scripts/stable_txt2img.py \\\n --ddim_eta 0.0 \\\n --n_samples 1 \\\n --n_iter 4 \\\n --scale 7.0 \\\n --ddim_steps 50 \\\n --ckpt \"/workspace/Dreambooth-Stable-Diffusion/trained_models/\" + {file_name} \\\n --prompt \"gijsbert person as a masterpiece portrait painting by John Singer Sargent in the style of Rembrandt\"",
+      "metadata": {
+        "trusted": true
+      },
+      "execution_count": 2,
+      "outputs": [
+        {
+          "ename": "<class 'AttributeError'>",
+          "evalue": "module 'pexpect' has no attribute 'TIMEOUT'",
+          "traceback": [
+            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+            "\u001b[0;31mAttributeError\u001b[0m                            Traceback (most recent call last)",
+            "Cell \u001b[0;32mIn [2], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m get_ipython()\u001b[38;5;241m.\u001b[39msystem(\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mchangeme python scripts/stable_txt2img.py   --ddim_eta 0.0   --n_samples 1   --n_iter 4   --scale 7.0   --ddim_steps 50   --ckpt \u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m/workspace/Dreambooth-Stable-Diffusion/trained_models/\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m + \u001b[39m\u001b[38;5;132;01m{file_name}\u001b[39;00m\u001b[38;5;124m   --prompt \u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mjoepenna person as a masterpiece portrait painting by John Singer Sargent in the style of Rembrandt\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m'\u001b[39m)\n",
+            "File \u001b[0;32m/lib/python3.10/site-packages/IPython/core/interactiveshell.py:2466\u001b[0m, in \u001b[0;36mInteractiveShell.system_piped\u001b[0;34m(self, cmd)\u001b[0m\n\u001b[1;32m   2461\u001b[0m     \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mOSError\u001b[39;00m(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mBackground processes not supported.\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[1;32m   2463\u001b[0m \u001b[38;5;66;03m# we explicitly do NOT return the subprocess status code, because\u001b[39;00m\n\u001b[1;32m   2464\u001b[0m \u001b[38;5;66;03m# a non-None value would trigger :func:`sys.displayhook` calls.\u001b[39;00m\n\u001b[1;32m   2465\u001b[0m \u001b[38;5;66;03m# Instead, we store the exit_code in user_ns.\u001b[39;00m\n\u001b[0;32m-> 2466\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39muser_ns[\u001b[38;5;124m'\u001b[39m\u001b[38;5;124m_exit_code\u001b[39m\u001b[38;5;124m'\u001b[39m] \u001b[38;5;241m=\u001b[39m \u001b[43msystem\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mvar_expand\u001b[49m\u001b[43m(\u001b[49m\u001b[43mcmd\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mdepth\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;241;43m1\u001b[39;49m\u001b[43m)\u001b[49m\u001b[43m)\u001b[49m\n",
+            "File \u001b[0;32m/lib/python3.10/site-packages/IPython/utils/_process_posix.py:129\u001b[0m, in \u001b[0;36mProcessHandler.system\u001b[0;34m(self, cmd)\u001b[0m\n\u001b[1;32m    125\u001b[0m enc \u001b[38;5;241m=\u001b[39m DEFAULT_ENCODING\n\u001b[1;32m    127\u001b[0m \u001b[38;5;66;03m# Patterns to match on the output, for pexpect.  We read input and\u001b[39;00m\n\u001b[1;32m    128\u001b[0m \u001b[38;5;66;03m# allow either a short timeout or EOF\u001b[39;00m\n\u001b[0;32m--> 129\u001b[0m patterns \u001b[38;5;241m=\u001b[39m [\u001b[43mpexpect\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mTIMEOUT\u001b[49m, pexpect\u001b[38;5;241m.\u001b[39mEOF]\n\u001b[1;32m    130\u001b[0m \u001b[38;5;66;03m# the index of the EOF pattern in the list.\u001b[39;00m\n\u001b[1;32m    131\u001b[0m \u001b[38;5;66;03m# even though we know it's 1, this call means we don't have to worry if\u001b[39;00m\n\u001b[1;32m    132\u001b[0m \u001b[38;5;66;03m# we change the above list, and forget to change this value:\u001b[39;00m\n\u001b[1;32m    133\u001b[0m EOF_index \u001b[38;5;241m=\u001b[39m patterns\u001b[38;5;241m.\u001b[39mindex(pexpect\u001b[38;5;241m.\u001b[39mEOF)\n",
+            "\u001b[0;31mAttributeError\u001b[0m: module 'pexpect' has no attribute 'TIMEOUT'"
+          ],
+          "output_type": "error"
+        }
+      ],
+      "id": "80ddb03b"
+    },
+    {
+      "cell_type": "code",
+      "source": "",
+      "metadata": {},
+      "execution_count": null,
+      "outputs": [],
+      "id": "0e3c10d9-2c40-4f50-9cf4-97e88a57288c"
+    }
+  ]
+}

environment.yaml

@@ -0,0 +1,31 @@
+name: ldm
+channels:
+  - pytorch
+  - defaults
+dependencies:
+  - python=3.8.10
+  - pip=20.3
+  - cudatoolkit=11.3
+  - pytorch=1.10.2
+  - torchvision=0.11.3
+  - numpy=1.22.3
+  - pip:
+    - albumentations==1.1.0
+    - opencv-python==4.2.0.34
+    - pudb==2019.2
+    - imageio==2.14.1
+    - imageio-ffmpeg==0.4.7
+    - pytorch-lightning==1.5.9
+    - omegaconf==2.1.1
+    - test-tube>=0.7.5
+    - streamlit>=0.73.1
+    - setuptools==59.5.0
+    - pillow==9.0.1
+    - einops==0.4.1
+    - torch-fidelity==0.3.0
+    - transformers==4.18.0
+    - torchmetrics==0.6.0
+    - kornia==0.6
+    - -e git+https://github.com/CompVis/taming-transformers.git@master#egg=taming-transformers
+    - -e git+https://github.com/openai/CLIP.git@main#egg=clip
+    - -e .

evaluation/clip_eval.py

@@ -0,0 +1,113 @@
+import clip
+import torch
+from torchvision import transforms
+
+from ldm.models.diffusion.ddim import DDIMSampler
+
+class CLIPEvaluator(object):
+    def __init__(self, device, clip_model='ViT-B/32') -> None:
+        self.device = device
+        self.model, clip_preprocess = clip.load(clip_model, device=self.device)
+
+        self.clip_preprocess = clip_preprocess
+        
+        self.preprocess = transforms.Compose([transforms.Normalize(mean=[-1.0, -1.0, -1.0], std=[2.0, 2.0, 2.0])] + # Un-normalize from [-1.0, 1.0] (generator output) to [0, 1].
+                                              clip_preprocess.transforms[:2] +                                      # to match CLIP input scale assumptions
+                                              clip_preprocess.transforms[4:])                                       # + skip convert PIL to tensor
+
+    def tokenize(self, strings: list):
+        return clip.tokenize(strings).to(self.device)
+
+    @torch.no_grad()
+    def encode_text(self, tokens: list) -> torch.Tensor:
+        return self.model.encode_text(tokens)
+
+    @torch.no_grad()
+    def encode_images(self, images: torch.Tensor) -> torch.Tensor:
+        images = self.preprocess(images).to(self.device)
+        return self.model.encode_image(images)
+
+    def get_text_features(self, text: str, norm: bool = True) -> torch.Tensor:
+
+        tokens = clip.tokenize(text).to(self.device)
+
+        text_features = self.encode_text(tokens).detach()
+
+        if norm:
+            text_features /= text_features.norm(dim=-1, keepdim=True)
+
+        return text_features
+
+    def get_image_features(self, img: torch.Tensor, norm: bool = True) -> torch.Tensor:
+        image_features = self.encode_images(img)
+        
+        if norm:
+            image_features /= image_features.clone().norm(dim=-1, keepdim=True)
+
+        return image_features
+
+    def img_to_img_similarity(self, src_images, generated_images):
+        src_img_features = self.get_image_features(src_images)
+        gen_img_features = self.get_image_features(generated_images)
+
+        return (src_img_features @ gen_img_features.T).mean()
+
+    def txt_to_img_similarity(self, text, generated_images):
+        text_features    = self.get_text_features(text)
+        gen_img_features = self.get_image_features(generated_images)
+
+        return (text_features @ gen_img_features.T).mean()
+
+
+class LDMCLIPEvaluator(CLIPEvaluator):
+    def __init__(self, device, clip_model='ViT-B/32') -> None:
+        super().__init__(device, clip_model)
+
+    def evaluate(self, ldm_model, src_images, target_text, n_samples=64, n_steps=50):
+        
+        sampler = DDIMSampler(ldm_model)
+
+        samples_per_batch = 8
+        n_batches         = n_samples // samples_per_batch
+
+        # generate samples
+        all_samples=list()
+        with torch.no_grad():
+            with ldm_model.ema_scope():                
+                uc = ldm_model.get_learned_conditioning(samples_per_batch * [""])
+
+                for batch in range(n_batches):
+                    c = ldm_model.get_learned_conditioning(samples_per_batch * [target_text])
+                    shape = [4, 256//8, 256//8]
+                    samples_ddim, _ = sampler.sample(S=n_steps,
+                                                    conditioning=c,
+                                                    batch_size=samples_per_batch,
+                                                    shape=shape,
+                                                    verbose=False,
+                                                    unconditional_guidance_scale=5.0,
+                                                    unconditional_conditioning=uc,
+                                                    eta=0.0)
+
+                    x_samples_ddim = ldm_model.decode_first_stage(samples_ddim)
+                    x_samples_ddim = torch.clamp(x_samples_ddim, min=-1.0, max=1.0)
+
+                    all_samples.append(x_samples_ddim)
+        
+        all_samples = torch.cat(all_samples, axis=0)
+
+        sim_samples_to_img  = self.img_to_img_similarity(src_images, all_samples)
+        sim_samples_to_text = self.txt_to_img_similarity(target_text.replace("*", ""), all_samples)
+
+        return sim_samples_to_img, sim_samples_to_text
+
+
+class ImageDirEvaluator(CLIPEvaluator):
+    def __init__(self, device, clip_model='ViT-B/32') -> None:
+        super().__init__(device, clip_model)
+
+    def evaluate(self, gen_samples, src_images, target_text):
+
+        sim_samples_to_img  = self.img_to_img_similarity(src_images, gen_samples)
+        sim_samples_to_text = self.txt_to_img_similarity(target_text.replace("*", ""), gen_samples)
+
+        return sim_samples_to_img, sim_samples_to_text

ldm/data/base.py

@@ -0,0 +1,23 @@
+from abc import abstractmethod
+from torch.utils.data import Dataset, ConcatDataset, ChainDataset, IterableDataset
+
+
+class Txt2ImgIterableBaseDataset(IterableDataset):
+    '''
+    Define an interface to make the IterableDatasets for text2img data chainable
+    '''
+    def __init__(self, num_records=0, valid_ids=None, size=256):
+        super().__init__()
+        self.num_records = num_records
+        self.valid_ids = valid_ids
+        self.sample_ids = valid_ids
+        self.size = size
+
+        print(f'{self.__class__.__name__} dataset contains {self.__len__()} examples.')
+
+    def __len__(self):
+        return self.num_records
+
+    @abstractmethod
+    def __iter__(self):
+        pass

ldm/data/imagenet.py

@@ -0,0 +1,394 @@
+import os, yaml, pickle, shutil, tarfile, glob
+import cv2
+import albumentations
+import PIL
+import numpy as np
+import torchvision.transforms.functional as TF
+from omegaconf import OmegaConf
+from functools import partial
+from PIL import Image
+from tqdm import tqdm
+from torch.utils.data import Dataset, Subset
+
+import taming.data.utils as tdu
+from taming.data.imagenet import str_to_indices, give_synsets_from_indices, download, retrieve
+from taming.data.imagenet import ImagePaths
+
+from ldm.modules.image_degradation import degradation_fn_bsr, degradation_fn_bsr_light
+
+
+def synset2idx(path_to_yaml="data/index_synset.yaml"):
+    with open(path_to_yaml) as f:
+        di2s = yaml.load(f)
+    return dict((v,k) for k,v in di2s.items())
+
+
+class ImageNetBase(Dataset):
+    def __init__(self, config=None):
+        self.config = config or OmegaConf.create()
+        if not type(self.config)==dict:
+            self.config = OmegaConf.to_container(self.config)
+        self.keep_orig_class_label = self.config.get("keep_orig_class_label", False)
+        self.process_images = True  # if False we skip loading & processing images and self.data contains filepaths
+        self._prepare()
+        self._prepare_synset_to_human()
+        self._prepare_idx_to_synset()
+        self._prepare_human_to_integer_label()
+        self._load()
+
+    def __len__(self):
+        return len(self.data)
+
+    def __getitem__(self, i):
+        return self.data[i]
+
+    def _prepare(self):
+        raise NotImplementedError()
+
+    def _filter_relpaths(self, relpaths):
+        ignore = set([
+            "n06596364_9591.JPEG",
+        ])
+        relpaths = [rpath for rpath in relpaths if not rpath.split("/")[-1] in ignore]
+        if "sub_indices" in self.config:
+            indices = str_to_indices(self.config["sub_indices"])
+            synsets = give_synsets_from_indices(indices, path_to_yaml=self.idx2syn)  # returns a list of strings
+            self.synset2idx = synset2idx(path_to_yaml=self.idx2syn)
+            files = []
+            for rpath in relpaths:
+                syn = rpath.split("/")[0]
+                if syn in synsets:
+                    files.append(rpath)
+            return files
+        else:
+            return relpaths
+
+    def _prepare_synset_to_human(self):
+        SIZE = 2655750
+        URL = "https://heibox.uni-heidelberg.de/f/9f28e956cd304264bb82/?dl=1"
+        self.human_dict = os.path.join(self.root, "synset_human.txt")
+        if (not os.path.exists(self.human_dict) or
+                not os.path.getsize(self.human_dict)==SIZE):
+            download(URL, self.human_dict)
+
+    def _prepare_idx_to_synset(self):
+        URL = "https://heibox.uni-heidelberg.de/f/d835d5b6ceda4d3aa910/?dl=1"
+        self.idx2syn = os.path.join(self.root, "index_synset.yaml")
+        if (not os.path.exists(self.idx2syn)):
+            download(URL, self.idx2syn)
+
+    def _prepare_human_to_integer_label(self):
+        URL = "https://heibox.uni-heidelberg.de/f/2362b797d5be43b883f6/?dl=1"
+        self.human2integer = os.path.join(self.root, "imagenet1000_clsidx_to_labels.txt")
+        if (not os.path.exists(self.human2integer)):
+            download(URL, self.human2integer)
+        with open(self.human2integer, "r") as f:
+            lines = f.read().splitlines()
+            assert len(lines) == 1000
+            self.human2integer_dict = dict()
+            for line in lines:
+                value, key = line.split(":")
+                self.human2integer_dict[key] = int(value)
+
+    def _load(self):
+        with open(self.txt_filelist, "r") as f:
+            self.relpaths = f.read().splitlines()
+            l1 = len(self.relpaths)
+            self.relpaths = self._filter_relpaths(self.relpaths)
+            print("Removed {} files from filelist during filtering.".format(l1 - len(self.relpaths)))
+
+        self.synsets = [p.split("/")[0] for p in self.relpaths]
+        self.abspaths = [os.path.join(self.datadir, p) for p in self.relpaths]
+
+        unique_synsets = np.unique(self.synsets)
+        class_dict = dict((synset, i) for i, synset in enumerate(unique_synsets))
+        if not self.keep_orig_class_label:
+            self.class_labels = [class_dict[s] for s in self.synsets]
+        else:
+            self.class_labels = [self.synset2idx[s] for s in self.synsets]
+
+        with open(self.human_dict, "r") as f:
+            human_dict = f.read().splitlines()
+            human_dict = dict(line.split(maxsplit=1) for line in human_dict)
+
+        self.human_labels = [human_dict[s] for s in self.synsets]
+
+        labels = {
+            "relpath": np.array(self.relpaths),
+            "synsets": np.array(self.synsets),
+            "class_label": np.array(self.class_labels),
+            "human_label": np.array(self.human_labels),
+        }
+
+        if self.process_images:
+            self.size = retrieve(self.config, "size", default=256)
+            self.data = ImagePaths(self.abspaths,
+                                   labels=labels,
+                                   size=self.size,
+                                   random_crop=self.random_crop,
+                                   )
+        else:
+            self.data = self.abspaths
+
+
+class ImageNetTrain(ImageNetBase):
+    NAME = "ILSVRC2012_train"
+    URL = "http://www.image-net.org/challenges/LSVRC/2012/"
+    AT_HASH = "a306397ccf9c2ead27155983c254227c0fd938e2"
+    FILES = [
+        "ILSVRC2012_img_train.tar",
+    ]
+    SIZES = [
+        147897477120,
+    ]
+
+    def __init__(self, process_images=True, data_root=None, **kwargs):
+        self.process_images = process_images
+        self.data_root = data_root
+        super().__init__(**kwargs)
+
+    def _prepare(self):
+        if self.data_root:
+            self.root = os.path.join(self.data_root, self.NAME)
+        else:
+            cachedir = os.environ.get("XDG_CACHE_HOME", os.path.expanduser("~/.cache"))
+            self.root = os.path.join(cachedir, "autoencoders/data", self.NAME)
+
+        self.datadir = os.path.join(self.root, "data")
+        self.txt_filelist = os.path.join(self.root, "filelist.txt")
+        self.expected_length = 1281167
+        self.random_crop = retrieve(self.config, "ImageNetTrain/random_crop",
+                                    default=True)
+        if not tdu.is_prepared(self.root):
+            # prep
+            print("Preparing dataset {} in {}".format(self.NAME, self.root))
+
+            datadir = self.datadir
+            if not os.path.exists(datadir):
+                path = os.path.join(self.root, self.FILES[0])
+                if not os.path.exists(path) or not os.path.getsize(path)==self.SIZES[0]:
+                    import academictorrents as at
+                    atpath = at.get(self.AT_HASH, datastore=self.root)
+                    assert atpath == path
+
+                print("Extracting {} to {}".format(path, datadir))
+                os.makedirs(datadir, exist_ok=True)
+                with tarfile.open(path, "r:") as tar:
+                    tar.extractall(path=datadir)
+
+                print("Extracting sub-tars.")
+                subpaths = sorted(glob.glob(os.path.join(datadir, "*.tar")))
+                for subpath in tqdm(subpaths):
+                    subdir = subpath[:-len(".tar")]
+                    os.makedirs(subdir, exist_ok=True)
+                    with tarfile.open(subpath, "r:") as tar:
+                        tar.extractall(path=subdir)
+
+            filelist = glob.glob(os.path.join(datadir, "**", "*.JPEG"))
+            filelist = [os.path.relpath(p, start=datadir) for p in filelist]
+            filelist = sorted(filelist)
+            filelist = "\n".join(filelist)+"\n"
+            with open(self.txt_filelist, "w") as f:
+                f.write(filelist)
+
+            tdu.mark_prepared(self.root)
+
+
+class ImageNetValidation(ImageNetBase):
+    NAME = "ILSVRC2012_validation"
+    URL = "http://www.image-net.org/challenges/LSVRC/2012/"
+    AT_HASH = "5d6d0df7ed81efd49ca99ea4737e0ae5e3a5f2e5"
+    VS_URL = "https://heibox.uni-heidelberg.de/f/3e0f6e9c624e45f2bd73/?dl=1"
+    FILES = [
+        "ILSVRC2012_img_val.tar",
+        "validation_synset.txt",
+    ]
+    SIZES = [
+        6744924160,
+        1950000,
+    ]
+
+    def __init__(self, process_images=True, data_root=None, **kwargs):
+        self.data_root = data_root
+        self.process_images = process_images
+        super().__init__(**kwargs)
+
+    def _prepare(self):
+        if self.data_root:
+            self.root = os.path.join(self.data_root, self.NAME)
+        else:
+            cachedir = os.environ.get("XDG_CACHE_HOME", os.path.expanduser("~/.cache"))
+            self.root = os.path.join(cachedir, "autoencoders/data", self.NAME)
+        self.datadir = os.path.join(self.root, "data")
+        self.txt_filelist = os.path.join(self.root, "filelist.txt")
+        self.expected_length = 50000
+        self.random_crop = retrieve(self.config, "ImageNetValidation/random_crop",
+                                    default=False)
+        if not tdu.is_prepared(self.root):
+            # prep
+            print("Preparing dataset {} in {}".format(self.NAME, self.root))
+
+            datadir = self.datadir
+            if not os.path.exists(datadir):
+                path = os.path.join(self.root, self.FILES[0])
+                if not os.path.exists(path) or not os.path.getsize(path)==self.SIZES[0]:
+                    import academictorrents as at
+                    atpath = at.get(self.AT_HASH, datastore=self.root)
+                    assert atpath == path
+
+                print("Extracting {} to {}".format(path, datadir))
+                os.makedirs(datadir, exist_ok=True)
+                with tarfile.open(path, "r:") as tar:
+                    tar.extractall(path=datadir)
+
+                vspath = os.path.join(self.root, self.FILES[1])
+                if not os.path.exists(vspath) or not os.path.getsize(vspath)==self.SIZES[1]:
+                    download(self.VS_URL, vspath)
+
+                with open(vspath, "r") as f:
+                    synset_dict = f.read().splitlines()
+                    synset_dict = dict(line.split() for line in synset_dict)
+
+                print("Reorganizing into synset folders")
+                synsets = np.unique(list(synset_dict.values()))
+                for s in synsets:
+                    os.makedirs(os.path.join(datadir, s), exist_ok=True)
+                for k, v in synset_dict.items():
+                    src = os.path.join(datadir, k)
+                    dst = os.path.join(datadir, v)
+                    shutil.move(src, dst)
+
+            filelist = glob.glob(os.path.join(datadir, "**", "*.JPEG"))
+            filelist = [os.path.relpath(p, start=datadir) for p in filelist]
+            filelist = sorted(filelist)
+            filelist = "\n".join(filelist)+"\n"
+            with open(self.txt_filelist, "w") as f:
+                f.write(filelist)
+
+            tdu.mark_prepared(self.root)
+
+
+
+class ImageNetSR(Dataset):
+    def __init__(self, size=None,
+                 degradation=None, downscale_f=4, min_crop_f=0.5, max_crop_f=1.,
+                 random_crop=True):
+        """
+        Imagenet Superresolution Dataloader
+        Performs following ops in order:
+        1.  crops a crop of size s from image either as random or center crop
+        2.  resizes crop to size with cv2.area_interpolation
+        3.  degrades resized crop with degradation_fn
+
+        :param size: resizing to size after cropping
+        :param degradation: degradation_fn, e.g. cv_bicubic or bsrgan_light
+        :param downscale_f: Low Resolution Downsample factor
+        :param min_crop_f: determines crop size s,
+          where s = c * min_img_side_len with c sampled from interval (min_crop_f, max_crop_f)
+        :param max_crop_f: ""
+        :param data_root:
+        :param random_crop:
+        """
+        self.base = self.get_base()
+        assert size
+        assert (size / downscale_f).is_integer()
+        self.size = size
+        self.LR_size = int(size / downscale_f)
+        self.min_crop_f = min_crop_f
+        self.max_crop_f = max_crop_f
+        assert(max_crop_f <= 1.)
+        self.center_crop = not random_crop
+
+        self.image_rescaler = albumentations.SmallestMaxSize(max_size=size, interpolation=cv2.INTER_AREA)
+
+        self.pil_interpolation = False # gets reset later if incase interp_op is from pillow
+
+        if degradation == "bsrgan":
+            self.degradation_process = partial(degradation_fn_bsr, sf=downscale_f)
+
+        elif degradation == "bsrgan_light":
+            self.degradation_process = partial(degradation_fn_bsr_light, sf=downscale_f)
+
+        else:
+            interpolation_fn = {
+            "cv_nearest": cv2.INTER_NEAREST,
+            "cv_bilinear": cv2.INTER_LINEAR,
+            "cv_bicubic": cv2.INTER_CUBIC,
+            "cv_area": cv2.INTER_AREA,
+            "cv_lanczos": cv2.INTER_LANCZOS4,
+            "pil_nearest": PIL.Image.NEAREST,
+            "pil_bilinear": PIL.Image.BILINEAR,
+            "pil_bicubic": PIL.Image.BICUBIC,
+            "pil_box": PIL.Image.BOX,
+            "pil_hamming": PIL.Image.HAMMING,
+            "pil_lanczos": PIL.Image.LANCZOS,
+            }[degradation]
+
+            self.pil_interpolation = degradation.startswith("pil_")
+
+            if self.pil_interpolation:
+                self.degradation_process = partial(TF.resize, size=self.LR_size, interpolation=interpolation_fn)
+
+            else:
+                self.degradation_process = albumentations.SmallestMaxSize(max_size=self.LR_size,
+                                                                          interpolation=interpolation_fn)
+
+    def __len__(self):
+        return len(self.base)
+
+    def __getitem__(self, i):
+        example = self.base[i]
+        image = Image.open(example["file_path_"])
+
+        if not image.mode == "RGB":
+            image = image.convert("RGB")
+
+        image = np.array(image).astype(np.uint8)
+
+        min_side_len = min(image.shape[:2])
+        crop_side_len = min_side_len * np.random.uniform(self.min_crop_f, self.max_crop_f, size=None)
+        crop_side_len = int(crop_side_len)
+
+        if self.center_crop:
+            self.cropper = albumentations.CenterCrop(height=crop_side_len, width=crop_side_len)
+
+        else:
+            self.cropper = albumentations.RandomCrop(height=crop_side_len, width=crop_side_len)
+
+        image = self.cropper(image=image)["image"]
+        image = self.image_rescaler(image=image)["image"]
+
+        if self.pil_interpolation:
+            image_pil = PIL.Image.fromarray(image)
+            LR_image = self.degradation_process(image_pil)
+            LR_image = np.array(LR_image).astype(np.uint8)
+
+        else:
+            LR_image = self.degradation_process(image=image)["image"]
+
+        example["image"] = (image/127.5 - 1.0).astype(np.float32)
+        example["LR_image"] = (LR_image/127.5 - 1.0).astype(np.float32)
+
+        return example
+
+
+class ImageNetSRTrain(ImageNetSR):
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+
+    def get_base(self):
+        with open("data/imagenet_train_hr_indices.p", "rb") as f:
+            indices = pickle.load(f)
+        dset = ImageNetTrain(process_images=False,)
+        return Subset(dset, indices)
+
+
+class ImageNetSRValidation(ImageNetSR):
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+
+    def get_base(self):
+        with open("data/imagenet_val_hr_indices.p", "rb") as f:
+            indices = pickle.load(f)
+        dset = ImageNetValidation(process_images=False,)
+        return Subset(dset, indices)

ldm/data/lsun.py

@@ -0,0 +1,92 @@
+import os
+import numpy as np
+import PIL
+from PIL import Image
+from torch.utils.data import Dataset
+from torchvision import transforms
+
+
+class LSUNBase(Dataset):
+    def __init__(self,
+                 txt_file,
+                 data_root,
+                 size=None,
+                 interpolation="bicubic",
+                 flip_p=0.5
+                 ):
+        self.data_paths = txt_file
+        self.data_root = data_root
+        with open(self.data_paths, "r") as f:
+            self.image_paths = f.read().splitlines()
+        self._length = len(self.image_paths)
+        self.labels = {
+            "relative_file_path_": [l for l in self.image_paths],
+            "file_path_": [os.path.join(self.data_root, l)
+                           for l in self.image_paths],
+        }
+
+        self.size = size
+        self.interpolation = {"linear": PIL.Image.LINEAR,
+                              "bilinear": PIL.Image.BILINEAR,
+                              "bicubic": PIL.Image.BICUBIC,
+                              "lanczos": PIL.Image.LANCZOS,
+                              }[interpolation]
+        self.flip = transforms.RandomHorizontalFlip(p=flip_p)
+
+    def __len__(self):
+        return self._length
+
+    def __getitem__(self, i):
+        example = dict((k, self.labels[k][i]) for k in self.labels)
+        image = Image.open(example["file_path_"])
+        if not image.mode == "RGB":
+            image = image.convert("RGB")
+
+        # default to score-sde preprocessing
+        img = np.array(image).astype(np.uint8)
+        crop = min(img.shape[0], img.shape[1])
+        h, w, = img.shape[0], img.shape[1]
+        img = img[(h - crop) // 2:(h + crop) // 2,
+              (w - crop) // 2:(w + crop) // 2]
+
+        image = Image.fromarray(img)
+        if self.size is not None:
+            image = image.resize((self.size, self.size), resample=self.interpolation)
+
+        image = self.flip(image)
+        image = np.array(image).astype(np.uint8)
+        example["image"] = (image / 127.5 - 1.0).astype(np.float32)
+        return example
+
+
+class LSUNChurchesTrain(LSUNBase):
+    def __init__(self, **kwargs):
+        super().__init__(txt_file="data/lsun/church_outdoor_train.txt", data_root="data/lsun/churches", **kwargs)
+
+
+class LSUNChurchesValidation(LSUNBase):
+    def __init__(self, flip_p=0., **kwargs):
+        super().__init__(txt_file="data/lsun/church_outdoor_val.txt", data_root="data/lsun/churches",
+                         flip_p=flip_p, **kwargs)
+
+
+class LSUNBedroomsTrain(LSUNBase):
+    def __init__(self, **kwargs):
+        super().__init__(txt_file="data/lsun/bedrooms_train.txt", data_root="data/lsun/bedrooms", **kwargs)
+
+
+class LSUNBedroomsValidation(LSUNBase):
+    def __init__(self, flip_p=0.0, **kwargs):
+        super().__init__(txt_file="data/lsun/bedrooms_val.txt", data_root="data/lsun/bedrooms",
+                         flip_p=flip_p, **kwargs)
+
+
+class LSUNCatsTrain(LSUNBase):
+    def __init__(self, **kwargs):
+        super().__init__(txt_file="data/lsun/cat_train.txt", data_root="data/lsun/cats", **kwargs)
+
+
+class LSUNCatsValidation(LSUNBase):
+    def __init__(self, flip_p=0., **kwargs):
+        super().__init__(txt_file="data/lsun/cat_val.txt", data_root="data/lsun/cats",
+                         flip_p=flip_p, **kwargs)

ldm/data/personalized.py

@@ -0,0 +1,115 @@
+import os
+import numpy as np
+import PIL
+from PIL import Image
+from torch.utils.data import Dataset
+from torchvision import transforms
+
+import random
+
+training_templates_smallest = [
+    'gijsbert {}',
+]
+
+reg_templates_smallest = [
+    '{}',
+]
+
+imagenet_templates_small = [
+
+    '{}',
+]
+
+imagenet_dual_templates_small = [
+    '{} with {}'
+]
+
+per_img_token_list = [
+    'א', 'ב', 'ג', 'ד', 'ה', 'ו', 'ז', 'ח', 'ט', 'י', 'כ', 'ל', 'מ', 'נ', 'ס', 'ע', 'פ', 'צ', 'ק', 'ר', 'ש', 'ת',
+]
+
+class PersonalizedBase(Dataset):
+    def __init__(self,
+                 data_root,
+                 size=None,
+                 repeats=100,
+                 interpolation="bicubic",
+                 flip_p=0.5,
+                 set="train",
+                 placeholder_token="dog",
+                 per_image_tokens=False,
+                 center_crop=False,
+                 mixing_prob=0.25,
+                 coarse_class_text=None,
+                 reg = False
+                 ):
+
+        self.data_root = data_root
+
+        self.image_paths = [os.path.join(self.data_root, file_path) for file_path in os.listdir(self.data_root)]
+
+        # self._length = len(self.image_paths)
+        self.num_images = len(self.image_paths)
+        self._length = self.num_images 
+
+        self.placeholder_token = placeholder_token
+
+        self.per_image_tokens = per_image_tokens
+        self.center_crop = center_crop
+        self.mixing_prob = mixing_prob
+
+        self.coarse_class_text = coarse_class_text
+
+        if per_image_tokens:
+            assert self.num_images < len(per_img_token_list), f"Can't use per-image tokens when the training set contains more than {len(per_img_token_list)} tokens. To enable larger sets, add more tokens to 'per_img_token_list'."
+
+        if set == "train":
+            self._length = self.num_images * repeats
+
+        self.size = size
+        self.interpolation = {"linear": PIL.Image.LINEAR,
+                              "bilinear": PIL.Image.BILINEAR,
+                              "bicubic": PIL.Image.BICUBIC,
+                              "lanczos": PIL.Image.LANCZOS,
+                              }[interpolation]
+        self.flip = transforms.RandomHorizontalFlip(p=flip_p)
+        self.reg = reg
+
+    def __len__(self):
+        return self._length
+
+    def __getitem__(self, i):
+        example = {}
+        image = Image.open(self.image_paths[i % self.num_images])
+
+        if not image.mode == "RGB":
+            image = image.convert("RGB")
+
+        placeholder_string = self.placeholder_token
+        if self.coarse_class_text:
+            placeholder_string = f"{self.coarse_class_text} {placeholder_string}"
+
+        if not self.reg:
+            text = random.choice(training_templates_smallest).format(placeholder_string)
+        else:
+            text = random.choice(reg_templates_smallest).format(placeholder_string)
+            
+        example["caption"] = text
+
+        # default to score-sde preprocessing
+        img = np.array(image).astype(np.uint8)
+        
+        if self.center_crop:
+            crop = min(img.shape[0], img.shape[1])
+            h, w, = img.shape[0], img.shape[1]
+            img = img[(h - crop) // 2:(h + crop) // 2,
+                (w - crop) // 2:(w + crop) // 2]
+
+        image = Image.fromarray(img)
+        if self.size is not None:
+            image = image.resize((self.size, self.size), resample=self.interpolation)
+
+        image = self.flip(image)
+        image = np.array(image).astype(np.uint8)
+        example["image"] = (image / 127.5 - 1.0).astype(np.float32)
+        return example

ldm/data/personalized_style.py

@@ -0,0 +1,129 @@
+import os
+import numpy as np
+import PIL
+from PIL import Image
+from torch.utils.data import Dataset
+from torchvision import transforms
+
+import random
+
+imagenet_templates_small = [
+    'a painting in the style of {}',
+    'a rendering in the style of {}',
+    'a cropped painting in the style of {}',
+    'the painting in the style of {}',
+    'a clean painting in the style of {}',
+    'a dirty painting in the style of {}',
+    'a dark painting in the style of {}',
+    'a picture in the style of {}',
+    'a cool painting in the style of {}',
+    'a close-up painting in the style of {}',
+    'a bright painting in the style of {}',
+    'a cropped painting in the style of {}',
+    'a good painting in the style of {}',
+    'a close-up painting in the style of {}',
+    'a rendition in the style of {}',
+    'a nice painting in the style of {}',
+    'a small painting in the style of {}',
+    'a weird painting in the style of {}',
+    'a large painting in the style of {}',
+]
+
+imagenet_dual_templates_small = [
+    'a painting in the style of {} with {}',
+    'a rendering in the style of {} with {}',
+    'a cropped painting in the style of {} with {}',
+    'the painting in the style of {} with {}',
+    'a clean painting in the style of {} with {}',
+    'a dirty painting in the style of {} with {}',
+    'a dark painting in the style of {} with {}',
+    'a cool painting in the style of {} with {}',
+    'a close-up painting in the style of {} with {}',
+    'a bright painting in the style of {} with {}',
+    'a cropped painting in the style of {} with {}',
+    'a good painting in the style of {} with {}',
+    'a painting of one {} in the style of {}',
+    'a nice painting in the style of {} with {}',
+    'a small painting in the style of {} with {}',
+    'a weird painting in the style of {} with {}',
+    'a large painting in the style of {} with {}',
+]
+
+per_img_token_list = [
+    'א', 'ב', 'ג', 'ד', 'ה', 'ו', 'ז', 'ח', 'ט', 'י', 'כ', 'ל', 'מ', 'נ', 'ס', 'ע', 'פ', 'צ', 'ק', 'ר', 'ש', 'ת',
+]
+
+class PersonalizedBase(Dataset):
+    def __init__(self,
+                 data_root,
+                 size=None,
+                 repeats=100,
+                 interpolation="bicubic",
+                 flip_p=0.5,
+                 set="train",
+                 placeholder_token="*",
+                 per_image_tokens=False,
+                 center_crop=False,
+                 ):
+
+        self.data_root = data_root
+
+        self.image_paths = [os.path.join(self.data_root, file_path) for file_path in os.listdir(self.data_root)]
+
+        # self._length = len(self.image_paths)
+        self.num_images = len(self.image_paths)
+        self._length = self.num_images 
+
+        self.placeholder_token = placeholder_token
+
+        self.per_image_tokens = per_image_tokens
+        self.center_crop = center_crop
+
+        if per_image_tokens:
+            assert self.num_images < len(per_img_token_list), f"Can't use per-image tokens when the training set contains more than {len(per_img_token_list)} tokens. To enable larger sets, add more tokens to 'per_img_token_list'."
+
+        if set == "train":
+            self._length = self.num_images * repeats
+
+        self.size = size
+        self.interpolation = {"linear": PIL.Image.LINEAR,
+                              "bilinear": PIL.Image.BILINEAR,
+                              "bicubic": PIL.Image.BICUBIC,
+                              "lanczos": PIL.Image.LANCZOS,
+                              }[interpolation]
+        self.flip = transforms.RandomHorizontalFlip(p=flip_p)
+
+    def __len__(self):
+        return self._length
+
+    def __getitem__(self, i):
+        example = {}
+        image = Image.open(self.image_paths[i % self.num_images])
+
+        if not image.mode == "RGB":
+            image = image.convert("RGB")
+
+        if self.per_image_tokens and np.random.uniform() < 0.25:
+            text = random.choice(imagenet_dual_templates_small).format(self.placeholder_token, per_img_token_list[i % self.num_images])
+        else:
+            text = random.choice(imagenet_templates_small).format(self.placeholder_token)
+            
+        example["caption"] = text
+
+        # default to score-sde preprocessing
+        img = np.array(image).astype(np.uint8)
+        
+        if self.center_crop:
+            crop = min(img.shape[0], img.shape[1])
+            h, w, = img.shape[0], img.shape[1]
+            img = img[(h - crop) // 2:(h + crop) // 2,
+                (w - crop) // 2:(w + crop) // 2]
+
+        image = Image.fromarray(img)
+        if self.size is not None:
+            image = image.resize((self.size, self.size), resample=self.interpolation)
+
+        image = self.flip(image)
+        image = np.array(image).astype(np.uint8)
+        example["image"] = (image / 127.5 - 1.0).astype(np.float32)
+        return example

ldm/lr_scheduler.py

@@ -0,0 +1,98 @@
+import numpy as np
+
+
+class LambdaWarmUpCosineScheduler:
+    """
+    note: use with a base_lr of 1.0
+    """
+    def __init__(self, warm_up_steps, lr_min, lr_max, lr_start, max_decay_steps, verbosity_interval=0):
+        self.lr_warm_up_steps = warm_up_steps
+        self.lr_start = lr_start
+        self.lr_min = lr_min
+        self.lr_max = lr_max
+        self.lr_max_decay_steps = max_decay_steps
+        self.last_lr = 0.
+        self.verbosity_interval = verbosity_interval
+
+    def schedule(self, n, **kwargs):
+        if self.verbosity_interval > 0:
+            if n % self.verbosity_interval == 0: print(f"current step: {n}, recent lr-multiplier: {self.last_lr}")
+        if n < self.lr_warm_up_steps:
+            lr = (self.lr_max - self.lr_start) / self.lr_warm_up_steps * n + self.lr_start
+            self.last_lr = lr
+            return lr
+        else:
+            t = (n - self.lr_warm_up_steps) / (self.lr_max_decay_steps - self.lr_warm_up_steps)
+            t = min(t, 1.0)
+            lr = self.lr_min + 0.5 * (self.lr_max - self.lr_min) * (
+                    1 + np.cos(t * np.pi))
+            self.last_lr = lr
+            return lr
+
+    def __call__(self, n, **kwargs):
+        return self.schedule(n,**kwargs)
+
+
+class LambdaWarmUpCosineScheduler2:
+    """
+    supports repeated iterations, configurable via lists
+    note: use with a base_lr of 1.0.
+    """
+    def __init__(self, warm_up_steps, f_min, f_max, f_start, cycle_lengths, verbosity_interval=0):
+        assert len(warm_up_steps) == len(f_min) == len(f_max) == len(f_start) == len(cycle_lengths)
+        self.lr_warm_up_steps = warm_up_steps
+        self.f_start = f_start
+        self.f_min = f_min
+        self.f_max = f_max
+        self.cycle_lengths = cycle_lengths
+        self.cum_cycles = np.cumsum([0] + list(self.cycle_lengths))
+        self.last_f = 0.
+        self.verbosity_interval = verbosity_interval
+
+    def find_in_interval(self, n):
+        interval = 0
+        for cl in self.cum_cycles[1:]:
+            if n <= cl:
+                return interval
+            interval += 1
+
+    def schedule(self, n, **kwargs):
+        cycle = self.find_in_interval(n)
+        n = n - self.cum_cycles[cycle]
+        if self.verbosity_interval > 0:
+            if n % self.verbosity_interval == 0: print(f"current step: {n}, recent lr-multiplier: {self.last_f}, "
+                                                       f"current cycle {cycle}")
+        if n < self.lr_warm_up_steps[cycle]:
+            f = (self.f_max[cycle] - self.f_start[cycle]) / self.lr_warm_up_steps[cycle] * n + self.f_start[cycle]
+            self.last_f = f
+            return f
+        else:
+            t = (n - self.lr_warm_up_steps[cycle]) / (self.cycle_lengths[cycle] - self.lr_warm_up_steps[cycle])
+            t = min(t, 1.0)
+            f = self.f_min[cycle] + 0.5 * (self.f_max[cycle] - self.f_min[cycle]) * (
+                    1 + np.cos(t * np.pi))
+            self.last_f = f
+            return f
+
+    def __call__(self, n, **kwargs):
+        return self.schedule(n, **kwargs)
+
+
+class LambdaLinearScheduler(LambdaWarmUpCosineScheduler2):
+
+    def schedule(self, n, **kwargs):
+        cycle = self.find_in_interval(n)
+        n = n - self.cum_cycles[cycle]
+        if self.verbosity_interval > 0:
+            if n % self.verbosity_interval == 0: print(f"current step: {n}, recent lr-multiplier: {self.last_f}, "
+                                                       f"current cycle {cycle}")
+
+        if n < self.lr_warm_up_steps[cycle]:
+            f = (self.f_max[cycle] - self.f_start[cycle]) / self.lr_warm_up_steps[cycle] * n + self.f_start[cycle]
+            self.last_f = f
+            return f
+        else:
+            f = self.f_min[cycle] + (self.f_max[cycle] - self.f_min[cycle]) * (self.cycle_lengths[cycle] - n) / (self.cycle_lengths[cycle])
+            self.last_f = f
+            return f
+

ldm/models/autoencoder.py

@@ -0,0 +1,443 @@
+import torch
+import pytorch_lightning as pl
+import torch.nn.functional as F
+from contextlib import contextmanager
+
+from taming.modules.vqvae.quantize import VectorQuantizer2 as VectorQuantizer
+
+from ldm.modules.diffusionmodules.model import Encoder, Decoder
+from ldm.modules.distributions.distributions import DiagonalGaussianDistribution
+
+from ldm.util import instantiate_from_config
+
+
+class VQModel(pl.LightningModule):
+    def __init__(self,
+                 ddconfig,
+                 lossconfig,
+                 n_embed,
+                 embed_dim,
+                 ckpt_path=None,
+                 ignore_keys=[],
+                 image_key="image",
+                 colorize_nlabels=None,
+                 monitor=None,
+                 batch_resize_range=None,
+                 scheduler_config=None,
+                 lr_g_factor=1.0,
+                 remap=None,
+                 sane_index_shape=False, # tell vector quantizer to return indices as bhw
+                 use_ema=False
+                 ):
+        super().__init__()
+        self.embed_dim = embed_dim
+        self.n_embed = n_embed
+        self.image_key = image_key
+        self.encoder = Encoder(**ddconfig)
+        self.decoder = Decoder(**ddconfig)
+        self.loss = instantiate_from_config(lossconfig)
+        self.quantize = VectorQuantizer(n_embed, embed_dim, beta=0.25,
+                                        remap=remap,
+                                        sane_index_shape=sane_index_shape)
+        self.quant_conv = torch.nn.Conv2d(ddconfig["z_channels"], embed_dim, 1)
+        self.post_quant_conv = torch.nn.Conv2d(embed_dim, ddconfig["z_channels"], 1)
+        if colorize_nlabels is not None:
+            assert type(colorize_nlabels)==int
+            self.register_buffer("colorize", torch.randn(3, colorize_nlabels, 1, 1))
+        if monitor is not None:
+            self.monitor = monitor
+        self.batch_resize_range = batch_resize_range
+        if self.batch_resize_range is not None:
+            print(f"{self.__class__.__name__}: Using per-batch resizing in range {batch_resize_range}.")
+
+        self.use_ema = use_ema
+        if self.use_ema:
+            self.model_ema = LitEma(self)
+            print(f"Keeping EMAs of {len(list(self.model_ema.buffers()))}.")
+
+        if ckpt_path is not None:
+            self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys)
+        self.scheduler_config = scheduler_config
+        self.lr_g_factor = lr_g_factor
+
+    @contextmanager
+    def ema_scope(self, context=None):
+        if self.use_ema:
+            self.model_ema.store(self.parameters())
+            self.model_ema.copy_to(self)
+            if context is not None:
+                print(f"{context}: Switched to EMA weights")
+        try:
+            yield None
+        finally:
+            if self.use_ema:
+                self.model_ema.restore(self.parameters())
+                if context is not None:
+                    print(f"{context}: Restored training weights")
+
+    def init_from_ckpt(self, path, ignore_keys=list()):
+        sd = torch.load(path, map_location="cpu")["state_dict"]
+        keys = list(sd.keys())
+        for k in keys:
+            for ik in ignore_keys:
+                if k.startswith(ik):
+                    print("Deleting key {} from state_dict.".format(k))
+                    del sd[k]
+        missing, unexpected = self.load_state_dict(sd, strict=False)
+        print(f"Restored from {path} with {len(missing)} missing and {len(unexpected)} unexpected keys")
+        if len(missing) > 0:
+            print(f"Missing Keys: {missing}")
+            print(f"Unexpected Keys: {unexpected}")
+
+    def on_train_batch_end(self, *args, **kwargs):
+        if self.use_ema:
+            self.model_ema(self)
+
+    def encode(self, x):
+        h = self.encoder(x)
+        h = self.quant_conv(h)
+        quant, emb_loss, info = self.quantize(h)
+        return quant, emb_loss, info
+
+    def encode_to_prequant(self, x):
+        h = self.encoder(x)
+        h = self.quant_conv(h)
+        return h
+
+    def decode(self, quant):
+        quant = self.post_quant_conv(quant)
+        dec = self.decoder(quant)
+        return dec
+
+    def decode_code(self, code_b):
+        quant_b = self.quantize.embed_code(code_b)
+        dec = self.decode(quant_b)
+        return dec
+
+    def forward(self, input, return_pred_indices=False):
+        quant, diff, (_,_,ind) = self.encode(input)
+        dec = self.decode(quant)
+        if return_pred_indices:
+            return dec, diff, ind
+        return dec, diff
+
+    def get_input(self, batch, k):
+        x = batch[k]
+        if len(x.shape) == 3:
+            x = x[..., None]
+        x = x.permute(0, 3, 1, 2).to(memory_format=torch.contiguous_format).float()
+        if self.batch_resize_range is not None:
+            lower_size = self.batch_resize_range[0]
+            upper_size = self.batch_resize_range[1]
+            if self.global_step <= 4:
+                # do the first few batches with max size to avoid later oom
+                new_resize = upper_size
+            else:
+                new_resize = np.random.choice(np.arange(lower_size, upper_size+16, 16))
+            if new_resize != x.shape[2]:
+                x = F.interpolate(x, size=new_resize, mode="bicubic")
+            x = x.detach()
+        return x
+
+    def training_step(self, batch, batch_idx, optimizer_idx):
+        # https://github.com/pytorch/pytorch/issues/37142
+        # try not to fool the heuristics
+        x = self.get_input(batch, self.image_key)
+        xrec, qloss, ind = self(x, return_pred_indices=True)
+
+        if optimizer_idx == 0:
+            # autoencode
+            aeloss, log_dict_ae = self.loss(qloss, x, xrec, optimizer_idx, self.global_step,
+                                            last_layer=self.get_last_layer(), split="train",
+                                            predicted_indices=ind)
+
+            self.log_dict(log_dict_ae, prog_bar=False, logger=True, on_step=True, on_epoch=True)
+            return aeloss
+
+        if optimizer_idx == 1:
+            # discriminator
+            discloss, log_dict_disc = self.loss(qloss, x, xrec, optimizer_idx, self.global_step,
+                                            last_layer=self.get_last_layer(), split="train")
+            self.log_dict(log_dict_disc, prog_bar=False, logger=True, on_step=True, on_epoch=True)
+            return discloss
+
+    def validation_step(self, batch, batch_idx):
+        log_dict = self._validation_step(batch, batch_idx)
+        with self.ema_scope():
+            log_dict_ema = self._validation_step(batch, batch_idx, suffix="_ema")
+        return log_dict
+
+    def _validation_step(self, batch, batch_idx, suffix=""):
+        x = self.get_input(batch, self.image_key)
+        xrec, qloss, ind = self(x, return_pred_indices=True)
+        aeloss, log_dict_ae = self.loss(qloss, x, xrec, 0,
+                                        self.global_step,
+                                        last_layer=self.get_last_layer(),
+                                        split="val"+suffix,
+                                        predicted_indices=ind
+                                        )
+
+        discloss, log_dict_disc = self.loss(qloss, x, xrec, 1,
+                                            self.global_step,
+                                            last_layer=self.get_last_layer(),
+                                            split="val"+suffix,
+                                            predicted_indices=ind
+                                            )
+        rec_loss = log_dict_ae[f"val{suffix}/rec_loss"]
+        self.log(f"val{suffix}/rec_loss", rec_loss,
+                   prog_bar=True, logger=True, on_step=False, on_epoch=True, sync_dist=True)
+        self.log(f"val{suffix}/aeloss", aeloss,
+                   prog_bar=True, logger=True, on_step=False, on_epoch=True, sync_dist=True)
+        if version.parse(pl.__version__) >= version.parse('1.4.0'):
+            del log_dict_ae[f"val{suffix}/rec_loss"]
+        self.log_dict(log_dict_ae)
+        self.log_dict(log_dict_disc)
+        return self.log_dict
+
+    def configure_optimizers(self):
+        lr_d = self.learning_rate
+        lr_g = self.lr_g_factor*self.learning_rate
+        print("lr_d", lr_d)
+        print("lr_g", lr_g)
+        opt_ae = torch.optim.Adam(list(self.encoder.parameters())+
+                                  list(self.decoder.parameters())+
+                                  list(self.quantize.parameters())+
+                                  list(self.quant_conv.parameters())+
+                                  list(self.post_quant_conv.parameters()),
+                                  lr=lr_g, betas=(0.5, 0.9))
+        opt_disc = torch.optim.Adam(self.loss.discriminator.parameters(),
+                                    lr=lr_d, betas=(0.5, 0.9))
+
+        if self.scheduler_config is not None:
+            scheduler = instantiate_from_config(self.scheduler_config)
+
+            print("Setting up LambdaLR scheduler...")
+            scheduler = [
+                {
+                    'scheduler': LambdaLR(opt_ae, lr_lambda=scheduler.schedule),
+                    'interval': 'step',
+                    'frequency': 1
+                },
+                {
+                    'scheduler': LambdaLR(opt_disc, lr_lambda=scheduler.schedule),
+                    'interval': 'step',
+                    'frequency': 1
+                },
+            ]
+            return [opt_ae, opt_disc], scheduler
+        return [opt_ae, opt_disc], []
+
+    def get_last_layer(self):
+        return self.decoder.conv_out.weight
+
+    def log_images(self, batch, only_inputs=False, plot_ema=False, **kwargs):
+        log = dict()
+        x = self.get_input(batch, self.image_key)
+        x = x.to(self.device)
+        if only_inputs:
+            log["inputs"] = x
+            return log
+        xrec, _ = self(x)
+        if x.shape[1] > 3:
+            # colorize with random projection
+            assert xrec.shape[1] > 3
+            x = self.to_rgb(x)
+            xrec = self.to_rgb(xrec)
+        log["inputs"] = x
+        log["reconstructions"] = xrec
+        if plot_ema:
+            with self.ema_scope():
+                xrec_ema, _ = self(x)
+                if x.shape[1] > 3: xrec_ema = self.to_rgb(xrec_ema)
+                log["reconstructions_ema"] = xrec_ema
+        return log
+
+    def to_rgb(self, x):
+        assert self.image_key == "segmentation"
+        if not hasattr(self, "colorize"):
+            self.register_buffer("colorize", torch.randn(3, x.shape[1], 1, 1).to(x))
+        x = F.conv2d(x, weight=self.colorize)
+        x = 2.*(x-x.min())/(x.max()-x.min()) - 1.
+        return x
+
+
+class VQModelInterface(VQModel):
+    def __init__(self, embed_dim, *args, **kwargs):
+        super().__init__(embed_dim=embed_dim, *args, **kwargs)
+        self.embed_dim = embed_dim
+
+    def encode(self, x):
+        h = self.encoder(x)
+        h = self.quant_conv(h)
+        return h
+
+    def decode(self, h, force_not_quantize=False):
+        # also go through quantization layer
+        if not force_not_quantize:
+            quant, emb_loss, info = self.quantize(h)
+        else:
+            quant = h
+        quant = self.post_quant_conv(quant)
+        dec = self.decoder(quant)
+        return dec
+
+
+class AutoencoderKL(pl.LightningModule):
+    def __init__(self,
+                 ddconfig,
+                 lossconfig,
+                 embed_dim,
+                 ckpt_path=None,
+                 ignore_keys=[],
+                 image_key="image",
+                 colorize_nlabels=None,
+                 monitor=None,
+                 ):
+        super().__init__()
+        self.image_key = image_key
+        self.encoder = Encoder(**ddconfig)
+        self.decoder = Decoder(**ddconfig)
+        self.loss = instantiate_from_config(lossconfig)
+        assert ddconfig["double_z"]
+        self.quant_conv = torch.nn.Conv2d(2*ddconfig["z_channels"], 2*embed_dim, 1)
+        self.post_quant_conv = torch.nn.Conv2d(embed_dim, ddconfig["z_channels"], 1)
+        self.embed_dim = embed_dim
+        if colorize_nlabels is not None:
+            assert type(colorize_nlabels)==int
+            self.register_buffer("colorize", torch.randn(3, colorize_nlabels, 1, 1))
+        if monitor is not None:
+            self.monitor = monitor
+        if ckpt_path is not None:
+            self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys)
+
+    def init_from_ckpt(self, path, ignore_keys=list()):
+        sd = torch.load(path, map_location="cpu")["state_dict"]
+        keys = list(sd.keys())
+        for k in keys:
+            for ik in ignore_keys:
+                if k.startswith(ik):
+                    print("Deleting key {} from state_dict.".format(k))
+                    del sd[k]
+        self.load_state_dict(sd, strict=False)
+        print(f"Restored from {path}")
+
+    def encode(self, x):
+        h = self.encoder(x)
+        moments = self.quant_conv(h)
+        posterior = DiagonalGaussianDistribution(moments)
+        return posterior
+
+    def decode(self, z):
+        z = self.post_quant_conv(z)
+        dec = self.decoder(z)
+        return dec
+
+    def forward(self, input, sample_posterior=True):
+        posterior = self.encode(input)
+        if sample_posterior:
+            z = posterior.sample()
+        else:
+            z = posterior.mode()
+        dec = self.decode(z)
+        return dec, posterior
+
+    def get_input(self, batch, k):
+        x = batch[k]
+        if len(x.shape) == 3:
+            x = x[..., None]
+        x = x.permute(0, 3, 1, 2).to(memory_format=torch.contiguous_format).float()
+        return x
+
+    def training_step(self, batch, batch_idx, optimizer_idx):
+        inputs = self.get_input(batch, self.image_key)
+        reconstructions, posterior = self(inputs)
+
+        if optimizer_idx == 0:
+            # train encoder+decoder+logvar
+            aeloss, log_dict_ae = self.loss(inputs, reconstructions, posterior, optimizer_idx, self.global_step,
+                                            last_layer=self.get_last_layer(), split="train")
+            self.log("aeloss", aeloss, prog_bar=True, logger=True, on_step=True, on_epoch=True)
+            self.log_dict(log_dict_ae, prog_bar=False, logger=True, on_step=True, on_epoch=False)
+            return aeloss
+
+        if optimizer_idx == 1:
+            # train the discriminator
+            discloss, log_dict_disc = self.loss(inputs, reconstructions, posterior, optimizer_idx, self.global_step,
+                                                last_layer=self.get_last_layer(), split="train")
+
+            self.log("discloss", discloss, prog_bar=True, logger=True, on_step=True, on_epoch=True)
+            self.log_dict(log_dict_disc, prog_bar=False, logger=True, on_step=True, on_epoch=False)
+            return discloss
+
+    def validation_step(self, batch, batch_idx):
+        inputs = self.get_input(batch, self.image_key)
+        reconstructions, posterior = self(inputs)
+        aeloss, log_dict_ae = self.loss(inputs, reconstructions, posterior, 0, self.global_step,
+                                        last_layer=self.get_last_layer(), split="val")
+
+        discloss, log_dict_disc = self.loss(inputs, reconstructions, posterior, 1, self.global_step,
+                                            last_layer=self.get_last_layer(), split="val")
+
+        self.log("val/rec_loss", log_dict_ae["val/rec_loss"])
+        self.log_dict(log_dict_ae)
+        self.log_dict(log_dict_disc)
+        return self.log_dict
+
+    def configure_optimizers(self):
+        lr = self.learning_rate
+        opt_ae = torch.optim.Adam(list(self.encoder.parameters())+
+                                  list(self.decoder.parameters())+
+                                  list(self.quant_conv.parameters())+
+                                  list(self.post_quant_conv.parameters()),
+                                  lr=lr, betas=(0.5, 0.9))
+        opt_disc = torch.optim.Adam(self.loss.discriminator.parameters(),
+                                    lr=lr, betas=(0.5, 0.9))
+        return [opt_ae, opt_disc], []
+
+    def get_last_layer(self):
+        return self.decoder.conv_out.weight
+
+    @torch.no_grad()
+    def log_images(self, batch, only_inputs=False, **kwargs):
+        log = dict()
+        x = self.get_input(batch, self.image_key)
+        x = x.to(self.device)
+        if not only_inputs:
+            xrec, posterior = self(x)
+            if x.shape[1] > 3:
+                # colorize with random projection
+                assert xrec.shape[1] > 3
+                x = self.to_rgb(x)
+                xrec = self.to_rgb(xrec)
+            log["samples"] = self.decode(torch.randn_like(posterior.sample()))
+            log["reconstructions"] = xrec
+        log["inputs"] = x
+        return log
+
+    def to_rgb(self, x):
+        assert self.image_key == "segmentation"
+        if not hasattr(self, "colorize"):
+            self.register_buffer("colorize", torch.randn(3, x.shape[1], 1, 1).to(x))
+        x = F.conv2d(x, weight=self.colorize)
+        x = 2.*(x-x.min())/(x.max()-x.min()) - 1.
+        return x
+
+
+class IdentityFirstStage(torch.nn.Module):
+    def __init__(self, *args, vq_interface=False, **kwargs):
+        self.vq_interface = vq_interface  # TODO: Should be true by default but check to not break older stuff
+        super().__init__()
+
+    def encode(self, x, *args, **kwargs):
+        return x
+
+    def decode(self, x, *args, **kwargs):
+        return x
+
+    def quantize(self, x, *args, **kwargs):
+        if self.vq_interface:
+            return x, None, [None, None, None]
+        return x
+
+    def forward(self, x, *args, **kwargs):
+        return x

ldm/models/diffusion/classifier.py

@@ -0,0 +1,267 @@
+import os
+import torch
+import pytorch_lightning as pl
+from omegaconf import OmegaConf
+from torch.nn import functional as F
+from torch.optim import AdamW
+from torch.optim.lr_scheduler import LambdaLR
+from copy import deepcopy
+from einops import rearrange
+from glob import glob
+from natsort import natsorted
+
+from ldm.modules.diffusionmodules.openaimodel import EncoderUNetModel, UNetModel
+from ldm.util import log_txt_as_img, default, ismap, instantiate_from_config
+
+__models__ = {
+    'class_label': EncoderUNetModel,
+    'segmentation': UNetModel
+}
+
+
+def disabled_train(self, mode=True):
+    """Overwrite model.train with this function to make sure train/eval mode
+    does not change anymore."""
+    return self
+
+
+class NoisyLatentImageClassifier(pl.LightningModule):
+
+    def __init__(self,
+                 diffusion_path,
+                 num_classes,
+                 ckpt_path=None,
+                 pool='attention',
+                 label_key=None,
+                 diffusion_ckpt_path=None,
+                 scheduler_config=None,
+                 weight_decay=1.e-2,
+                 log_steps=10,
+                 monitor='val/loss',
+                 *args,
+                 **kwargs):
+        super().__init__(*args, **kwargs)
+        self.num_classes = num_classes
+        # get latest config of diffusion model
+        diffusion_config = natsorted(glob(os.path.join(diffusion_path, 'configs', '*-project.yaml')))[-1]
+        self.diffusion_config = OmegaConf.load(diffusion_config).model
+        self.diffusion_config.params.ckpt_path = diffusion_ckpt_path
+        self.load_diffusion()
+
+        self.monitor = monitor
+        self.numd = self.diffusion_model.first_stage_model.encoder.num_resolutions - 1
+        self.log_time_interval = self.diffusion_model.num_timesteps // log_steps
+        self.log_steps = log_steps
+
+        self.label_key = label_key if not hasattr(self.diffusion_model, 'cond_stage_key') \
+            else self.diffusion_model.cond_stage_key
+
+        assert self.label_key is not None, 'label_key neither in diffusion model nor in model.params'
+
+        if self.label_key not in __models__:
+            raise NotImplementedError()
+
+        self.load_classifier(ckpt_path, pool)
+
+        self.scheduler_config = scheduler_config
+        self.use_scheduler = self.scheduler_config is not None
+        self.weight_decay = weight_decay
+
+    def init_from_ckpt(self, path, ignore_keys=list(), only_model=False):
+        sd = torch.load(path, map_location="cpu")
+        if "state_dict" in list(sd.keys()):
+            sd = sd["state_dict"]
+        keys = list(sd.keys())
+        for k in keys:
+            for ik in ignore_keys:
+                if k.startswith(ik):
+                    print("Deleting key {} from state_dict.".format(k))
+                    del sd[k]
+        missing, unexpected = self.load_state_dict(sd, strict=False) if not only_model else self.model.load_state_dict(
+            sd, strict=False)
+        print(f"Restored from {path} with {len(missing)} missing and {len(unexpected)} unexpected keys")
+        if len(missing) > 0:
+            print(f"Missing Keys: {missing}")
+        if len(unexpected) > 0:
+            print(f"Unexpected Keys: {unexpected}")
+
+    def load_diffusion(self):
+        model = instantiate_from_config(self.diffusion_config)
+        self.diffusion_model = model.eval()
+        self.diffusion_model.train = disabled_train
+        for param in self.diffusion_model.parameters():
+            param.requires_grad = False
+
+    def load_classifier(self, ckpt_path, pool):
+        model_config = deepcopy(self.diffusion_config.params.unet_config.params)
+        model_config.in_channels = self.diffusion_config.params.unet_config.params.out_channels
+        model_config.out_channels = self.num_classes
+        if self.label_key == 'class_label':
+            model_config.pool = pool
+
+        self.model = __models__[self.label_key](**model_config)
+        if ckpt_path is not None:
+            print('#####################################################################')
+            print(f'load from ckpt "{ckpt_path}"')
+            print('#####################################################################')
+            self.init_from_ckpt(ckpt_path)
+
+    @torch.no_grad()
+    def get_x_noisy(self, x, t, noise=None):
+        noise = default(noise, lambda: torch.randn_like(x))
+        continuous_sqrt_alpha_cumprod = None
+        if self.diffusion_model.use_continuous_noise:
+            continuous_sqrt_alpha_cumprod = self.diffusion_model.sample_continuous_noise_level(x.shape[0], t + 1)
+            # todo: make sure t+1 is correct here
+
+        return self.diffusion_model.q_sample(x_start=x, t=t, noise=noise,
+                                             continuous_sqrt_alpha_cumprod=continuous_sqrt_alpha_cumprod)
+
+    def forward(self, x_noisy, t, *args, **kwargs):
+        return self.model(x_noisy, t)
+
+    @torch.no_grad()
+    def get_input(self, batch, k):
+        x = batch[k]
+        if len(x.shape) == 3:
+            x = x[..., None]
+        x = rearrange(x, 'b h w c -> b c h w')
+        x = x.to(memory_format=torch.contiguous_format).float()
+        return x
+
+    @torch.no_grad()
+    def get_conditioning(self, batch, k=None):
+        if k is None:
+            k = self.label_key
+        assert k is not None, 'Needs to provide label key'
+
+        targets = batch[k].to(self.device)
+
+        if self.label_key == 'segmentation':
+            targets = rearrange(targets, 'b h w c -> b c h w')
+            for down in range(self.numd):
+                h, w = targets.shape[-2:]
+                targets = F.interpolate(targets, size=(h // 2, w // 2), mode='nearest')
+
+            # targets = rearrange(targets,'b c h w -> b h w c')
+
+        return targets
+
+    def compute_top_k(self, logits, labels, k, reduction="mean"):
+        _, top_ks = torch.topk(logits, k, dim=1)
+        if reduction == "mean":
+            return (top_ks == labels[:, None]).float().sum(dim=-1).mean().item()
+        elif reduction == "none":
+            return (top_ks == labels[:, None]).float().sum(dim=-1)
+
+    def on_train_epoch_start(self):
+        # save some memory
+        self.diffusion_model.model.to('cpu')
+
+    @torch.no_grad()
+    def write_logs(self, loss, logits, targets):
+        log_prefix = 'train' if self.training else 'val'
+        log = {}
+        log[f"{log_prefix}/loss"] = loss.mean()
+        log[f"{log_prefix}/acc@1"] = self.compute_top_k(
+            logits, targets, k=1, reduction="mean"
+        )
+        log[f"{log_prefix}/acc@5"] = self.compute_top_k(
+            logits, targets, k=5, reduction="mean"
+        )
+
+        self.log_dict(log, prog_bar=False, logger=True, on_step=self.training, on_epoch=True)
+        self.log('loss', log[f"{log_prefix}/loss"], prog_bar=True, logger=False)
+        self.log('global_step', self.global_step, logger=False, on_epoch=False, prog_bar=True)
+        lr = self.optimizers().param_groups[0]['lr']
+        self.log('lr_abs', lr, on_step=True, logger=True, on_epoch=False, prog_bar=True)
+
+    def shared_step(self, batch, t=None):
+        x, *_ = self.diffusion_model.get_input(batch, k=self.diffusion_model.first_stage_key)
+        targets = self.get_conditioning(batch)
+        if targets.dim() == 4:
+            targets = targets.argmax(dim=1)
+        if t is None:
+            t = torch.randint(0, self.diffusion_model.num_timesteps, (x.shape[0],), device=self.device).long()
+        else:
+            t = torch.full(size=(x.shape[0],), fill_value=t, device=self.device).long()
+        x_noisy = self.get_x_noisy(x, t)
+        logits = self(x_noisy, t)
+
+        loss = F.cross_entropy(logits, targets, reduction='none')
+
+        self.write_logs(loss.detach(), logits.detach(), targets.detach())
+
+        loss = loss.mean()
+        return loss, logits, x_noisy, targets
+
+    def training_step(self, batch, batch_idx):
+        loss, *_ = self.shared_step(batch)
+        return loss
+
+    def reset_noise_accs(self):
+        self.noisy_acc = {t: {'acc@1': [], 'acc@5': []} for t in
+                          range(0, self.diffusion_model.num_timesteps, self.diffusion_model.log_every_t)}
+
+    def on_validation_start(self):
+        self.reset_noise_accs()
+
+    @torch.no_grad()
+    def validation_step(self, batch, batch_idx):
+        loss, *_ = self.shared_step(batch)
+
+        for t in self.noisy_acc:
+            _, logits, _, targets = self.shared_step(batch, t)
+            self.noisy_acc[t]['acc@1'].append(self.compute_top_k(logits, targets, k=1, reduction='mean'))
+            self.noisy_acc[t]['acc@5'].append(self.compute_top_k(logits, targets, k=5, reduction='mean'))
+
+        return loss
+
+    def configure_optimizers(self):
+        optimizer = AdamW(self.model.parameters(), lr=self.learning_rate, weight_decay=self.weight_decay)
+
+        if self.use_scheduler:
+            scheduler = instantiate_from_config(self.scheduler_config)
+
+            print("Setting up LambdaLR scheduler...")
+            scheduler = [
+                {
+                    'scheduler': LambdaLR(optimizer, lr_lambda=scheduler.schedule),
+                    'interval': 'step',
+                    'frequency': 1
+                }]
+            return [optimizer], scheduler
+
+        return optimizer
+
+    @torch.no_grad()
+    def log_images(self, batch, N=8, *args, **kwargs):
+        log = dict()
+        x = self.get_input(batch, self.diffusion_model.first_stage_key)
+        log['inputs'] = x
+
+        y = self.get_conditioning(batch)
+
+        if self.label_key == 'class_label':
+            y = log_txt_as_img((x.shape[2], x.shape[3]), batch["human_label"])
+            log['labels'] = y
+
+        if ismap(y):
+            log['labels'] = self.diffusion_model.to_rgb(y)
+
+            for step in range(self.log_steps):
+                current_time = step * self.log_time_interval
+
+                _, logits, x_noisy, _ = self.shared_step(batch, t=current_time)
+
+                log[f'inputs@t{current_time}'] = x_noisy
+
+                pred = F.one_hot(logits.argmax(dim=1), num_classes=self.num_classes)
+                pred = rearrange(pred, 'b h w c -> b c h w')
+
+                log[f'pred@t{current_time}'] = self.diffusion_model.to_rgb(pred)
+
+        for key in log:
+            log[key] = log[key][:N]
+
+        return log

ldm/models/diffusion/ddim.py

@@ -0,0 +1,241 @@
+"""SAMPLING ONLY."""
+
+import torch
+import numpy as np
+from tqdm import tqdm
+from functools import partial
+
+from ldm.modules.diffusionmodules.util import make_ddim_sampling_parameters, make_ddim_timesteps, noise_like, \
+    extract_into_tensor
+
+
+class DDIMSampler(object):
+    def __init__(self, model, schedule="linear", **kwargs):
+        super().__init__()
+        self.model = model
+        self.ddpm_num_timesteps = model.num_timesteps
+        self.schedule = schedule
+
+    def register_buffer(self, name, attr):
+        if type(attr) == torch.Tensor:
+            if attr.device != torch.device("cuda"):
+                attr = attr.to(torch.device("cuda"))
+        setattr(self, name, attr)
+
+    def make_schedule(self, ddim_num_steps, ddim_discretize="uniform", ddim_eta=0., verbose=True):
+        self.ddim_timesteps = make_ddim_timesteps(ddim_discr_method=ddim_discretize, num_ddim_timesteps=ddim_num_steps,
+                                                  num_ddpm_timesteps=self.ddpm_num_timesteps,verbose=verbose)
+        alphas_cumprod = self.model.alphas_cumprod
+        assert alphas_cumprod.shape[0] == self.ddpm_num_timesteps, 'alphas have to be defined for each timestep'
+        to_torch = lambda x: x.clone().detach().to(torch.float32).to(self.model.device)
+
+        self.register_buffer('betas', to_torch(self.model.betas))
+        self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod))
+        self.register_buffer('alphas_cumprod_prev', to_torch(self.model.alphas_cumprod_prev))
+
+        # calculations for diffusion q(x_t | x_{t-1}) and others
+        self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod.cpu())))
+        self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod.cpu())))
+        self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod.cpu())))
+        self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu())))
+        self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu() - 1)))
+
+        # ddim sampling parameters
+        ddim_sigmas, ddim_alphas, ddim_alphas_prev = make_ddim_sampling_parameters(alphacums=alphas_cumprod.cpu(),
+                                                                                   ddim_timesteps=self.ddim_timesteps,
+                                                                                   eta=ddim_eta,verbose=verbose)
+        self.register_buffer('ddim_sigmas', ddim_sigmas)
+        self.register_buffer('ddim_alphas', ddim_alphas)
+        self.register_buffer('ddim_alphas_prev', ddim_alphas_prev)
+        self.register_buffer('ddim_sqrt_one_minus_alphas', np.sqrt(1. - ddim_alphas))
+        sigmas_for_original_sampling_steps = ddim_eta * torch.sqrt(
+            (1 - self.alphas_cumprod_prev) / (1 - self.alphas_cumprod) * (
+                        1 - self.alphas_cumprod / self.alphas_cumprod_prev))
+        self.register_buffer('ddim_sigmas_for_original_num_steps', sigmas_for_original_sampling_steps)
+
+    @torch.no_grad()
+    def sample(self,
+               S,
+               batch_size,
+               shape,
+               conditioning=None,
+               callback=None,
+               normals_sequence=None,
+               img_callback=None,
+               quantize_x0=False,
+               eta=0.,
+               mask=None,
+               x0=None,
+               temperature=1.,
+               noise_dropout=0.,
+               score_corrector=None,
+               corrector_kwargs=None,
+               verbose=True,
+               x_T=None,
+               log_every_t=100,
+               unconditional_guidance_scale=1.,
+               unconditional_conditioning=None,
+               # this has to come in the same format as the conditioning, # e.g. as encoded tokens, ...
+               **kwargs
+               ):
+        if conditioning is not None:
+            if isinstance(conditioning, dict):
+                cbs = conditioning[list(conditioning.keys())[0]].shape[0]
+                if cbs != batch_size:
+                    print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}")
+            else:
+                if conditioning.shape[0] != batch_size:
+                    print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}")
+
+        self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=verbose)
+        # sampling
+        C, H, W = shape
+        size = (batch_size, C, H, W)
+        print(f'Data shape for DDIM sampling is {size}, eta {eta}')
+
+        samples, intermediates = self.ddim_sampling(conditioning, size,
+                                                    callback=callback,
+                                                    img_callback=img_callback,
+                                                    quantize_denoised=quantize_x0,
+                                                    mask=mask, x0=x0,
+                                                    ddim_use_original_steps=False,
+                                                    noise_dropout=noise_dropout,
+                                                    temperature=temperature,
+                                                    score_corrector=score_corrector,
+                                                    corrector_kwargs=corrector_kwargs,
+                                                    x_T=x_T,
+                                                    log_every_t=log_every_t,
+                                                    unconditional_guidance_scale=unconditional_guidance_scale,
+                                                    unconditional_conditioning=unconditional_conditioning,
+                                                    )
+        return samples, intermediates
+
+    @torch.no_grad()
+    def ddim_sampling(self, cond, shape,
+                      x_T=None, ddim_use_original_steps=False,
+                      callback=None, timesteps=None, quantize_denoised=False,
+                      mask=None, x0=None, img_callback=None, log_every_t=100,
+                      temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
+                      unconditional_guidance_scale=1., unconditional_conditioning=None,):
+        device = self.model.betas.device
+        b = shape[0]
+        if x_T is None:
+            img = torch.randn(shape, device=device)
+        else:
+            img = x_T
+
+        if timesteps is None:
+            timesteps = self.ddpm_num_timesteps if ddim_use_original_steps else self.ddim_timesteps
+        elif timesteps is not None and not ddim_use_original_steps:
+            subset_end = int(min(timesteps / self.ddim_timesteps.shape[0], 1) * self.ddim_timesteps.shape[0]) - 1
+            timesteps = self.ddim_timesteps[:subset_end]
+
+        intermediates = {'x_inter': [img], 'pred_x0': [img]}
+        time_range = reversed(range(0,timesteps)) if ddim_use_original_steps else np.flip(timesteps)
+        total_steps = timesteps if ddim_use_original_steps else timesteps.shape[0]
+        print(f"Running DDIM Sampling with {total_steps} timesteps")
+
+        iterator = tqdm(time_range, desc='DDIM Sampler', total=total_steps)
+
+        for i, step in enumerate(iterator):
+            index = total_steps - i - 1
+            ts = torch.full((b,), step, device=device, dtype=torch.long)
+
+            if mask is not None:
+                assert x0 is not None
+                img_orig = self.model.q_sample(x0, ts)  # TODO: deterministic forward pass?
+                img = img_orig * mask + (1. - mask) * img
+
+            outs = self.p_sample_ddim(img, cond, ts, index=index, use_original_steps=ddim_use_original_steps,
+                                      quantize_denoised=quantize_denoised, temperature=temperature,
+                                      noise_dropout=noise_dropout, score_corrector=score_corrector,
+                                      corrector_kwargs=corrector_kwargs,
+                                      unconditional_guidance_scale=unconditional_guidance_scale,
+                                      unconditional_conditioning=unconditional_conditioning)
+            img, pred_x0 = outs
+            if callback: callback(i)
+            if img_callback: img_callback(pred_x0, i)
+
+            if index % log_every_t == 0 or index == total_steps - 1:
+                intermediates['x_inter'].append(img)
+                intermediates['pred_x0'].append(pred_x0)
+
+        return img, intermediates
+
+    @torch.no_grad()
+    def p_sample_ddim(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False,
+                      temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
+                      unconditional_guidance_scale=1., unconditional_conditioning=None):
+        b, *_, device = *x.shape, x.device
+
+        if unconditional_conditioning is None or unconditional_guidance_scale == 1.:
+            e_t = self.model.apply_model(x, t, c)
+        else:
+            x_in = torch.cat([x] * 2)
+            t_in = torch.cat([t] * 2)
+            c_in = torch.cat([unconditional_conditioning, c])
+            e_t_uncond, e_t = self.model.apply_model(x_in, t_in, c_in).chunk(2)
+            e_t = e_t_uncond + unconditional_guidance_scale * (e_t - e_t_uncond)
+
+        if score_corrector is not None:
+            assert self.model.parameterization == "eps"
+            e_t = score_corrector.modify_score(self.model, e_t, x, t, c, **corrector_kwargs)
+
+        alphas = self.model.alphas_cumprod if use_original_steps else self.ddim_alphas
+        alphas_prev = self.model.alphas_cumprod_prev if use_original_steps else self.ddim_alphas_prev
+        sqrt_one_minus_alphas = self.model.sqrt_one_minus_alphas_cumprod if use_original_steps else self.ddim_sqrt_one_minus_alphas
+        sigmas = self.model.ddim_sigmas_for_original_num_steps if use_original_steps else self.ddim_sigmas
+        # select parameters corresponding to the currently considered timestep
+        a_t = torch.full((b, 1, 1, 1), alphas[index], device=device)
+        a_prev = torch.full((b, 1, 1, 1), alphas_prev[index], device=device)
+        sigma_t = torch.full((b, 1, 1, 1), sigmas[index], device=device)
+        sqrt_one_minus_at = torch.full((b, 1, 1, 1), sqrt_one_minus_alphas[index],device=device)
+
+        # current prediction for x_0
+        pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt()
+        if quantize_denoised:
+            pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0)
+        # direction pointing to x_t
+        dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t
+        noise = sigma_t * noise_like(x.shape, device, repeat_noise) * temperature
+        if noise_dropout > 0.:
+            noise = torch.nn.functional.dropout(noise, p=noise_dropout)
+        x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise
+        return x_prev, pred_x0
+
+    @torch.no_grad()
+    def stochastic_encode(self, x0, t, use_original_steps=False, noise=None):
+        # fast, but does not allow for exact reconstruction
+        # t serves as an index to gather the correct alphas
+        if use_original_steps:
+            sqrt_alphas_cumprod = self.sqrt_alphas_cumprod
+            sqrt_one_minus_alphas_cumprod = self.sqrt_one_minus_alphas_cumprod
+        else:
+            sqrt_alphas_cumprod = torch.sqrt(self.ddim_alphas)
+            sqrt_one_minus_alphas_cumprod = self.ddim_sqrt_one_minus_alphas
+
+        if noise is None:
+            noise = torch.randn_like(x0)
+        return (extract_into_tensor(sqrt_alphas_cumprod, t, x0.shape) * x0 +
+                extract_into_tensor(sqrt_one_minus_alphas_cumprod, t, x0.shape) * noise)
+
+    @torch.no_grad()
+    def decode(self, x_latent, cond, t_start, unconditional_guidance_scale=1.0, unconditional_conditioning=None,
+               use_original_steps=False):
+
+        timesteps = np.arange(self.ddpm_num_timesteps) if use_original_steps else self.ddim_timesteps
+        timesteps = timesteps[:t_start]
+
+        time_range = np.flip(timesteps)
+        total_steps = timesteps.shape[0]
+        print(f"Running DDIM Sampling with {total_steps} timesteps")
+
+        iterator = tqdm(time_range, desc='Decoding image', total=total_steps)
+        x_dec = x_latent
+        for i, step in enumerate(iterator):
+            index = total_steps - i - 1
+            ts = torch.full((x_latent.shape[0],), step, device=x_latent.device, dtype=torch.long)
+            x_dec, _ = self.p_sample_ddim(x_dec, cond, ts, index=index, use_original_steps=use_original_steps,
+                                          unconditional_guidance_scale=unconditional_guidance_scale,
+                                          unconditional_conditioning=unconditional_conditioning)
+        return x_dec

ldm/models/diffusion/plms.py

@@ -0,0 +1,236 @@
+"""SAMPLING ONLY."""
+
+import torch
+import numpy as np
+from tqdm import tqdm
+from functools import partial
+
+from ldm.modules.diffusionmodules.util import make_ddim_sampling_parameters, make_ddim_timesteps, noise_like
+
+
+class PLMSSampler(object):
+    def __init__(self, model, schedule="linear", **kwargs):
+        super().__init__()
+        self.model = model
+        self.ddpm_num_timesteps = model.num_timesteps
+        self.schedule = schedule
+
+    def register_buffer(self, name, attr):
+        if type(attr) == torch.Tensor:
+            if attr.device != torch.device("cuda"):
+                attr = attr.to(torch.device("cuda"))
+        setattr(self, name, attr)
+
+    def make_schedule(self, ddim_num_steps, ddim_discretize="uniform", ddim_eta=0., verbose=True):
+        if ddim_eta != 0:
+            raise ValueError('ddim_eta must be 0 for PLMS')
+        self.ddim_timesteps = make_ddim_timesteps(ddim_discr_method=ddim_discretize, num_ddim_timesteps=ddim_num_steps,
+                                                  num_ddpm_timesteps=self.ddpm_num_timesteps,verbose=verbose)
+        alphas_cumprod = self.model.alphas_cumprod
+        assert alphas_cumprod.shape[0] == self.ddpm_num_timesteps, 'alphas have to be defined for each timestep'
+        to_torch = lambda x: x.clone().detach().to(torch.float32).to(self.model.device)
+
+        self.register_buffer('betas', to_torch(self.model.betas))
+        self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod))
+        self.register_buffer('alphas_cumprod_prev', to_torch(self.model.alphas_cumprod_prev))
+
+        # calculations for diffusion q(x_t | x_{t-1}) and others
+        self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod.cpu())))
+        self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod.cpu())))
+        self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod.cpu())))
+        self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu())))
+        self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu() - 1)))
+
+        # ddim sampling parameters
+        ddim_sigmas, ddim_alphas, ddim_alphas_prev = make_ddim_sampling_parameters(alphacums=alphas_cumprod.cpu(),
+                                                                                   ddim_timesteps=self.ddim_timesteps,
+                                                                                   eta=ddim_eta,verbose=verbose)
+        self.register_buffer('ddim_sigmas', ddim_sigmas)
+        self.register_buffer('ddim_alphas', ddim_alphas)
+        self.register_buffer('ddim_alphas_prev', ddim_alphas_prev)
+        self.register_buffer('ddim_sqrt_one_minus_alphas', np.sqrt(1. - ddim_alphas))
+        sigmas_for_original_sampling_steps = ddim_eta * torch.sqrt(
+            (1 - self.alphas_cumprod_prev) / (1 - self.alphas_cumprod) * (
+                        1 - self.alphas_cumprod / self.alphas_cumprod_prev))
+        self.register_buffer('ddim_sigmas_for_original_num_steps', sigmas_for_original_sampling_steps)
+
+    @torch.no_grad()
+    def sample(self,
+               S,
+               batch_size,
+               shape,
+               conditioning=None,
+               callback=None,
+               normals_sequence=None,
+               img_callback=None,
+               quantize_x0=False,
+               eta=0.,
+               mask=None,
+               x0=None,
+               temperature=1.,
+               noise_dropout=0.,
+               score_corrector=None,
+               corrector_kwargs=None,
+               verbose=True,
+               x_T=None,
+               log_every_t=100,
+               unconditional_guidance_scale=1.,
+               unconditional_conditioning=None,
+               # this has to come in the same format as the conditioning, # e.g. as encoded tokens, ...
+               **kwargs
+               ):
+        if conditioning is not None:
+            if isinstance(conditioning, dict):
+                cbs = conditioning[list(conditioning.keys())[0]].shape[0]
+                if cbs != batch_size:
+                    print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}")
+            else:
+                if conditioning.shape[0] != batch_size:
+                    print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}")
+
+        self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=verbose)
+        # sampling
+        C, H, W = shape
+        size = (batch_size, C, H, W)
+        print(f'Data shape for PLMS sampling is {size}')
+
+        samples, intermediates = self.plms_sampling(conditioning, size,
+                                                    callback=callback,
+                                                    img_callback=img_callback,
+                                                    quantize_denoised=quantize_x0,
+                                                    mask=mask, x0=x0,
+                                                    ddim_use_original_steps=False,
+                                                    noise_dropout=noise_dropout,
+                                                    temperature=temperature,
+                                                    score_corrector=score_corrector,
+                                                    corrector_kwargs=corrector_kwargs,
+                                                    x_T=x_T,
+                                                    log_every_t=log_every_t,
+                                                    unconditional_guidance_scale=unconditional_guidance_scale,
+                                                    unconditional_conditioning=unconditional_conditioning,
+                                                    )
+        return samples, intermediates
+
+    @torch.no_grad()
+    def plms_sampling(self, cond, shape,
+                      x_T=None, ddim_use_original_steps=False,
+                      callback=None, timesteps=None, quantize_denoised=False,
+                      mask=None, x0=None, img_callback=None, log_every_t=100,
+                      temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
+                      unconditional_guidance_scale=1., unconditional_conditioning=None,):
+        device = self.model.betas.device
+        b = shape[0]
+        if x_T is None:
+            img = torch.randn(shape, device=device)
+        else:
+            img = x_T
+
+        if timesteps is None:
+            timesteps = self.ddpm_num_timesteps if ddim_use_original_steps else self.ddim_timesteps
+        elif timesteps is not None and not ddim_use_original_steps:
+            subset_end = int(min(timesteps / self.ddim_timesteps.shape[0], 1) * self.ddim_timesteps.shape[0]) - 1
+            timesteps = self.ddim_timesteps[:subset_end]
+
+        intermediates = {'x_inter': [img], 'pred_x0': [img]}
+        time_range = list(reversed(range(0,timesteps))) if ddim_use_original_steps else np.flip(timesteps)
+        total_steps = timesteps if ddim_use_original_steps else timesteps.shape[0]
+        print(f"Running PLMS Sampling with {total_steps} timesteps")
+
+        iterator = tqdm(time_range, desc='PLMS Sampler', total=total_steps)
+        old_eps = []
+
+        for i, step in enumerate(iterator):
+            index = total_steps - i - 1
+            ts = torch.full((b,), step, device=device, dtype=torch.long)
+            ts_next = torch.full((b,), time_range[min(i + 1, len(time_range) - 1)], device=device, dtype=torch.long)
+
+            if mask is not None:
+                assert x0 is not None
+                img_orig = self.model.q_sample(x0, ts)  # TODO: deterministic forward pass?
+                img = img_orig * mask + (1. - mask) * img
+
+            outs = self.p_sample_plms(img, cond, ts, index=index, use_original_steps=ddim_use_original_steps,
+                                      quantize_denoised=quantize_denoised, temperature=temperature,
+                                      noise_dropout=noise_dropout, score_corrector=score_corrector,
+                                      corrector_kwargs=corrector_kwargs,
+                                      unconditional_guidance_scale=unconditional_guidance_scale,
+                                      unconditional_conditioning=unconditional_conditioning,
+                                      old_eps=old_eps, t_next=ts_next)
+            img, pred_x0, e_t = outs
+            old_eps.append(e_t)
+            if len(old_eps) >= 4:
+                old_eps.pop(0)
+            if callback: callback(i)
+            if img_callback: img_callback(pred_x0, i)
+
+            if index % log_every_t == 0 or index == total_steps - 1:
+                intermediates['x_inter'].append(img)
+                intermediates['pred_x0'].append(pred_x0)
+
+        return img, intermediates
+
+    @torch.no_grad()
+    def p_sample_plms(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False,
+                      temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
+                      unconditional_guidance_scale=1., unconditional_conditioning=None, old_eps=None, t_next=None):
+        b, *_, device = *x.shape, x.device
+
+        def get_model_output(x, t):
+            if unconditional_conditioning is None or unconditional_guidance_scale == 1.:
+                e_t = self.model.apply_model(x, t, c)
+            else:
+                x_in = torch.cat([x] * 2)
+                t_in = torch.cat([t] * 2)
+                c_in = torch.cat([unconditional_conditioning, c])
+                e_t_uncond, e_t = self.model.apply_model(x_in, t_in, c_in).chunk(2)
+                e_t = e_t_uncond + unconditional_guidance_scale * (e_t - e_t_uncond)
+
+            if score_corrector is not None:
+                assert self.model.parameterization == "eps"
+                e_t = score_corrector.modify_score(self.model, e_t, x, t, c, **corrector_kwargs)
+
+            return e_t
+
+        alphas = self.model.alphas_cumprod if use_original_steps else self.ddim_alphas
+        alphas_prev = self.model.alphas_cumprod_prev if use_original_steps else self.ddim_alphas_prev
+        sqrt_one_minus_alphas = self.model.sqrt_one_minus_alphas_cumprod if use_original_steps else self.ddim_sqrt_one_minus_alphas
+        sigmas = self.model.ddim_sigmas_for_original_num_steps if use_original_steps else self.ddim_sigmas
+
+        def get_x_prev_and_pred_x0(e_t, index):
+            # select parameters corresponding to the currently considered timestep
+            a_t = torch.full((b, 1, 1, 1), alphas[index], device=device)
+            a_prev = torch.full((b, 1, 1, 1), alphas_prev[index], device=device)
+            sigma_t = torch.full((b, 1, 1, 1), sigmas[index], device=device)
+            sqrt_one_minus_at = torch.full((b, 1, 1, 1), sqrt_one_minus_alphas[index],device=device)
+
+            # current prediction for x_0
+            pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt()
+            if quantize_denoised:
+                pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0)
+            # direction pointing to x_t
+            dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t
+            noise = sigma_t * noise_like(x.shape, device, repeat_noise) * temperature
+            if noise_dropout > 0.:
+                noise = torch.nn.functional.dropout(noise, p=noise_dropout)
+            x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise
+            return x_prev, pred_x0
+
+        e_t = get_model_output(x, t)
+        if len(old_eps) == 0:
+            # Pseudo Improved Euler (2nd order)
+            x_prev, pred_x0 = get_x_prev_and_pred_x0(e_t, index)
+            e_t_next = get_model_output(x_prev, t_next)
+            e_t_prime = (e_t + e_t_next) / 2
+        elif len(old_eps) == 1:
+            # 2nd order Pseudo Linear Multistep (Adams-Bashforth)
+            e_t_prime = (3 * e_t - old_eps[-1]) / 2
+        elif len(old_eps) == 2:
+            # 3nd order Pseudo Linear Multistep (Adams-Bashforth)
+            e_t_prime = (23 * e_t - 16 * old_eps[-1] + 5 * old_eps[-2]) / 12
+        elif len(old_eps) >= 3:
+            # 4nd order Pseudo Linear Multistep (Adams-Bashforth)
+            e_t_prime = (55 * e_t - 59 * old_eps[-1] + 37 * old_eps[-2] - 9 * old_eps[-3]) / 24
+
+        x_prev, pred_x0 = get_x_prev_and_pred_x0(e_t_prime, index)
+
+        return x_prev, pred_x0, e_t

ldm/modules/attention.py

@@ -0,0 +1,268 @@
+from inspect import isfunction
+import math
+import torch
+import torch.nn.functional as F
+from torch import nn, einsum
+from einops import rearrange, repeat
+
+from ldm.modules.diffusionmodules.util import checkpoint
+
+
+def exists(val):
+    return val is not None
+
+
+def uniq(arr):
+    return{el: True for el in arr}.keys()
+
+
+def default(val, d):
+    if exists(val):
+        return val
+    return d() if isfunction(d) else d
+
+
+def max_neg_value(t):
+    return -torch.finfo(t.dtype).max
+
+
+def init_(tensor):
+    dim = tensor.shape[-1]
+    std = 1 / math.sqrt(dim)
+    tensor.uniform_(-std, std)
+    return tensor
+
+
+# feedforward
+class GEGLU(nn.Module):
+    def __init__(self, dim_in, dim_out):
+        super().__init__()
+        self.proj = nn.Linear(dim_in, dim_out * 2)
+
+    def forward(self, x):
+        x, gate = self.proj(x).chunk(2, dim=-1)
+        return x * F.gelu(gate)
+
+
+class FeedForward(nn.Module):
+    def __init__(self, dim, dim_out=None, mult=4, glu=False, dropout=0.):
+        super().__init__()
+        inner_dim = int(dim * mult)
+        dim_out = default(dim_out, dim)
+        project_in = nn.Sequential(
+            nn.Linear(dim, inner_dim),
+            nn.GELU()
+        ) if not glu else GEGLU(dim, inner_dim)
+
+        self.net = nn.Sequential(
+            project_in,
+            nn.Dropout(dropout),
+            nn.Linear(inner_dim, dim_out)
+        )
+
+    def forward(self, x):
+        return self.net(x)
+
+
+def zero_module(module):
+    """
+    Zero out the parameters of a module and return it.
+    """
+    for p in module.parameters():
+        p.detach().zero_()
+    return module
+
+
+def Normalize(in_channels):
+    return torch.nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True)
+
+
+class LinearAttention(nn.Module):
+    def __init__(self, dim, heads=4, dim_head=32):
+        super().__init__()
+        self.heads = heads
+        hidden_dim = dim_head * heads
+        self.to_qkv = nn.Conv2d(dim, hidden_dim * 3, 1, bias = False)
+        self.to_out = nn.Conv2d(hidden_dim, dim, 1)
+
+    def forward(self, x):
+        b, c, h, w = x.shape
+        qkv = self.to_qkv(x)
+        q, k, v = rearrange(qkv, 'b (qkv heads c) h w -> qkv b heads c (h w)', heads = self.heads, qkv=3)
+        k = k.softmax(dim=-1)  
+        context = torch.einsum('bhdn,bhen->bhde', k, v)
+        out = torch.einsum('bhde,bhdn->bhen', context, q)
+        out = rearrange(out, 'b heads c (h w) -> b (heads c) h w', heads=self.heads, h=h, w=w)
+        return self.to_out(out)
+
+
+class SpatialSelfAttention(nn.Module):
+    def __init__(self, in_channels):
+        super().__init__()
+        self.in_channels = in_channels
+
+        self.norm = Normalize(in_channels)
+        self.q = torch.nn.Conv2d(in_channels,
+                                 in_channels,
+                                 kernel_size=1,
+                                 stride=1,
+                                 padding=0)
+        self.k = torch.nn.Conv2d(in_channels,
+                                 in_channels,
+                                 kernel_size=1,
+                                 stride=1,
+                                 padding=0)
+        self.v = torch.nn.Conv2d(in_channels,
+                                 in_channels,
+                                 kernel_size=1,
+                                 stride=1,
+                                 padding=0)
+        self.proj_out = torch.nn.Conv2d(in_channels,
+                                        in_channels,
+                                        kernel_size=1,
+                                        stride=1,
+                                        padding=0)
+
+    def forward(self, x):
+        h_ = x
+        h_ = self.norm(h_)
+        q = self.q(h_)
+        k = self.k(h_)
+        v = self.v(h_)
+
+        # compute attention
+        b,c,h,w = q.shape
+        q = rearrange(q, 'b c h w -> b (h w) c')
+        k = rearrange(k, 'b c h w -> b c (h w)')
+        w_ = torch.einsum('bij,bjk->bik', q, k)
+
+        w_ = w_ * (int(c)**(-0.5))
+        w_ = torch.nn.functional.softmax(w_, dim=2)
+
+        # attend to values
+        v = rearrange(v, 'b c h w -> b c (h w)')
+        w_ = rearrange(w_, 'b i j -> b j i')
+        h_ = torch.einsum('bij,bjk->bik', v, w_)
+        h_ = rearrange(h_, 'b c (h w) -> b c h w', h=h)
+        h_ = self.proj_out(h_)
+
+        return x+h_
+
+
+class CrossAttention(nn.Module):
+    def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0.):
+        super().__init__()
+        inner_dim = dim_head * heads
+        context_dim = default(context_dim, query_dim)
+
+        self.scale = dim_head ** -0.5
+        self.heads = heads
+
+        self.to_q = nn.Linear(query_dim, inner_dim, bias=False)
+        self.to_k = nn.Linear(context_dim, inner_dim, bias=False)
+        self.to_v = nn.Linear(context_dim, inner_dim, bias=False)
+
+        self.to_out = nn.Sequential(
+            nn.Linear(inner_dim, query_dim),
+            nn.Dropout(dropout)
+        )
+
+    def forward(self, x, context=None, mask=None):
+        h = self.heads
+
+        q = self.to_q(x)
+        context = default(context, x)
+        k = self.to_k(context)
+        v = self.to_v(context)
+        del context, x
+
+        q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v))
+
+        r1 = torch.zeros(q.shape[0], q.shape[1], v.shape[2], device=q.device)
+
+        # valid values for steps = 2,4,8,16,32,64
+        # higher steps is slower but less memory usage
+        # at 16 can run 1920x1536 on a 3090, at 64 can run over 1920x1920
+        # speed seems to be impacted more on 30x series cards
+        steps = 16
+        slice_size = q.shape[1] // steps if q.shape[1] % steps == 0 else q.shape[1]
+        for i in range(0, q.shape[1], slice_size):
+            end = i + slice_size
+            s1 = einsum('b i d, b j d -> b i j', q[:, i:end], k)
+            s1 *= self.scale
+            s2 = s1.softmax(dim=-1)
+            del s1
+            r1[:, i:end] = einsum('b i j, b j d -> b i d', s2, v)
+            del s2
+        r2 = rearrange(r1, '(b h) n d -> b n (h d)', h=h)
+        del r1
+
+        return self.to_out(r2)
+
+
+class BasicTransformerBlock(nn.Module):
+    def __init__(self, dim, n_heads, d_head, dropout=0., context_dim=None, gated_ff=True, checkpoint=True):
+        super().__init__()
+        self.attn1 = CrossAttention(query_dim=dim, heads=n_heads, dim_head=d_head, dropout=dropout)  # is a self-attention
+        self.ff = FeedForward(dim, dropout=dropout, glu=gated_ff)
+        self.attn2 = CrossAttention(query_dim=dim, context_dim=context_dim,
+                                    heads=n_heads, dim_head=d_head, dropout=dropout)  # is self-attn if context is none
+        self.norm1 = nn.LayerNorm(dim)
+        self.norm2 = nn.LayerNorm(dim)
+        self.norm3 = nn.LayerNorm(dim)
+        self.checkpoint = checkpoint
+
+    def forward(self, x, context=None):
+        return checkpoint(self._forward, (x, context), self.parameters(), self.checkpoint)
+
+    def _forward(self, x, context=None):
+        x = self.attn1(self.norm1(x)) + x
+        x = self.attn2(self.norm2(x), context=context) + x
+        x = self.ff(self.norm3(x)) + x
+        return x
+
+
+class SpatialTransformer(nn.Module):
+    """
+    Transformer block for image-like data.
+    First, project the input (aka embedding)
+    and reshape to b, t, d.
+    Then apply standard transformer action.
+    Finally, reshape to image
+    """
+    def __init__(self, in_channels, n_heads, d_head,
+                 depth=1, dropout=0., context_dim=None):
+        super().__init__()
+        self.in_channels = in_channels
+        inner_dim = n_heads * d_head
+        self.norm = Normalize(in_channels)
+
+        self.proj_in = nn.Conv2d(in_channels,
+                                 inner_dim,
+                                 kernel_size=1,
+                                 stride=1,
+                                 padding=0)
+
+        self.transformer_blocks = nn.ModuleList(
+            [BasicTransformerBlock(inner_dim, n_heads, d_head, dropout=dropout, context_dim=context_dim)
+                for d in range(depth)]
+        )
+
+        self.proj_out = zero_module(nn.Conv2d(inner_dim,
+                                              in_channels,
+                                              kernel_size=1,
+                                              stride=1,
+                                              padding=0))
+
+    def forward(self, x, context=None):
+        # note: if no context is given, cross-attention defaults to self-attention
+        b, c, h, w = x.shape
+        x_in = x
+        x = self.norm(x)
+        x = self.proj_in(x)
+        x = rearrange(x, 'b c h w -> b (h w) c')
+        for block in self.transformer_blocks:
+            x = block(x, context=context)
+        x = rearrange(x, 'b (h w) c -> b c h w', h=h, w=w)
+        x = self.proj_out(x)
+        return x + x_in

ldm/modules/diffusionmodules/model.py

@@ -0,0 +1,835 @@
+# pytorch_diffusion + derived encoder decoder
+import math
+import torch
+import torch.nn as nn
+import numpy as np
+from einops import rearrange
+
+from ldm.util import instantiate_from_config
+from ldm.modules.attention import LinearAttention
+
+
+def get_timestep_embedding(timesteps, embedding_dim):
+    """
+    This matches the implementation in Denoising Diffusion Probabilistic Models:
+    From Fairseq.
+    Build sinusoidal embeddings.
+    This matches the implementation in tensor2tensor, but differs slightly
+    from the description in Section 3.5 of "Attention Is All You Need".
+    """
+    assert len(timesteps.shape) == 1
+
+    half_dim = embedding_dim // 2
+    emb = math.log(10000) / (half_dim - 1)
+    emb = torch.exp(torch.arange(half_dim, dtype=torch.float32) * -emb)
+    emb = emb.to(device=timesteps.device)
+    emb = timesteps.float()[:, None] * emb[None, :]
+    emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1)
+    if embedding_dim % 2 == 1:  # zero pad
+        emb = torch.nn.functional.pad(emb, (0,1,0,0))
+    return emb
+
+
+def nonlinearity(x):
+    # swish
+    return x*torch.sigmoid(x)
+
+
+def Normalize(in_channels, num_groups=32):
+    return torch.nn.GroupNorm(num_groups=num_groups, num_channels=in_channels, eps=1e-6, affine=True)
+
+
+class Upsample(nn.Module):
+    def __init__(self, in_channels, with_conv):
+        super().__init__()
+        self.with_conv = with_conv
+        if self.with_conv:
+            self.conv = torch.nn.Conv2d(in_channels,
+                                        in_channels,
+                                        kernel_size=3,
+                                        stride=1,
+                                        padding=1)
+
+    def forward(self, x):
+        x = torch.nn.functional.interpolate(x, scale_factor=2.0, mode="nearest")
+        if self.with_conv:
+            x = self.conv(x)
+        return x
+
+
+class Downsample(nn.Module):
+    def __init__(self, in_channels, with_conv):
+        super().__init__()
+        self.with_conv = with_conv
+        if self.with_conv:
+            # no asymmetric padding in torch conv, must do it ourselves
+            self.conv = torch.nn.Conv2d(in_channels,
+                                        in_channels,
+                                        kernel_size=3,
+                                        stride=2,
+                                        padding=0)
+
+    def forward(self, x):
+        if self.with_conv:
+            pad = (0,1,0,1)
+            x = torch.nn.functional.pad(x, pad, mode="constant", value=0)
+            x = self.conv(x)
+        else:
+            x = torch.nn.functional.avg_pool2d(x, kernel_size=2, stride=2)
+        return x
+
+
+class ResnetBlock(nn.Module):
+    def __init__(self, *, in_channels, out_channels=None, conv_shortcut=False,
+                 dropout, temb_channels=512):
+        super().__init__()
+        self.in_channels = in_channels
+        out_channels = in_channels if out_channels is None else out_channels
+        self.out_channels = out_channels
+        self.use_conv_shortcut = conv_shortcut
+
+        self.norm1 = Normalize(in_channels)
+        self.conv1 = torch.nn.Conv2d(in_channels,
+                                     out_channels,
+                                     kernel_size=3,
+                                     stride=1,
+                                     padding=1)
+        if temb_channels > 0:
+            self.temb_proj = torch.nn.Linear(temb_channels,
+                                             out_channels)
+        self.norm2 = Normalize(out_channels)
+        self.dropout = torch.nn.Dropout(dropout)
+        self.conv2 = torch.nn.Conv2d(out_channels,
+                                     out_channels,
+                                     kernel_size=3,
+                                     stride=1,
+                                     padding=1)
+        if self.in_channels != self.out_channels:
+            if self.use_conv_shortcut:
+                self.conv_shortcut = torch.nn.Conv2d(in_channels,
+                                                     out_channels,
+                                                     kernel_size=3,
+                                                     stride=1,
+                                                     padding=1)
+            else:
+                self.nin_shortcut = torch.nn.Conv2d(in_channels,
+                                                    out_channels,
+                                                    kernel_size=1,
+                                                    stride=1,
+                                                    padding=0)
+
+    def forward(self, x, temb):
+        h = x
+        h = self.norm1(h)
+        h = nonlinearity(h)
+        h = self.conv1(h)
+
+        if temb is not None:
+            h = h + self.temb_proj(nonlinearity(temb))[:,:,None,None]
+
+        h = self.norm2(h)
+        h = nonlinearity(h)
+        h = self.dropout(h)
+        h = self.conv2(h)
+
+        if self.in_channels != self.out_channels:
+            if self.use_conv_shortcut:
+                x = self.conv_shortcut(x)
+            else:
+                x = self.nin_shortcut(x)
+
+        return x+h
+
+
+class LinAttnBlock(LinearAttention):
+    """to match AttnBlock usage"""
+    def __init__(self, in_channels):
+        super().__init__(dim=in_channels, heads=1, dim_head=in_channels)
+
+
+class AttnBlock(nn.Module):
+    def __init__(self, in_channels):
+        super().__init__()
+        self.in_channels = in_channels
+
+        self.norm = Normalize(in_channels)
+        self.q = torch.nn.Conv2d(in_channels,
+                                 in_channels,
+                                 kernel_size=1,
+                                 stride=1,
+                                 padding=0)
+        self.k = torch.nn.Conv2d(in_channels,
+                                 in_channels,
+                                 kernel_size=1,
+                                 stride=1,
+                                 padding=0)
+        self.v = torch.nn.Conv2d(in_channels,
+                                 in_channels,
+                                 kernel_size=1,
+                                 stride=1,
+                                 padding=0)
+        self.proj_out = torch.nn.Conv2d(in_channels,
+                                        in_channels,
+                                        kernel_size=1,
+                                        stride=1,
+                                        padding=0)
+
+
+    def forward(self, x):
+        h_ = x
+        h_ = self.norm(h_)
+        q = self.q(h_)
+        k = self.k(h_)
+        v = self.v(h_)
+
+        # compute attention
+        b,c,h,w = q.shape
+        q = q.reshape(b,c,h*w)
+        q = q.permute(0,2,1)   # b,hw,c
+        k = k.reshape(b,c,h*w) # b,c,hw
+        w_ = torch.bmm(q,k)     # b,hw,hw    w[b,i,j]=sum_c q[b,i,c]k[b,c,j]
+        w_ = w_ * (int(c)**(-0.5))
+        w_ = torch.nn.functional.softmax(w_, dim=2)
+
+        # attend to values
+        v = v.reshape(b,c,h*w)
+        w_ = w_.permute(0,2,1)   # b,hw,hw (first hw of k, second of q)
+        h_ = torch.bmm(v,w_)     # b, c,hw (hw of q) h_[b,c,j] = sum_i v[b,c,i] w_[b,i,j]
+        h_ = h_.reshape(b,c,h,w)
+
+        h_ = self.proj_out(h_)
+
+        return x+h_
+
+
+def make_attn(in_channels, attn_type="vanilla"):
+    assert attn_type in ["vanilla", "linear", "none"], f'attn_type {attn_type} unknown'
+    print(f"making attention of type '{attn_type}' with {in_channels} in_channels")
+    if attn_type == "vanilla":
+        return AttnBlock(in_channels)
+    elif attn_type == "none":
+        return nn.Identity(in_channels)
+    else:
+        return LinAttnBlock(in_channels)
+
+
+class Model(nn.Module):
+    def __init__(self, *, ch, out_ch, ch_mult=(1,2,4,8), num_res_blocks,
+                 attn_resolutions, dropout=0.0, resamp_with_conv=True, in_channels,
+                 resolution, use_timestep=True, use_linear_attn=False, attn_type="vanilla"):
+        super().__init__()
+        if use_linear_attn: attn_type = "linear"
+        self.ch = ch
+        self.temb_ch = self.ch*4
+        self.num_resolutions = len(ch_mult)
+        self.num_res_blocks = num_res_blocks
+        self.resolution = resolution
+        self.in_channels = in_channels
+
+        self.use_timestep = use_timestep
+        if self.use_timestep:
+            # timestep embedding
+            self.temb = nn.Module()
+            self.temb.dense = nn.ModuleList([
+                torch.nn.Linear(self.ch,
+                                self.temb_ch),
+                torch.nn.Linear(self.temb_ch,
+                                self.temb_ch),
+            ])
+
+        # downsampling
+        self.conv_in = torch.nn.Conv2d(in_channels,
+                                       self.ch,
+                                       kernel_size=3,
+                                       stride=1,
+                                       padding=1)
+
+        curr_res = resolution
+        in_ch_mult = (1,)+tuple(ch_mult)
+        self.down = nn.ModuleList()
+        for i_level in range(self.num_resolutions):
+            block = nn.ModuleList()
+            attn = nn.ModuleList()
+            block_in = ch*in_ch_mult[i_level]
+            block_out = ch*ch_mult[i_level]
+            for i_block in range(self.num_res_blocks):
+                block.append(ResnetBlock(in_channels=block_in,
+                                         out_channels=block_out,
+                                         temb_channels=self.temb_ch,
+                                         dropout=dropout))
+                block_in = block_out
+                if curr_res in attn_resolutions:
+                    attn.append(make_attn(block_in, attn_type=attn_type))
+            down = nn.Module()
+            down.block = block
+            down.attn = attn
+            if i_level != self.num_resolutions-1:
+                down.downsample = Downsample(block_in, resamp_with_conv)
+                curr_res = curr_res // 2
+            self.down.append(down)
+
+        # middle
+        self.mid = nn.Module()
+        self.mid.block_1 = ResnetBlock(in_channels=block_in,
+                                       out_channels=block_in,
+                                       temb_channels=self.temb_ch,
+                                       dropout=dropout)
+        self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
+        self.mid.block_2 = ResnetBlock(in_channels=block_in,
+                                       out_channels=block_in,
+                                       temb_channels=self.temb_ch,
+                                       dropout=dropout)
+
+        # upsampling
+        self.up = nn.ModuleList()
+        for i_level in reversed(range(self.num_resolutions)):
+            block = nn.ModuleList()
+            attn = nn.ModuleList()
+            block_out = ch*ch_mult[i_level]
+            skip_in = ch*ch_mult[i_level]
+            for i_block in range(self.num_res_blocks+1):
+                if i_block == self.num_res_blocks:
+                    skip_in = ch*in_ch_mult[i_level]
+                block.append(ResnetBlock(in_channels=block_in+skip_in,
+                                         out_channels=block_out,
+                                         temb_channels=self.temb_ch,
+                                         dropout=dropout))
+                block_in = block_out
+                if curr_res in attn_resolutions:
+                    attn.append(make_attn(block_in, attn_type=attn_type))
+            up = nn.Module()
+            up.block = block
+            up.attn = attn
+            if i_level != 0:
+                up.upsample = Upsample(block_in, resamp_with_conv)
+                curr_res = curr_res * 2
+            self.up.insert(0, up) # prepend to get consistent order
+
+        # end
+        self.norm_out = Normalize(block_in)
+        self.conv_out = torch.nn.Conv2d(block_in,
+                                        out_ch,
+                                        kernel_size=3,
+                                        stride=1,
+                                        padding=1)
+
+    def forward(self, x, t=None, context=None):
+        #assert x.shape[2] == x.shape[3] == self.resolution
+        if context is not None:
+            # assume aligned context, cat along channel axis
+            x = torch.cat((x, context), dim=1)
+        if self.use_timestep:
+            # timestep embedding
+            assert t is not None
+            temb = get_timestep_embedding(t, self.ch)
+            temb = self.temb.dense[0](temb)
+            temb = nonlinearity(temb)
+            temb = self.temb.dense[1](temb)
+        else:
+            temb = None
+
+        # downsampling
+        hs = [self.conv_in(x)]
+        for i_level in range(self.num_resolutions):
+            for i_block in range(self.num_res_blocks):
+                h = self.down[i_level].block[i_block](hs[-1], temb)
+                if len(self.down[i_level].attn) > 0:
+                    h = self.down[i_level].attn[i_block](h)
+                hs.append(h)
+            if i_level != self.num_resolutions-1:
+                hs.append(self.down[i_level].downsample(hs[-1]))
+
+        # middle
+        h = hs[-1]
+        h = self.mid.block_1(h, temb)
+        h = self.mid.attn_1(h)
+        h = self.mid.block_2(h, temb)
+
+        # upsampling
+        for i_level in reversed(range(self.num_resolutions)):
+            for i_block in range(self.num_res_blocks+1):
+                h = self.up[i_level].block[i_block](
+                    torch.cat([h, hs.pop()], dim=1), temb)
+                if len(self.up[i_level].attn) > 0:
+                    h = self.up[i_level].attn[i_block](h)
+            if i_level != 0:
+                h = self.up[i_level].upsample(h)
+
+        # end
+        h = self.norm_out(h)
+        h = nonlinearity(h)
+        h = self.conv_out(h)
+        return h
+
+    def get_last_layer(self):
+        return self.conv_out.weight
+
+
+class Encoder(nn.Module):
+    def __init__(self, *, ch, out_ch, ch_mult=(1,2,4,8), num_res_blocks,
+                 attn_resolutions, dropout=0.0, resamp_with_conv=True, in_channels,
+                 resolution, z_channels, double_z=True, use_linear_attn=False, attn_type="vanilla",
+                 **ignore_kwargs):
+        super().__init__()
+        if use_linear_attn: attn_type = "linear"
+        self.ch = ch
+        self.temb_ch = 0
+        self.num_resolutions = len(ch_mult)
+        self.num_res_blocks = num_res_blocks
+        self.resolution = resolution
+        self.in_channels = in_channels
+
+        # downsampling
+        self.conv_in = torch.nn.Conv2d(in_channels,
+                                       self.ch,
+                                       kernel_size=3,
+                                       stride=1,
+                                       padding=1)
+
+        curr_res = resolution
+        in_ch_mult = (1,)+tuple(ch_mult)
+        self.in_ch_mult = in_ch_mult
+        self.down = nn.ModuleList()
+        for i_level in range(self.num_resolutions):
+            block = nn.ModuleList()
+            attn = nn.ModuleList()
+            block_in = ch*in_ch_mult[i_level]
+            block_out = ch*ch_mult[i_level]
+            for i_block in range(self.num_res_blocks):
+                block.append(ResnetBlock(in_channels=block_in,
+                                         out_channels=block_out,
+                                         temb_channels=self.temb_ch,
+                                         dropout=dropout))
+                block_in = block_out
+                if curr_res in attn_resolutions:
+                    attn.append(make_attn(block_in, attn_type=attn_type))
+            down = nn.Module()
+            down.block = block
+            down.attn = attn
+            if i_level != self.num_resolutions-1:
+                down.downsample = Downsample(block_in, resamp_with_conv)
+                curr_res = curr_res // 2
+            self.down.append(down)
+
+        # middle
+        self.mid = nn.Module()
+        self.mid.block_1 = ResnetBlock(in_channels=block_in,
+                                       out_channels=block_in,
+                                       temb_channels=self.temb_ch,
+                                       dropout=dropout)
+        self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
+        self.mid.block_2 = ResnetBlock(in_channels=block_in,
+                                       out_channels=block_in,
+                                       temb_channels=self.temb_ch,
+                                       dropout=dropout)
+
+        # end
+        self.norm_out = Normalize(block_in)
+        self.conv_out = torch.nn.Conv2d(block_in,
+                                        2*z_channels if double_z else z_channels,
+                                        kernel_size=3,
+                                        stride=1,
+                                        padding=1)
+
+    def forward(self, x):
+        # timestep embedding
+        temb = None
+
+        # downsampling
+        hs = [self.conv_in(x)]
+        for i_level in range(self.num_resolutions):
+            for i_block in range(self.num_res_blocks):
+                h = self.down[i_level].block[i_block](hs[-1], temb)
+                if len(self.down[i_level].attn) > 0:
+                    h = self.down[i_level].attn[i_block](h)
+                hs.append(h)
+            if i_level != self.num_resolutions-1:
+                hs.append(self.down[i_level].downsample(hs[-1]))
+
+        # middle
+        h = hs[-1]
+        h = self.mid.block_1(h, temb)
+        h = self.mid.attn_1(h)
+        h = self.mid.block_2(h, temb)
+
+        # end
+        h = self.norm_out(h)
+        h = nonlinearity(h)
+        h = self.conv_out(h)
+        return h
+
+
+class Decoder(nn.Module):
+    def __init__(self, *, ch, out_ch, ch_mult=(1,2,4,8), num_res_blocks,
+                 attn_resolutions, dropout=0.0, resamp_with_conv=True, in_channels,
+                 resolution, z_channels, give_pre_end=False, tanh_out=False, use_linear_attn=False,
+                 attn_type="vanilla", **ignorekwargs):
+        super().__init__()
+        if use_linear_attn: attn_type = "linear"
+        self.ch = ch
+        self.temb_ch = 0
+        self.num_resolutions = len(ch_mult)
+        self.num_res_blocks = num_res_blocks
+        self.resolution = resolution
+        self.in_channels = in_channels
+        self.give_pre_end = give_pre_end
+        self.tanh_out = tanh_out
+
+        # compute in_ch_mult, block_in and curr_res at lowest res
+        in_ch_mult = (1,)+tuple(ch_mult)
+        block_in = ch*ch_mult[self.num_resolutions-1]
+        curr_res = resolution // 2**(self.num_resolutions-1)
+        self.z_shape = (1,z_channels,curr_res,curr_res)
+        print("Working with z of shape {} = {} dimensions.".format(
+            self.z_shape, np.prod(self.z_shape)))
+
+        # z to block_in
+        self.conv_in = torch.nn.Conv2d(z_channels,
+                                       block_in,
+                                       kernel_size=3,
+                                       stride=1,
+                                       padding=1)
+
+        # middle
+        self.mid = nn.Module()
+        self.mid.block_1 = ResnetBlock(in_channels=block_in,
+                                       out_channels=block_in,
+                                       temb_channels=self.temb_ch,
+                                       dropout=dropout)
+        self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
+        self.mid.block_2 = ResnetBlock(in_channels=block_in,
+                                       out_channels=block_in,
+                                       temb_channels=self.temb_ch,
+                                       dropout=dropout)
+
+        # upsampling
+        self.up = nn.ModuleList()
+        for i_level in reversed(range(self.num_resolutions)):
+            block = nn.ModuleList()
+            attn = nn.ModuleList()
+            block_out = ch*ch_mult[i_level]
+            for i_block in range(self.num_res_blocks+1):
+                block.append(ResnetBlock(in_channels=block_in,
+                                         out_channels=block_out,
+                                         temb_channels=self.temb_ch,
+                                         dropout=dropout))
+                block_in = block_out
+                if curr_res in attn_resolutions:
+                    attn.append(make_attn(block_in, attn_type=attn_type))
+            up = nn.Module()
+            up.block = block
+            up.attn = attn
+            if i_level != 0:
+                up.upsample = Upsample(block_in, resamp_with_conv)
+                curr_res = curr_res * 2
+            self.up.insert(0, up) # prepend to get consistent order
+
+        # end
+        self.norm_out = Normalize(block_in)
+        self.conv_out = torch.nn.Conv2d(block_in,
+                                        out_ch,
+                                        kernel_size=3,
+                                        stride=1,
+                                        padding=1)
+
+    def forward(self, z):
+        #assert z.shape[1:] == self.z_shape[1:]
+        self.last_z_shape = z.shape
+
+        # timestep embedding
+        temb = None
+
+        # z to block_in
+        h = self.conv_in(z)
+
+        # middle
+        h = self.mid.block_1(h, temb)
+        h = self.mid.attn_1(h)
+        h = self.mid.block_2(h, temb)
+
+        # upsampling
+        for i_level in reversed(range(self.num_resolutions)):
+            for i_block in range(self.num_res_blocks+1):
+                h = self.up[i_level].block[i_block](h, temb)
+                if len(self.up[i_level].attn) > 0:
+                    h = self.up[i_level].attn[i_block](h)
+            if i_level != 0:
+                h = self.up[i_level].upsample(h)
+
+        # end
+        if self.give_pre_end:
+            return h
+
+        h = self.norm_out(h)
+        h = nonlinearity(h)
+        h = self.conv_out(h)
+        if self.tanh_out:
+            h = torch.tanh(h)
+        return h
+
+
+class SimpleDecoder(nn.Module):
+    def __init__(self, in_channels, out_channels, *args, **kwargs):
+        super().__init__()
+        self.model = nn.ModuleList([nn.Conv2d(in_channels, in_channels, 1),
+                                     ResnetBlock(in_channels=in_channels,
+                                                 out_channels=2 * in_channels,
+                                                 temb_channels=0, dropout=0.0),
+                                     ResnetBlock(in_channels=2 * in_channels,
+                                                out_channels=4 * in_channels,
+                                                temb_channels=0, dropout=0.0),
+                                     ResnetBlock(in_channels=4 * in_channels,
+                                                out_channels=2 * in_channels,
+                                                temb_channels=0, dropout=0.0),
+                                     nn.Conv2d(2*in_channels, in_channels, 1),
+                                     Upsample(in_channels, with_conv=True)])
+        # end
+        self.norm_out = Normalize(in_channels)
+        self.conv_out = torch.nn.Conv2d(in_channels,
+                                        out_channels,
+                                        kernel_size=3,
+                                        stride=1,
+                                        padding=1)
+
+    def forward(self, x):
+        for i, layer in enumerate(self.model):
+            if i in [1,2,3]:
+                x = layer(x, None)
+            else:
+                x = layer(x)
+
+        h = self.norm_out(x)
+        h = nonlinearity(h)
+        x = self.conv_out(h)
+        return x
+
+
+class UpsampleDecoder(nn.Module):
+    def __init__(self, in_channels, out_channels, ch, num_res_blocks, resolution,
+                 ch_mult=(2,2), dropout=0.0):
+        super().__init__()
+        # upsampling
+        self.temb_ch = 0
+        self.num_resolutions = len(ch_mult)
+        self.num_res_blocks = num_res_blocks
+        block_in = in_channels
+        curr_res = resolution // 2 ** (self.num_resolutions - 1)
+        self.res_blocks = nn.ModuleList()
+        self.upsample_blocks = nn.ModuleList()
+        for i_level in range(self.num_resolutions):
+            res_block = []
+            block_out = ch * ch_mult[i_level]
+            for i_block in range(self.num_res_blocks + 1):
+                res_block.append(ResnetBlock(in_channels=block_in,
+                                         out_channels=block_out,
+                                         temb_channels=self.temb_ch,
+                                         dropout=dropout))
+                block_in = block_out
+            self.res_blocks.append(nn.ModuleList(res_block))
+            if i_level != self.num_resolutions - 1:
+                self.upsample_blocks.append(Upsample(block_in, True))
+                curr_res = curr_res * 2
+
+        # end
+        self.norm_out = Normalize(block_in)
+        self.conv_out = torch.nn.Conv2d(block_in,
+                                        out_channels,
+                                        kernel_size=3,
+                                        stride=1,
+                                        padding=1)
+
+    def forward(self, x):
+        # upsampling
+        h = x
+        for k, i_level in enumerate(range(self.num_resolutions)):
+            for i_block in range(self.num_res_blocks + 1):
+                h = self.res_blocks[i_level][i_block](h, None)
+            if i_level != self.num_resolutions - 1:
+                h = self.upsample_blocks[k](h)
+        h = self.norm_out(h)
+        h = nonlinearity(h)
+        h = self.conv_out(h)
+        return h
+
+
+class LatentRescaler(nn.Module):
+    def __init__(self, factor, in_channels, mid_channels, out_channels, depth=2):
+        super().__init__()
+        # residual block, interpolate, residual block
+        self.factor = factor
+        self.conv_in = nn.Conv2d(in_channels,
+                                 mid_channels,
+                                 kernel_size=3,
+                                 stride=1,
+                                 padding=1)
+        self.res_block1 = nn.ModuleList([ResnetBlock(in_channels=mid_channels,
+                                                     out_channels=mid_channels,
+                                                     temb_channels=0,
+                                                     dropout=0.0) for _ in range(depth)])
+        self.attn = AttnBlock(mid_channels)
+        self.res_block2 = nn.ModuleList([ResnetBlock(in_channels=mid_channels,
+                                                     out_channels=mid_channels,
+                                                     temb_channels=0,
+                                                     dropout=0.0) for _ in range(depth)])
+
+        self.conv_out = nn.Conv2d(mid_channels,
+                                  out_channels,
+                                  kernel_size=1,
+                                  )
+
+    def forward(self, x):
+        x = self.conv_in(x)
+        for block in self.res_block1:
+            x = block(x, None)
+        x = torch.nn.functional.interpolate(x, size=(int(round(x.shape[2]*self.factor)), int(round(x.shape[3]*self.factor))))
+        x = self.attn(x)
+        for block in self.res_block2:
+            x = block(x, None)
+        x = self.conv_out(x)
+        return x
+
+
+class MergedRescaleEncoder(nn.Module):
+    def __init__(self, in_channels, ch, resolution, out_ch, num_res_blocks,
+                 attn_resolutions, dropout=0.0, resamp_with_conv=True,
+                 ch_mult=(1,2,4,8), rescale_factor=1.0, rescale_module_depth=1):
+        super().__init__()
+        intermediate_chn = ch * ch_mult[-1]
+        self.encoder = Encoder(in_channels=in_channels, num_res_blocks=num_res_blocks, ch=ch, ch_mult=ch_mult,
+                               z_channels=intermediate_chn, double_z=False, resolution=resolution,
+                               attn_resolutions=attn_resolutions, dropout=dropout, resamp_with_conv=resamp_with_conv,
+                               out_ch=None)
+        self.rescaler = LatentRescaler(factor=rescale_factor, in_channels=intermediate_chn,
+                                       mid_channels=intermediate_chn, out_channels=out_ch, depth=rescale_module_depth)
+
+    def forward(self, x):
+        x = self.encoder(x)
+        x = self.rescaler(x)
+        return x
+
+
+class MergedRescaleDecoder(nn.Module):
+    def __init__(self, z_channels, out_ch, resolution, num_res_blocks, attn_resolutions, ch, ch_mult=(1,2,4,8),
+                 dropout=0.0, resamp_with_conv=True, rescale_factor=1.0, rescale_module_depth=1):
+        super().__init__()
+        tmp_chn = z_channels*ch_mult[-1]
+        self.decoder = Decoder(out_ch=out_ch, z_channels=tmp_chn, attn_resolutions=attn_resolutions, dropout=dropout,
+                               resamp_with_conv=resamp_with_conv, in_channels=None, num_res_blocks=num_res_blocks,
+                               ch_mult=ch_mult, resolution=resolution, ch=ch)
+        self.rescaler = LatentRescaler(factor=rescale_factor, in_channels=z_channels, mid_channels=tmp_chn,
+                                       out_channels=tmp_chn, depth=rescale_module_depth)
+
+    def forward(self, x):
+        x = self.rescaler(x)
+        x = self.decoder(x)
+        return x
+
+
+class Upsampler(nn.Module):
+    def __init__(self, in_size, out_size, in_channels, out_channels, ch_mult=2):
+        super().__init__()
+        assert out_size >= in_size
+        num_blocks = int(np.log2(out_size//in_size))+1
+        factor_up = 1.+ (out_size % in_size)
+        print(f"Building {self.__class__.__name__} with in_size: {in_size} --> out_size {out_size} and factor {factor_up}")
+        self.rescaler = LatentRescaler(factor=factor_up, in_channels=in_channels, mid_channels=2*in_channels,
+                                       out_channels=in_channels)
+        self.decoder = Decoder(out_ch=out_channels, resolution=out_size, z_channels=in_channels, num_res_blocks=2,
+                               attn_resolutions=[], in_channels=None, ch=in_channels,
+                               ch_mult=[ch_mult for _ in range(num_blocks)])
+
+    def forward(self, x):
+        x = self.rescaler(x)
+        x = self.decoder(x)
+        return x
+
+
+class Resize(nn.Module):
+    def __init__(self, in_channels=None, learned=False, mode="bilinear"):
+        super().__init__()
+        self.with_conv = learned
+        self.mode = mode
+        if self.with_conv:
+            print(f"Note: {self.__class__.__name} uses learned downsampling and will ignore the fixed {mode} mode")
+            raise NotImplementedError()
+            assert in_channels is not None
+            # no asymmetric padding in torch conv, must do it ourselves
+            self.conv = torch.nn.Conv2d(in_channels,
+                                        in_channels,
+                                        kernel_size=4,
+                                        stride=2,
+                                        padding=1)
+
+    def forward(self, x, scale_factor=1.0):
+        if scale_factor==1.0:
+            return x
+        else:
+            x = torch.nn.functional.interpolate(x, mode=self.mode, align_corners=False, scale_factor=scale_factor)
+        return x
+
+class FirstStagePostProcessor(nn.Module):
+
+    def __init__(self, ch_mult:list, in_channels,
+                 pretrained_model:nn.Module=None,
+                 reshape=False,
+                 n_channels=None,
+                 dropout=0.,
+                 pretrained_config=None):
+        super().__init__()
+        if pretrained_config is None:
+            assert pretrained_model is not None, 'Either "pretrained_model" or "pretrained_config" must not be None'
+            self.pretrained_model = pretrained_model
+        else:
+            assert pretrained_config is not None, 'Either "pretrained_model" or "pretrained_config" must not be None'
+            self.instantiate_pretrained(pretrained_config)
+
+        self.do_reshape = reshape
+
+        if n_channels is None:
+            n_channels = self.pretrained_model.encoder.ch
+
+        self.proj_norm = Normalize(in_channels,num_groups=in_channels//2)
+        self.proj = nn.Conv2d(in_channels,n_channels,kernel_size=3,
+                            stride=1,padding=1)
+
+        blocks = []
+        downs = []
+        ch_in = n_channels
+        for m in ch_mult:
+            blocks.append(ResnetBlock(in_channels=ch_in,out_channels=m*n_channels,dropout=dropout))
+            ch_in = m * n_channels
+            downs.append(Downsample(ch_in, with_conv=False))
+
+        self.model = nn.ModuleList(blocks)
+        self.downsampler = nn.ModuleList(downs)
+
+
+    def instantiate_pretrained(self, config):
+        model = instantiate_from_config(config)
+        self.pretrained_model = model.eval()
+        # self.pretrained_model.train = False
+        for param in self.pretrained_model.parameters():
+            param.requires_grad = False
+
+
+    @torch.no_grad()
+    def encode_with_pretrained(self,x):
+        c = self.pretrained_model.encode(x)
+        if isinstance(c, DiagonalGaussianDistribution):
+            c = c.mode()
+        return  c
+
+    def forward(self,x):
+        z_fs = self.encode_with_pretrained(x)
+        z = self.proj_norm(z_fs)
+        z = self.proj(z)
+        z = nonlinearity(z)
+
+        for submodel, downmodel in zip(self.model,self.downsampler):
+            z = submodel(z,temb=None)
+            z = downmodel(z)
+
+        if self.do_reshape:
+            z = rearrange(z,'b c h w -> b (h w) c')
+        return z
+

ldm/modules/diffusionmodules/openaimodel.py

@@ -0,0 +1,961 @@
+from abc import abstractmethod
+from functools import partial
+import math
+from typing import Iterable
+
+import numpy as np
+import torch as th
+import torch.nn as nn
+import torch.nn.functional as F
+
+from ldm.modules.diffusionmodules.util import (
+    checkpoint,
+    conv_nd,
+    linear,
+    avg_pool_nd,
+    zero_module,
+    normalization,
+    timestep_embedding,
+)
+from ldm.modules.attention import SpatialTransformer
+
+
+# dummy replace
+def convert_module_to_f16(x):
+    pass
+
+def convert_module_to_f32(x):
+    pass
+
+
+## go
+class AttentionPool2d(nn.Module):
+    """
+    Adapted from CLIP: https://github.com/openai/CLIP/blob/main/clip/model.py
+    """
+
+    def __init__(
+        self,
+        spacial_dim: int,
+        embed_dim: int,
+        num_heads_channels: int,
+        output_dim: int = None,
+    ):
+        super().__init__()
+        self.positional_embedding = nn.Parameter(th.randn(embed_dim, spacial_dim ** 2 + 1) / embed_dim ** 0.5)
+        self.qkv_proj = conv_nd(1, embed_dim, 3 * embed_dim, 1)
+        self.c_proj = conv_nd(1, embed_dim, output_dim or embed_dim, 1)
+        self.num_heads = embed_dim // num_heads_channels
+        self.attention = QKVAttention(self.num_heads)
+
+    def forward(self, x):
+        b, c, *_spatial = x.shape
+        x = x.reshape(b, c, -1)  # NC(HW)
+        x = th.cat([x.mean(dim=-1, keepdim=True), x], dim=-1)  # NC(HW+1)
+        x = x + self.positional_embedding[None, :, :].to(x.dtype)  # NC(HW+1)
+        x = self.qkv_proj(x)
+        x = self.attention(x)
+        x = self.c_proj(x)
+        return x[:, :, 0]
+
+
+class TimestepBlock(nn.Module):
+    """
+    Any module where forward() takes timestep embeddings as a second argument.
+    """
+
+    @abstractmethod
+    def forward(self, x, emb):
+        """
+        Apply the module to `x` given `emb` timestep embeddings.
+        """
+
+
+class TimestepEmbedSequential(nn.Sequential, TimestepBlock):
+    """
+    A sequential module that passes timestep embeddings to the children that
+    support it as an extra input.
+    """
+
+    def forward(self, x, emb, context=None):
+        for layer in self:
+            if isinstance(layer, TimestepBlock):
+                x = layer(x, emb)
+            elif isinstance(layer, SpatialTransformer):
+                x = layer(x, context)
+            else:
+                x = layer(x)
+        return x
+
+
+class Upsample(nn.Module):
+    """
+    An upsampling layer with an optional convolution.
+    :param channels: channels in the inputs and outputs.
+    :param use_conv: a bool determining if a convolution is applied.
+    :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
+                 upsampling occurs in the inner-two dimensions.
+    """
+
+    def __init__(self, channels, use_conv, dims=2, out_channels=None, padding=1):
+        super().__init__()
+        self.channels = channels
+        self.out_channels = out_channels or channels
+        self.use_conv = use_conv
+        self.dims = dims
+        if use_conv:
+            self.conv = conv_nd(dims, self.channels, self.out_channels, 3, padding=padding)
+
+    def forward(self, x):
+        assert x.shape[1] == self.channels
+        if self.dims == 3:
+            x = F.interpolate(
+                x, (x.shape[2], x.shape[3] * 2, x.shape[4] * 2), mode="nearest"
+            )
+        else:
+            x = F.interpolate(x, scale_factor=2, mode="nearest")
+        if self.use_conv:
+            x = self.conv(x)
+        return x
+
+class TransposedUpsample(nn.Module):
+    'Learned 2x upsampling without padding'
+    def __init__(self, channels, out_channels=None, ks=5):
+        super().__init__()
+        self.channels = channels
+        self.out_channels = out_channels or channels
+
+        self.up = nn.ConvTranspose2d(self.channels,self.out_channels,kernel_size=ks,stride=2)
+
+    def forward(self,x):
+        return self.up(x)
+
+
+class Downsample(nn.Module):
+    """
+    A downsampling layer with an optional convolution.
+    :param channels: channels in the inputs and outputs.
+    :param use_conv: a bool determining if a convolution is applied.
+    :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
+                 downsampling occurs in the inner-two dimensions.
+    """
+
+    def __init__(self, channels, use_conv, dims=2, out_channels=None,padding=1):
+        super().__init__()
+        self.channels = channels
+        self.out_channels = out_channels or channels
+        self.use_conv = use_conv
+        self.dims = dims
+        stride = 2 if dims != 3 else (1, 2, 2)
+        if use_conv:
+            self.op = conv_nd(
+                dims, self.channels, self.out_channels, 3, stride=stride, padding=padding
+            )
+        else:
+            assert self.channels == self.out_channels
+            self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride)
+
+    def forward(self, x):
+        assert x.shape[1] == self.channels
+        return self.op(x)
+
+
+class ResBlock(TimestepBlock):
+    """
+    A residual block that can optionally change the number of channels.
+    :param channels: the number of input channels.
+    :param emb_channels: the number of timestep embedding channels.
+    :param dropout: the rate of dropout.
+    :param out_channels: if specified, the number of out channels.
+    :param use_conv: if True and out_channels is specified, use a spatial
+        convolution instead of a smaller 1x1 convolution to change the
+        channels in the skip connection.
+    :param dims: determines if the signal is 1D, 2D, or 3D.
+    :param use_checkpoint: if True, use gradient checkpointing on this module.
+    :param up: if True, use this block for upsampling.
+    :param down: if True, use this block for downsampling.
+    """
+
+    def __init__(
+        self,
+        channels,
+        emb_channels,
+        dropout,
+        out_channels=None,
+        use_conv=False,
+        use_scale_shift_norm=False,
+        dims=2,
+        use_checkpoint=False,
+        up=False,
+        down=False,
+    ):
+        super().__init__()
+        self.channels = channels
+        self.emb_channels = emb_channels
+        self.dropout = dropout
+        self.out_channels = out_channels or channels
+        self.use_conv = use_conv
+        self.use_checkpoint = use_checkpoint
+        self.use_scale_shift_norm = use_scale_shift_norm
+
+        self.in_layers = nn.Sequential(
+            normalization(channels),
+            nn.SiLU(),
+            conv_nd(dims, channels, self.out_channels, 3, padding=1),
+        )
+
+        self.updown = up or down
+
+        if up:
+            self.h_upd = Upsample(channels, False, dims)
+            self.x_upd = Upsample(channels, False, dims)
+        elif down:
+            self.h_upd = Downsample(channels, False, dims)
+            self.x_upd = Downsample(channels, False, dims)
+        else:
+            self.h_upd = self.x_upd = nn.Identity()
+
+        self.emb_layers = nn.Sequential(
+            nn.SiLU(),
+            linear(
+                emb_channels,
+                2 * self.out_channels if use_scale_shift_norm else self.out_channels,
+            ),
+        )
+        self.out_layers = nn.Sequential(
+            normalization(self.out_channels),
+            nn.SiLU(),
+            nn.Dropout(p=dropout),
+            zero_module(
+                conv_nd(dims, self.out_channels, self.out_channels, 3, padding=1)
+            ),
+        )
+
+        if self.out_channels == channels:
+            self.skip_connection = nn.Identity()
+        elif use_conv:
+            self.skip_connection = conv_nd(
+                dims, channels, self.out_channels, 3, padding=1
+            )
+        else:
+            self.skip_connection = conv_nd(dims, channels, self.out_channels, 1)
+
+    def forward(self, x, emb):
+        """
+        Apply the block to a Tensor, conditioned on a timestep embedding.
+        :param x: an [N x C x ...] Tensor of features.
+        :param emb: an [N x emb_channels] Tensor of timestep embeddings.
+        :return: an [N x C x ...] Tensor of outputs.
+        """
+        return checkpoint(
+            self._forward, (x, emb), self.parameters(), self.use_checkpoint
+        )
+
+
+    def _forward(self, x, emb):
+        if self.updown:
+            in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1]
+            h = in_rest(x)
+            h = self.h_upd(h)
+            x = self.x_upd(x)
+            h = in_conv(h)
+        else:
+            h = self.in_layers(x)
+        emb_out = self.emb_layers(emb).type(h.dtype)
+        while len(emb_out.shape) < len(h.shape):
+            emb_out = emb_out[..., None]
+        if self.use_scale_shift_norm:
+            out_norm, out_rest = self.out_layers[0], self.out_layers[1:]
+            scale, shift = th.chunk(emb_out, 2, dim=1)
+            h = out_norm(h) * (1 + scale) + shift
+            h = out_rest(h)
+        else:
+            h = h + emb_out
+            h = self.out_layers(h)
+        return self.skip_connection(x) + h
+
+
+class AttentionBlock(nn.Module):
+    """
+    An attention block that allows spatial positions to attend to each other.
+    Originally ported from here, but adapted to the N-d case.
+    https://github.com/hojonathanho/diffusion/blob/1e0dceb3b3495bbe19116a5e1b3596cd0706c543/diffusion_tf/models/unet.py#L66.
+    """
+
+    def __init__(
+        self,
+        channels,
+        num_heads=1,
+        num_head_channels=-1,
+        use_checkpoint=False,
+        use_new_attention_order=False,
+    ):
+        super().__init__()
+        self.channels = channels
+        if num_head_channels == -1:
+            self.num_heads = num_heads
+        else:
+            assert (
+                channels % num_head_channels == 0
+            ), f"q,k,v channels {channels} is not divisible by num_head_channels {num_head_channels}"
+            self.num_heads = channels // num_head_channels
+        self.use_checkpoint = use_checkpoint
+        self.norm = normalization(channels)
+        self.qkv = conv_nd(1, channels, channels * 3, 1)
+        if use_new_attention_order:
+            # split qkv before split heads
+            self.attention = QKVAttention(self.num_heads)
+        else:
+            # split heads before split qkv
+            self.attention = QKVAttentionLegacy(self.num_heads)
+
+        self.proj_out = zero_module(conv_nd(1, channels, channels, 1))
+
+    def forward(self, x):
+        return checkpoint(self._forward, (x,), self.parameters(), True)   # TODO: check checkpoint usage, is True # TODO: fix the .half call!!!
+        #return pt_checkpoint(self._forward, x)  # pytorch
+
+    def _forward(self, x):
+        b, c, *spatial = x.shape
+        x = x.reshape(b, c, -1)
+        qkv = self.qkv(self.norm(x))
+        h = self.attention(qkv)
+        h = self.proj_out(h)
+        return (x + h).reshape(b, c, *spatial)
+
+
+def count_flops_attn(model, _x, y):
+    """
+    A counter for the `thop` package to count the operations in an
+    attention operation.
+    Meant to be used like:
+        macs, params = thop.profile(
+            model,
+            inputs=(inputs, timestamps),
+            custom_ops={QKVAttention: QKVAttention.count_flops},
+        )
+    """
+    b, c, *spatial = y[0].shape
+    num_spatial = int(np.prod(spatial))
+    # We perform two matmuls with the same number of ops.
+    # The first computes the weight matrix, the second computes
+    # the combination of the value vectors.
+    matmul_ops = 2 * b * (num_spatial ** 2) * c
+    model.total_ops += th.DoubleTensor([matmul_ops])
+
+
+class QKVAttentionLegacy(nn.Module):
+    """
+    A module which performs QKV attention. Matches legacy QKVAttention + input/ouput heads shaping
+    """
+
+    def __init__(self, n_heads):
+        super().__init__()
+        self.n_heads = n_heads
+
+    def forward(self, qkv):
+        """
+        Apply QKV attention.
+        :param qkv: an [N x (H * 3 * C) x T] tensor of Qs, Ks, and Vs.
+        :return: an [N x (H * C) x T] tensor after attention.
+        """
+        bs, width, length = qkv.shape
+        assert width % (3 * self.n_heads) == 0
+        ch = width // (3 * self.n_heads)
+        q, k, v = qkv.reshape(bs * self.n_heads, ch * 3, length).split(ch, dim=1)
+        scale = 1 / math.sqrt(math.sqrt(ch))
+        weight = th.einsum(
+            "bct,bcs->bts", q * scale, k * scale
+        )  # More stable with f16 than dividing afterwards
+        weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
+        a = th.einsum("bts,bcs->bct", weight, v)
+        return a.reshape(bs, -1, length)
+
+    @staticmethod
+    def count_flops(model, _x, y):
+        return count_flops_attn(model, _x, y)
+
+
+class QKVAttention(nn.Module):
+    """
+    A module which performs QKV attention and splits in a different order.
+    """
+
+    def __init__(self, n_heads):
+        super().__init__()
+        self.n_heads = n_heads
+
+    def forward(self, qkv):
+        """
+        Apply QKV attention.
+        :param qkv: an [N x (3 * H * C) x T] tensor of Qs, Ks, and Vs.
+        :return: an [N x (H * C) x T] tensor after attention.
+        """
+        bs, width, length = qkv.shape
+        assert width % (3 * self.n_heads) == 0
+        ch = width // (3 * self.n_heads)
+        q, k, v = qkv.chunk(3, dim=1)
+        scale = 1 / math.sqrt(math.sqrt(ch))
+        weight = th.einsum(
+            "bct,bcs->bts",
+            (q * scale).view(bs * self.n_heads, ch, length),
+            (k * scale).view(bs * self.n_heads, ch, length),
+        )  # More stable with f16 than dividing afterwards
+        weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
+        a = th.einsum("bts,bcs->bct", weight, v.reshape(bs * self.n_heads, ch, length))
+        return a.reshape(bs, -1, length)
+
+    @staticmethod
+    def count_flops(model, _x, y):
+        return count_flops_attn(model, _x, y)
+
+
+class UNetModel(nn.Module):
+    """
+    The full UNet model with attention and timestep embedding.
+    :param in_channels: channels in the input Tensor.
+    :param model_channels: base channel count for the model.
+    :param out_channels: channels in the output Tensor.
+    :param num_res_blocks: number of residual blocks per downsample.
+    :param attention_resolutions: a collection of downsample rates at which
+        attention will take place. May be a set, list, or tuple.
+        For example, if this contains 4, then at 4x downsampling, attention
+        will be used.
+    :param dropout: the dropout probability.
+    :param channel_mult: channel multiplier for each level of the UNet.
+    :param conv_resample: if True, use learned convolutions for upsampling and
+        downsampling.
+    :param dims: determines if the signal is 1D, 2D, or 3D.
+    :param num_classes: if specified (as an int), then this model will be
+        class-conditional with `num_classes` classes.
+    :param use_checkpoint: use gradient checkpointing to reduce memory usage.
+    :param num_heads: the number of attention heads in each attention layer.
+    :param num_heads_channels: if specified, ignore num_heads and instead use
+                               a fixed channel width per attention head.
+    :param num_heads_upsample: works with num_heads to set a different number
+                               of heads for upsampling. Deprecated.
+    :param use_scale_shift_norm: use a FiLM-like conditioning mechanism.
+    :param resblock_updown: use residual blocks for up/downsampling.
+    :param use_new_attention_order: use a different attention pattern for potentially
+                                    increased efficiency.
+    """
+
+    def __init__(
+        self,
+        image_size,
+        in_channels,
+        model_channels,
+        out_channels,
+        num_res_blocks,
+        attention_resolutions,
+        dropout=0,
+        channel_mult=(1, 2, 4, 8),
+        conv_resample=True,
+        dims=2,
+        num_classes=None,
+        use_checkpoint=False,
+        use_fp16=False,
+        num_heads=-1,
+        num_head_channels=-1,
+        num_heads_upsample=-1,
+        use_scale_shift_norm=False,
+        resblock_updown=False,
+        use_new_attention_order=False,
+        use_spatial_transformer=False,    # custom transformer support
+        transformer_depth=1,              # custom transformer support
+        context_dim=None,                 # custom transformer support
+        n_embed=None,                     # custom support for prediction of discrete ids into codebook of first stage vq model
+        legacy=True,
+    ):
+        super().__init__()
+        if use_spatial_transformer:
+            assert context_dim is not None, 'Fool!! You forgot to include the dimension of your cross-attention conditioning...'
+
+        if context_dim is not None:
+            assert use_spatial_transformer, 'Fool!! You forgot to use the spatial transformer for your cross-attention conditioning...'
+            from omegaconf.listconfig import ListConfig
+            if type(context_dim) == ListConfig:
+                context_dim = list(context_dim)
+
+        if num_heads_upsample == -1:
+            num_heads_upsample = num_heads
+
+        if num_heads == -1:
+            assert num_head_channels != -1, 'Either num_heads or num_head_channels has to be set'
+
+        if num_head_channels == -1:
+            assert num_heads != -1, 'Either num_heads or num_head_channels has to be set'
+
+        self.image_size = image_size
+        self.in_channels = in_channels
+        self.model_channels = model_channels
+        self.out_channels = out_channels
+        self.num_res_blocks = num_res_blocks
+        self.attention_resolutions = attention_resolutions
+        self.dropout = dropout
+        self.channel_mult = channel_mult
+        self.conv_resample = conv_resample
+        self.num_classes = num_classes
+        self.use_checkpoint = use_checkpoint
+        self.dtype = th.float16 if use_fp16 else th.float32
+        self.num_heads = num_heads
+        self.num_head_channels = num_head_channels
+        self.num_heads_upsample = num_heads_upsample
+        self.predict_codebook_ids = n_embed is not None
+
+        time_embed_dim = model_channels * 4
+        self.time_embed = nn.Sequential(
+            linear(model_channels, time_embed_dim),
+            nn.SiLU(),
+            linear(time_embed_dim, time_embed_dim),
+        )
+
+        if self.num_classes is not None:
+            self.label_emb = nn.Embedding(num_classes, time_embed_dim)
+
+        self.input_blocks = nn.ModuleList(
+            [
+                TimestepEmbedSequential(
+                    conv_nd(dims, in_channels, model_channels, 3, padding=1)
+                )
+            ]
+        )
+        self._feature_size = model_channels
+        input_block_chans = [model_channels]
+        ch = model_channels
+        ds = 1
+        for level, mult in enumerate(channel_mult):
+            for _ in range(num_res_blocks):
+                layers = [
+                    ResBlock(
+                        ch,
+                        time_embed_dim,
+                        dropout,
+                        out_channels=mult * model_channels,
+                        dims=dims,
+                        use_checkpoint=use_checkpoint,
+                        use_scale_shift_norm=use_scale_shift_norm,
+                    )
+                ]
+                ch = mult * model_channels
+                if ds in attention_resolutions:
+                    if num_head_channels == -1:
+                        dim_head = ch // num_heads
+                    else:
+                        num_heads = ch // num_head_channels
+                        dim_head = num_head_channels
+                    if legacy:
+                        #num_heads = 1
+                        dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
+                    layers.append(
+                        AttentionBlock(
+                            ch,
+                            use_checkpoint=use_checkpoint,
+                            num_heads=num_heads,
+                            num_head_channels=dim_head,
+                            use_new_attention_order=use_new_attention_order,
+                        ) if not use_spatial_transformer else SpatialTransformer(
+                            ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim
+                        )
+                    )
+                self.input_blocks.append(TimestepEmbedSequential(*layers))
+                self._feature_size += ch
+                input_block_chans.append(ch)
+            if level != len(channel_mult) - 1:
+                out_ch = ch
+                self.input_blocks.append(
+                    TimestepEmbedSequential(
+                        ResBlock(
+                            ch,
+                            time_embed_dim,
+                            dropout,
+                            out_channels=out_ch,
+                            dims=dims,
+                            use_checkpoint=use_checkpoint,
+                            use_scale_shift_norm=use_scale_shift_norm,
+                            down=True,
+                        )
+                        if resblock_updown
+                        else Downsample(
+                            ch, conv_resample, dims=dims, out_channels=out_ch
+                        )
+                    )
+                )
+                ch = out_ch
+                input_block_chans.append(ch)
+                ds *= 2
+                self._feature_size += ch
+
+        if num_head_channels == -1:
+            dim_head = ch // num_heads
+        else:
+            num_heads = ch // num_head_channels
+            dim_head = num_head_channels
+        if legacy:
+            #num_heads = 1
+            dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
+        self.middle_block = TimestepEmbedSequential(
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+            ),
+            AttentionBlock(
+                ch,
+                use_checkpoint=use_checkpoint,
+                num_heads=num_heads,
+                num_head_channels=dim_head,
+                use_new_attention_order=use_new_attention_order,
+            ) if not use_spatial_transformer else SpatialTransformer(
+                            ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim
+                        ),
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+            ),
+        )
+        self._feature_size += ch
+
+        self.output_blocks = nn.ModuleList([])
+        for level, mult in list(enumerate(channel_mult))[::-1]:
+            for i in range(num_res_blocks + 1):
+                ich = input_block_chans.pop()
+                layers = [
+                    ResBlock(
+                        ch + ich,
+                        time_embed_dim,
+                        dropout,
+                        out_channels=model_channels * mult,
+                        dims=dims,
+                        use_checkpoint=use_checkpoint,
+                        use_scale_shift_norm=use_scale_shift_norm,
+                    )
+                ]
+                ch = model_channels * mult
+                if ds in attention_resolutions:
+                    if num_head_channels == -1:
+                        dim_head = ch // num_heads
+                    else:
+                        num_heads = ch // num_head_channels
+                        dim_head = num_head_channels
+                    if legacy:
+                        #num_heads = 1
+                        dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
+                    layers.append(
+                        AttentionBlock(
+                            ch,
+                            use_checkpoint=use_checkpoint,
+                            num_heads=num_heads_upsample,
+                            num_head_channels=dim_head,
+                            use_new_attention_order=use_new_attention_order,
+                        ) if not use_spatial_transformer else SpatialTransformer(
+                            ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim
+                        )
+                    )
+                if level and i == num_res_blocks:
+                    out_ch = ch
+                    layers.append(
+                        ResBlock(
+                            ch,
+                            time_embed_dim,
+                            dropout,
+                            out_channels=out_ch,
+                            dims=dims,
+                            use_checkpoint=use_checkpoint,
+                            use_scale_shift_norm=use_scale_shift_norm,
+                            up=True,
+                        )
+                        if resblock_updown
+                        else Upsample(ch, conv_resample, dims=dims, out_channels=out_ch)
+                    )
+                    ds //= 2
+                self.output_blocks.append(TimestepEmbedSequential(*layers))
+                self._feature_size += ch
+
+        self.out = nn.Sequential(
+            normalization(ch),
+            nn.SiLU(),
+            zero_module(conv_nd(dims, model_channels, out_channels, 3, padding=1)),
+        )
+        if self.predict_codebook_ids:
+            self.id_predictor = nn.Sequential(
+            normalization(ch),
+            conv_nd(dims, model_channels, n_embed, 1),
+            #nn.LogSoftmax(dim=1)  # change to cross_entropy and produce non-normalized logits
+        )
+
+    def convert_to_fp16(self):
+        """
+        Convert the torso of the model to float16.
+        """
+        self.input_blocks.apply(convert_module_to_f16)
+        self.middle_block.apply(convert_module_to_f16)
+        self.output_blocks.apply(convert_module_to_f16)
+
+    def convert_to_fp32(self):
+        """
+        Convert the torso of the model to float32.
+        """
+        self.input_blocks.apply(convert_module_to_f32)
+        self.middle_block.apply(convert_module_to_f32)
+        self.output_blocks.apply(convert_module_to_f32)
+
+    def forward(self, x, timesteps=None, context=None, y=None,**kwargs):
+        """
+        Apply the model to an input batch.
+        :param x: an [N x C x ...] Tensor of inputs.
+        :param timesteps: a 1-D batch of timesteps.
+        :param context: conditioning plugged in via crossattn
+        :param y: an [N] Tensor of labels, if class-conditional.
+        :return: an [N x C x ...] Tensor of outputs.
+        """
+        assert (y is not None) == (
+            self.num_classes is not None
+        ), "must specify y if and only if the model is class-conditional"
+        hs = []
+        t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False)
+        emb = self.time_embed(t_emb)
+
+        if self.num_classes is not None:
+            assert y.shape == (x.shape[0],)
+            emb = emb + self.label_emb(y)
+
+        h = x.type(self.dtype)
+        for module in self.input_blocks:
+            h = module(h, emb, context)
+            hs.append(h)
+        h = self.middle_block(h, emb, context)
+        for module in self.output_blocks:
+            h = th.cat([h, hs.pop()], dim=1)
+            h = module(h, emb, context)
+        h = h.type(x.dtype)
+        if self.predict_codebook_ids:
+            return self.id_predictor(h)
+        else:
+            return self.out(h)
+
+
+class EncoderUNetModel(nn.Module):
+    """
+    The half UNet model with attention and timestep embedding.
+    For usage, see UNet.
+    """
+
+    def __init__(
+        self,
+        image_size,
+        in_channels,
+        model_channels,
+        out_channels,
+        num_res_blocks,
+        attention_resolutions,
+        dropout=0,
+        channel_mult=(1, 2, 4, 8),
+        conv_resample=True,
+        dims=2,
+        use_checkpoint=False,
+        use_fp16=False,
+        num_heads=1,
+        num_head_channels=-1,
+        num_heads_upsample=-1,
+        use_scale_shift_norm=False,
+        resblock_updown=False,
+        use_new_attention_order=False,
+        pool="adaptive",
+        *args,
+        **kwargs
+    ):
+        super().__init__()
+
+        if num_heads_upsample == -1:
+            num_heads_upsample = num_heads
+
+        self.in_channels = in_channels
+        self.model_channels = model_channels
+        self.out_channels = out_channels
+        self.num_res_blocks = num_res_blocks
+        self.attention_resolutions = attention_resolutions
+        self.dropout = dropout
+        self.channel_mult = channel_mult
+        self.conv_resample = conv_resample
+        self.use_checkpoint = use_checkpoint
+        self.dtype = th.float16 if use_fp16 else th.float32
+        self.num_heads = num_heads
+        self.num_head_channels = num_head_channels
+        self.num_heads_upsample = num_heads_upsample
+
+        time_embed_dim = model_channels * 4
+        self.time_embed = nn.Sequential(
+            linear(model_channels, time_embed_dim),
+            nn.SiLU(),
+            linear(time_embed_dim, time_embed_dim),
+        )
+
+        self.input_blocks = nn.ModuleList(
+            [
+                TimestepEmbedSequential(
+                    conv_nd(dims, in_channels, model_channels, 3, padding=1)
+                )
+            ]
+        )
+        self._feature_size = model_channels
+        input_block_chans = [model_channels]
+        ch = model_channels
+        ds = 1
+        for level, mult in enumerate(channel_mult):
+            for _ in range(num_res_blocks):
+                layers = [
+                    ResBlock(
+                        ch,
+                        time_embed_dim,
+                        dropout,
+                        out_channels=mult * model_channels,
+                        dims=dims,
+                        use_checkpoint=use_checkpoint,
+                        use_scale_shift_norm=use_scale_shift_norm,
+                    )
+                ]
+                ch = mult * model_channels
+                if ds in attention_resolutions:
+                    layers.append(
+                        AttentionBlock(
+                            ch,
+                            use_checkpoint=use_checkpoint,
+                            num_heads=num_heads,
+                            num_head_channels=num_head_channels,
+                            use_new_attention_order=use_new_attention_order,
+                        )
+                    )
+                self.input_blocks.append(TimestepEmbedSequential(*layers))
+                self._feature_size += ch
+                input_block_chans.append(ch)
+            if level != len(channel_mult) - 1:
+                out_ch = ch
+                self.input_blocks.append(
+                    TimestepEmbedSequential(
+                        ResBlock(
+                            ch,
+                            time_embed_dim,
+                            dropout,
+                            out_channels=out_ch,
+                            dims=dims,
+                            use_checkpoint=use_checkpoint,
+                            use_scale_shift_norm=use_scale_shift_norm,
+                            down=True,
+                        )
+                        if resblock_updown
+                        else Downsample(
+                            ch, conv_resample, dims=dims, out_channels=out_ch
+                        )
+                    )
+                )
+                ch = out_ch
+                input_block_chans.append(ch)
+                ds *= 2
+                self._feature_size += ch
+
+        self.middle_block = TimestepEmbedSequential(
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+            ),
+            AttentionBlock(
+                ch,
+                use_checkpoint=use_checkpoint,
+                num_heads=num_heads,
+                num_head_channels=num_head_channels,
+                use_new_attention_order=use_new_attention_order,
+            ),
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+            ),
+        )
+        self._feature_size += ch
+        self.pool = pool
+        if pool == "adaptive":
+            self.out = nn.Sequential(
+                normalization(ch),
+                nn.SiLU(),
+                nn.AdaptiveAvgPool2d((1, 1)),
+                zero_module(conv_nd(dims, ch, out_channels, 1)),
+                nn.Flatten(),
+            )
+        elif pool == "attention":
+            assert num_head_channels != -1
+            self.out = nn.Sequential(
+                normalization(ch),
+                nn.SiLU(),
+                AttentionPool2d(
+                    (image_size // ds), ch, num_head_channels, out_channels
+                ),
+            )
+        elif pool == "spatial":
+            self.out = nn.Sequential(
+                nn.Linear(self._feature_size, 2048),
+                nn.ReLU(),
+                nn.Linear(2048, self.out_channels),
+            )
+        elif pool == "spatial_v2":
+            self.out = nn.Sequential(
+                nn.Linear(self._feature_size, 2048),
+                normalization(2048),
+                nn.SiLU(),
+                nn.Linear(2048, self.out_channels),
+            )
+        else:
+            raise NotImplementedError(f"Unexpected {pool} pooling")
+
+    def convert_to_fp16(self):
+        """
+        Convert the torso of the model to float16.
+        """
+        self.input_blocks.apply(convert_module_to_f16)
+        self.middle_block.apply(convert_module_to_f16)
+
+    def convert_to_fp32(self):
+        """
+        Convert the torso of the model to float32.
+        """
+        self.input_blocks.apply(convert_module_to_f32)
+        self.middle_block.apply(convert_module_to_f32)
+
+    def forward(self, x, timesteps):
+        """
+        Apply the model to an input batch.
+        :param x: an [N x C x ...] Tensor of inputs.
+        :param timesteps: a 1-D batch of timesteps.
+        :return: an [N x K] Tensor of outputs.
+        """
+        emb = self.time_embed(timestep_embedding(timesteps, self.model_channels))
+
+        results = []
+        h = x.type(self.dtype)
+        for module in self.input_blocks:
+            h = module(h, emb)
+            if self.pool.startswith("spatial"):
+                results.append(h.type(x.dtype).mean(dim=(2, 3)))
+        h = self.middle_block(h, emb)
+        if self.pool.startswith("spatial"):
+            results.append(h.type(x.dtype).mean(dim=(2, 3)))
+            h = th.cat(results, axis=-1)
+            return self.out(h)
+        else:
+            h = h.type(x.dtype)
+            return self.out(h)
+

ldm/modules/diffusionmodules/util.py

@@ -0,0 +1,267 @@
+# adopted from
+# https://github.com/openai/improved-diffusion/blob/main/improved_diffusion/gaussian_diffusion.py
+# and
+# https://github.com/lucidrains/denoising-diffusion-pytorch/blob/7706bdfc6f527f58d33f84b7b522e61e6e3164b3/denoising_diffusion_pytorch/denoising_diffusion_pytorch.py
+# and
+# https://github.com/openai/guided-diffusion/blob/0ba878e517b276c45d1195eb29f6f5f72659a05b/guided_diffusion/nn.py
+#
+# thanks!
+
+
+import os
+import math
+import torch
+import torch.nn as nn
+import numpy as np
+from einops import repeat
+
+from ldm.util import instantiate_from_config
+
+
+def make_beta_schedule(schedule, n_timestep, linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3):
+    if schedule == "linear":
+        betas = (
+                torch.linspace(linear_start ** 0.5, linear_end ** 0.5, n_timestep, dtype=torch.float64) ** 2
+        )
+
+    elif schedule == "cosine":
+        timesteps = (
+                torch.arange(n_timestep + 1, dtype=torch.float64) / n_timestep + cosine_s
+        )
+        alphas = timesteps / (1 + cosine_s) * np.pi / 2
+        alphas = torch.cos(alphas).pow(2)
+        alphas = alphas / alphas[0]
+        betas = 1 - alphas[1:] / alphas[:-1]
+        betas = np.clip(betas, a_min=0, a_max=0.999)
+
+    elif schedule == "sqrt_linear":
+        betas = torch.linspace(linear_start, linear_end, n_timestep, dtype=torch.float64)
+    elif schedule == "sqrt":
+        betas = torch.linspace(linear_start, linear_end, n_timestep, dtype=torch.float64) ** 0.5
+    else:
+        raise ValueError(f"schedule '{schedule}' unknown.")
+    return betas.numpy()
+
+
+def make_ddim_timesteps(ddim_discr_method, num_ddim_timesteps, num_ddpm_timesteps, verbose=True):
+    if ddim_discr_method == 'uniform':
+        c = num_ddpm_timesteps // num_ddim_timesteps
+        ddim_timesteps = np.asarray(list(range(0, num_ddpm_timesteps, c)))
+    elif ddim_discr_method == 'quad':
+        ddim_timesteps = ((np.linspace(0, np.sqrt(num_ddpm_timesteps * .8), num_ddim_timesteps)) ** 2).astype(int)
+    else:
+        raise NotImplementedError(f'There is no ddim discretization method called "{ddim_discr_method}"')
+
+    # assert ddim_timesteps.shape[0] == num_ddim_timesteps
+    # add one to get the final alpha values right (the ones from first scale to data during sampling)
+    steps_out = ddim_timesteps + 1
+    if verbose:
+        print(f'Selected timesteps for ddim sampler: {steps_out}')
+    return steps_out
+
+
+def make_ddim_sampling_parameters(alphacums, ddim_timesteps, eta, verbose=True):
+    # select alphas for computing the variance schedule
+    alphas = alphacums[ddim_timesteps]
+    alphas_prev = np.asarray([alphacums[0]] + alphacums[ddim_timesteps[:-1]].tolist())
+
+    # according the the formula provided in https://arxiv.org/abs/2010.02502
+    sigmas = eta * np.sqrt((1 - alphas_prev) / (1 - alphas) * (1 - alphas / alphas_prev))
+    if verbose:
+        print(f'Selected alphas for ddim sampler: a_t: {alphas}; a_(t-1): {alphas_prev}')
+        print(f'For the chosen value of eta, which is {eta}, '
+              f'this results in the following sigma_t schedule for ddim sampler {sigmas}')
+    return sigmas, alphas, alphas_prev
+
+
+def betas_for_alpha_bar(num_diffusion_timesteps, alpha_bar, max_beta=0.999):
+    """
+    Create a beta schedule that discretizes the given alpha_t_bar function,
+    which defines the cumulative product of (1-beta) over time from t = [0,1].
+    :param num_diffusion_timesteps: the number of betas to produce.
+    :param alpha_bar: a lambda that takes an argument t from 0 to 1 and
+                      produces the cumulative product of (1-beta) up to that
+                      part of the diffusion process.
+    :param max_beta: the maximum beta to use; use values lower than 1 to
+                     prevent singularities.
+    """
+    betas = []
+    for i in range(num_diffusion_timesteps):
+        t1 = i / num_diffusion_timesteps
+        t2 = (i + 1) / num_diffusion_timesteps
+        betas.append(min(1 - alpha_bar(t2) / alpha_bar(t1), max_beta))
+    return np.array(betas)
+
+
+def extract_into_tensor(a, t, x_shape):
+    b, *_ = t.shape
+    out = a.gather(-1, t)
+    return out.reshape(b, *((1,) * (len(x_shape) - 1)))
+
+
+def checkpoint(func, inputs, params, flag):
+    """
+    Evaluate a function without caching intermediate activations, allowing for
+    reduced memory at the expense of extra compute in the backward pass.
+    :param func: the function to evaluate.
+    :param inputs: the argument sequence to pass to `func`.
+    :param params: a sequence of parameters `func` depends on but does not
+                   explicitly take as arguments.
+    :param flag: if False, disable gradient checkpointing.
+    """
+    if flag: # disabled checkpointing to allow requires_grad = False for main model
+        args = tuple(inputs) + tuple(params)
+        return CheckpointFunction.apply(func, len(inputs), *args)
+    else:
+        return func(*inputs)
+
+
+class CheckpointFunction(torch.autograd.Function):
+    @staticmethod
+    def forward(ctx, run_function, length, *args):
+        ctx.run_function = run_function
+        ctx.input_tensors = list(args[:length])
+        ctx.input_params = list(args[length:])
+
+        with torch.no_grad():
+            output_tensors = ctx.run_function(*ctx.input_tensors)
+        return output_tensors
+
+    @staticmethod
+    def backward(ctx, *output_grads):
+        ctx.input_tensors = [x.detach().requires_grad_(True) for x in ctx.input_tensors]
+        with torch.enable_grad():
+            # Fixes a bug where the first op in run_function modifies the
+            # Tensor storage in place, which is not allowed for detach()'d
+            # Tensors.
+            shallow_copies = [x.view_as(x) for x in ctx.input_tensors]
+            output_tensors = ctx.run_function(*shallow_copies)
+        input_grads = torch.autograd.grad(
+            output_tensors,
+            ctx.input_tensors + ctx.input_params,
+            output_grads,
+            allow_unused=True,
+        )
+        del ctx.input_tensors
+        del ctx.input_params
+        del output_tensors
+        return (None, None) + input_grads
+
+
+def timestep_embedding(timesteps, dim, max_period=10000, repeat_only=False):
+    """
+    Create sinusoidal timestep embeddings.
+    :param timesteps: a 1-D Tensor of N indices, one per batch element.
+                      These may be fractional.
+    :param dim: the dimension of the output.
+    :param max_period: controls the minimum frequency of the embeddings.
+    :return: an [N x dim] Tensor of positional embeddings.
+    """
+    if not repeat_only:
+        half = dim // 2
+        freqs = torch.exp(
+            -math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32) / half
+        ).to(device=timesteps.device)
+        args = timesteps[:, None].float() * freqs[None]
+        embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1)
+        if dim % 2:
+            embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
+    else:
+        embedding = repeat(timesteps, 'b -> b d', d=dim)
+    return embedding
+
+
+def zero_module(module):
+    """
+    Zero out the parameters of a module and return it.
+    """
+    for p in module.parameters():
+        p.detach().zero_()
+    return module
+
+
+def scale_module(module, scale):
+    """
+    Scale the parameters of a module and return it.
+    """
+    for p in module.parameters():
+        p.detach().mul_(scale)
+    return module
+
+
+def mean_flat(tensor):
+    """
+    Take the mean over all non-batch dimensions.
+    """
+    return tensor.mean(dim=list(range(1, len(tensor.shape))))
+
+
+def normalization(channels):
+    """
+    Make a standard normalization layer.
+    :param channels: number of input channels.
+    :return: an nn.Module for normalization.
+    """
+    return GroupNorm32(32, channels)
+
+
+# PyTorch 1.7 has SiLU, but we support PyTorch 1.5.
+class SiLU(nn.Module):
+    def forward(self, x):
+        return x * torch.sigmoid(x)
+
+
+class GroupNorm32(nn.GroupNorm):
+    def forward(self, x):
+        return super().forward(x.float()).type(x.dtype)
+
+def conv_nd(dims, *args, **kwargs):
+    """
+    Create a 1D, 2D, or 3D convolution module.
+    """
+    if dims == 1:
+        return nn.Conv1d(*args, **kwargs)
+    elif dims == 2:
+        return nn.Conv2d(*args, **kwargs)
+    elif dims == 3:
+        return nn.Conv3d(*args, **kwargs)
+    raise ValueError(f"unsupported dimensions: {dims}")
+
+
+def linear(*args, **kwargs):
+    """
+    Create a linear module.
+    """
+    return nn.Linear(*args, **kwargs)
+
+
+def avg_pool_nd(dims, *args, **kwargs):
+    """
+    Create a 1D, 2D, or 3D average pooling module.
+    """
+    if dims == 1:
+        return nn.AvgPool1d(*args, **kwargs)
+    elif dims == 2:
+        return nn.AvgPool2d(*args, **kwargs)
+    elif dims == 3:
+        return nn.AvgPool3d(*args, **kwargs)
+    raise ValueError(f"unsupported dimensions: {dims}")
+
+
+class HybridConditioner(nn.Module):
+
+    def __init__(self, c_concat_config, c_crossattn_config):
+        super().__init__()
+        self.concat_conditioner = instantiate_from_config(c_concat_config)
+        self.crossattn_conditioner = instantiate_from_config(c_crossattn_config)
+
+    def forward(self, c_concat, c_crossattn):
+        c_concat = self.concat_conditioner(c_concat)
+        c_crossattn = self.crossattn_conditioner(c_crossattn)
+        return {'c_concat': [c_concat], 'c_crossattn': [c_crossattn]}
+
+
+def noise_like(shape, device, repeat=False):
+    repeat_noise = lambda: torch.randn((1, *shape[1:]), device=device).repeat(shape[0], *((1,) * (len(shape) - 1)))
+    noise = lambda: torch.randn(shape, device=device)
+    return repeat_noise() if repeat else noise()

ldm/modules/distributions/distributions.py

@@ -0,0 +1,92 @@
+import torch
+import numpy as np
+
+
+class AbstractDistribution:
+    def sample(self):
+        raise NotImplementedError()
+
+    def mode(self):
+        raise NotImplementedError()
+
+
+class DiracDistribution(AbstractDistribution):
+    def __init__(self, value):
+        self.value = value
+
+    def sample(self):
+        return self.value
+
+    def mode(self):
+        return self.value
+
+
+class DiagonalGaussianDistribution(object):
+    def __init__(self, parameters, deterministic=False):
+        self.parameters = parameters
+        self.mean, self.logvar = torch.chunk(parameters, 2, dim=1)
+        self.logvar = torch.clamp(self.logvar, -30.0, 20.0)
+        self.deterministic = deterministic
+        self.std = torch.exp(0.5 * self.logvar)
+        self.var = torch.exp(self.logvar)
+        if self.deterministic:
+            self.var = self.std = torch.zeros_like(self.mean).to(device=self.parameters.device)
+
+    def sample(self):
+        x = self.mean + self.std * torch.randn(self.mean.shape).to(device=self.parameters.device)
+        return x
+
+    def kl(self, other=None):
+        if self.deterministic:
+            return torch.Tensor([0.])
+        else:
+            if other is None:
+                return 0.5 * torch.sum(torch.pow(self.mean, 2)
+                                       + self.var - 1.0 - self.logvar,
+                                       dim=[1, 2, 3])
+            else:
+                return 0.5 * torch.sum(
+                    torch.pow(self.mean - other.mean, 2) / other.var
+                    + self.var / other.var - 1.0 - self.logvar + other.logvar,
+                    dim=[1, 2, 3])
+
+    def nll(self, sample, dims=[1,2,3]):
+        if self.deterministic:
+            return torch.Tensor([0.])
+        logtwopi = np.log(2.0 * np.pi)
+        return 0.5 * torch.sum(
+            logtwopi + self.logvar + torch.pow(sample - self.mean, 2) / self.var,
+            dim=dims)
+
+    def mode(self):
+        return self.mean
+
+
+def normal_kl(mean1, logvar1, mean2, logvar2):
+    """
+    source: https://github.com/openai/guided-diffusion/blob/27c20a8fab9cb472df5d6bdd6c8d11c8f430b924/guided_diffusion/losses.py#L12
+    Compute the KL divergence between two gaussians.
+    Shapes are automatically broadcasted, so batches can be compared to
+    scalars, among other use cases.
+    """
+    tensor = None
+    for obj in (mean1, logvar1, mean2, logvar2):
+        if isinstance(obj, torch.Tensor):
+            tensor = obj
+            break
+    assert tensor is not None, "at least one argument must be a Tensor"
+
+    # Force variances to be Tensors. Broadcasting helps convert scalars to
+    # Tensors, but it does not work for torch.exp().
+    logvar1, logvar2 = [
+        x if isinstance(x, torch.Tensor) else torch.tensor(x).to(tensor)
+        for x in (logvar1, logvar2)
+    ]
+
+    return 0.5 * (
+        -1.0
+        + logvar2
+        - logvar1
+        + torch.exp(logvar1 - logvar2)
+        + ((mean1 - mean2) ** 2) * torch.exp(-logvar2)
+    )

ldm/modules/ema.py

@@ -0,0 +1,76 @@
+import torch
+from torch import nn
+
+
+class LitEma(nn.Module):
+    def __init__(self, model, decay=0.9999, use_num_upates=True):
+        super().__init__()
+        if decay < 0.0 or decay > 1.0:
+            raise ValueError('Decay must be between 0 and 1')
+
+        self.m_name2s_name = {}
+        self.register_buffer('decay', torch.tensor(decay, dtype=torch.float32))
+        self.register_buffer('num_updates', torch.tensor(0,dtype=torch.int) if use_num_upates
+                             else torch.tensor(-1,dtype=torch.int))
+
+        for name, p in model.named_parameters():
+            if p.requires_grad:
+                #remove as '.'-character is not allowed in buffers
+                s_name = name.replace('.','')
+                self.m_name2s_name.update({name:s_name})
+                self.register_buffer(s_name,p.clone().detach().data)
+
+        self.collected_params = []
+
+    def forward(self,model):
+        decay = self.decay
+
+        if self.num_updates >= 0:
+            self.num_updates += 1
+            decay = min(self.decay,(1 + self.num_updates) / (10 + self.num_updates))
+
+        one_minus_decay = 1.0 - decay
+
+        with torch.no_grad():
+            m_param = dict(model.named_parameters())
+            shadow_params = dict(self.named_buffers())
+
+            for key in m_param:
+                if m_param[key].requires_grad:
+                    sname = self.m_name2s_name[key]
+                    shadow_params[sname] = shadow_params[sname].type_as(m_param[key])
+                    shadow_params[sname].sub_(one_minus_decay * (shadow_params[sname] - m_param[key]))
+                else:
+                    assert not key in self.m_name2s_name
+
+    def copy_to(self, model):
+        m_param = dict(model.named_parameters())
+        shadow_params = dict(self.named_buffers())
+        for key in m_param:
+            if m_param[key].requires_grad:
+                m_param[key].data.copy_(shadow_params[self.m_name2s_name[key]].data)
+            else:
+                assert not key in self.m_name2s_name
+
+    def store(self, parameters):
+        """
+        Save the current parameters for restoring later.
+        Args:
+          parameters: Iterable of `torch.nn.Parameter`; the parameters to be
+            temporarily stored.
+        """
+        self.collected_params = [param.clone() for param in parameters]
+
+    def restore(self, parameters):
+        """
+        Restore the parameters stored with the `store` method.
+        Useful to validate the model with EMA parameters without affecting the
+        original optimization process. Store the parameters before the
+        `copy_to` method. After validation (or model saving), use this to
+        restore the former parameters.
+        Args:
+          parameters: Iterable of `torch.nn.Parameter`; the parameters to be
+            updated with the stored parameters.
+        """
+        for c_param, param in zip(self.collected_params, parameters):
+            param.data.copy_(c_param.data)

ldm/modules/embedding_manager.py

@@ -0,0 +1,161 @@
+import torch
+from torch import nn
+
+from ldm.data.personalized import per_img_token_list
+from transformers import CLIPTokenizer
+from functools import partial
+
+DEFAULT_PLACEHOLDER_TOKEN = ["*"]
+
+PROGRESSIVE_SCALE = 2000
+
+def get_clip_token_for_string(tokenizer, string):
+    batch_encoding = tokenizer(string, truncation=True, max_length=77, return_length=True,
+                               return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
+    tokens = batch_encoding["input_ids"]
+    assert torch.count_nonzero(tokens - 49407) == 2, f"String '{string}' maps to more than a single token. Please use another string"
+
+    return tokens[0, 1]
+
+def get_bert_token_for_string(tokenizer, string):
+    token = tokenizer(string)
+    assert torch.count_nonzero(token) == 3, f"String '{string}' maps to more than a single token. Please use another string"
+
+    token = token[0, 1]
+
+    return token
+
+def get_embedding_for_clip_token(embedder, token):
+    return embedder(token.unsqueeze(0))[0, 0]
+
+
+class EmbeddingManager(nn.Module):
+    def __init__(
+            self,
+            embedder,
+            placeholder_strings=None,
+            initializer_words=None,
+            per_image_tokens=False,
+            num_vectors_per_token=1,
+            progressive_words=False,
+            **kwargs
+    ):
+        super().__init__()
+
+        self.string_to_token_dict = {}
+        
+        self.string_to_param_dict = nn.ParameterDict()
+
+        self.initial_embeddings = nn.ParameterDict() # These should not be optimized
+
+        self.progressive_words = progressive_words
+        self.progressive_counter = 0
+
+        self.max_vectors_per_token = num_vectors_per_token
+
+        if hasattr(embedder, 'tokenizer'): # using Stable Diffusion's CLIP encoder
+            self.is_clip = True
+            get_token_for_string = partial(get_clip_token_for_string, embedder.tokenizer)
+            get_embedding_for_tkn = partial(get_embedding_for_clip_token, embedder.transformer.text_model.embeddings)
+            token_dim = 768
+        else: # using LDM's BERT encoder
+            self.is_clip = False
+            get_token_for_string = partial(get_bert_token_for_string, embedder.tknz_fn)
+            get_embedding_for_tkn = embedder.transformer.token_emb
+            token_dim = 1280
+
+        if per_image_tokens:
+            placeholder_strings.extend(per_img_token_list)
+
+        for idx, placeholder_string in enumerate(placeholder_strings):
+            
+            token = get_token_for_string(placeholder_string)
+
+            if initializer_words and idx < len(initializer_words):
+                init_word_token = get_token_for_string(initializer_words[idx])
+
+                with torch.no_grad():
+                    init_word_embedding = get_embedding_for_tkn(init_word_token.cpu())
+
+                token_params = torch.nn.Parameter(init_word_embedding.unsqueeze(0).repeat(num_vectors_per_token, 1), requires_grad=True)
+                self.initial_embeddings[placeholder_string] = torch.nn.Parameter(init_word_embedding.unsqueeze(0).repeat(num_vectors_per_token, 1), requires_grad=False)
+            else:
+                token_params = torch.nn.Parameter(torch.rand(size=(num_vectors_per_token, token_dim), requires_grad=True))
+            
+            self.string_to_token_dict[placeholder_string] = token
+            self.string_to_param_dict[placeholder_string] = token_params
+
+    def forward(
+            self,
+            tokenized_text,
+            embedded_text,
+    ):
+        b, n, device = *tokenized_text.shape, tokenized_text.device
+
+        for placeholder_string, placeholder_token in self.string_to_token_dict.items():
+
+            placeholder_embedding = self.string_to_param_dict[placeholder_string].to(device)
+
+            if self.max_vectors_per_token == 1: # If there's only one vector per token, we can do a simple replacement
+                placeholder_idx = torch.where(tokenized_text == placeholder_token.to(device))
+                embedded_text[placeholder_idx] = placeholder_embedding
+            else: # otherwise, need to insert and keep track of changing indices
+                if self.progressive_words:
+                    self.progressive_counter += 1
+                    max_step_tokens = 1 + self.progressive_counter // PROGRESSIVE_SCALE
+                else:
+                    max_step_tokens = self.max_vectors_per_token
+
+                num_vectors_for_token = min(placeholder_embedding.shape[0], max_step_tokens)
+
+                placeholder_rows, placeholder_cols = torch.where(tokenized_text == placeholder_token.to(device))
+
+                if placeholder_rows.nelement() == 0:
+                    continue
+
+                sorted_cols, sort_idx = torch.sort(placeholder_cols, descending=True)
+                sorted_rows = placeholder_rows[sort_idx]
+
+                for idx in range(len(sorted_rows)):
+                    row = sorted_rows[idx]
+                    col = sorted_cols[idx]
+
+                    new_token_row = torch.cat([tokenized_text[row][:col], placeholder_token.repeat(num_vectors_for_token).to(device), tokenized_text[row][col + 1:]], axis=0)[:n]
+                    new_embed_row = torch.cat([embedded_text[row][:col], placeholder_embedding[:num_vectors_for_token], embedded_text[row][col + 1:]], axis=0)[:n]
+
+                    embedded_text[row]  = new_embed_row
+                    tokenized_text[row] = new_token_row
+
+        return embedded_text
+
+    def save(self, ckpt_path):
+        torch.save({"string_to_token": self.string_to_token_dict,
+                    "string_to_param": self.string_to_param_dict}, ckpt_path)
+
+    def load(self, ckpt_path):
+        ckpt = torch.load(ckpt_path, map_location='cpu')
+
+        self.string_to_token_dict = ckpt["string_to_token"]
+        self.string_to_param_dict = ckpt["string_to_param"]
+
+    def get_embedding_norms_squared(self):
+        all_params = torch.cat(list(self.string_to_param_dict.values()), axis=0) # num_placeholders x embedding_dim
+        param_norm_squared = (all_params * all_params).sum(axis=-1)              # num_placeholders
+
+        return param_norm_squared
+
+    def embedding_parameters(self):
+        return self.string_to_param_dict.parameters()
+
+    def embedding_to_coarse_loss(self):
+        
+        loss = 0.
+        num_embeddings = len(self.initial_embeddings)
+
+        for key in self.initial_embeddings:
+            optimized = self.string_to_param_dict[key]
+            coarse = self.initial_embeddings[key].clone().to(optimized.device)
+
+            loss = loss + (optimized - coarse) @ (optimized - coarse).T / num_embeddings
+
+        return loss

ldm/modules/encoders/modules.py

@@ -0,0 +1,396 @@
+import torch
+import torch.nn as nn
+from functools import partial
+import clip
+from einops import rearrange, repeat
+from transformers import CLIPTokenizer, CLIPTextModel
+import kornia
+
+from ldm.modules.x_transformer import Encoder, TransformerWrapper  # TODO: can we directly rely on lucidrains code and simply add this as a reuirement? --> test
+
+def _expand_mask(mask, dtype, tgt_len = None):
+    """
+    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
+    """
+    bsz, src_len = mask.size()
+    tgt_len = tgt_len if tgt_len is not None else src_len
+
+    expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)
+
+    inverted_mask = 1.0 - expanded_mask
+
+    return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)
+
+def _build_causal_attention_mask(bsz, seq_len, dtype):
+        # lazily create causal attention mask, with full attention between the vision tokens
+        # pytorch uses additive attention mask; fill with -inf
+        mask = torch.empty(bsz, seq_len, seq_len, dtype=dtype)
+        mask.fill_(torch.tensor(torch.finfo(dtype).min))
+        mask.triu_(1)  # zero out the lower diagonal
+        mask = mask.unsqueeze(1)  # expand mask
+        return mask
+
+class AbstractEncoder(nn.Module):
+    def __init__(self):
+        super().__init__()
+
+    def encode(self, *args, **kwargs):
+        raise NotImplementedError
+
+
+
+class ClassEmbedder(nn.Module):
+    def __init__(self, embed_dim, n_classes=1000, key='class'):
+        super().__init__()
+        self.key = key
+        self.embedding = nn.Embedding(n_classes, embed_dim)
+
+    def forward(self, batch, key=None):
+        if key is None:
+            key = self.key
+        # this is for use in crossattn
+        c = batch[key][:, None]
+        c = self.embedding(c)
+        return c
+
+
+class TransformerEmbedder(AbstractEncoder):
+    """Some transformer encoder layers"""
+    def __init__(self, n_embed, n_layer, vocab_size, max_seq_len=77, device="cuda"):
+        super().__init__()
+        self.device = device
+        self.transformer = TransformerWrapper(num_tokens=vocab_size, max_seq_len=max_seq_len,
+                                              attn_layers=Encoder(dim=n_embed, depth=n_layer))
+
+    def forward(self, tokens):
+        tokens = tokens.to(self.device)  # meh
+        z = self.transformer(tokens, return_embeddings=True)
+        return z
+
+    def encode(self, x):
+        return self(x)
+
+
+class BERTTokenizer(AbstractEncoder):
+    """ Uses a pretrained BERT tokenizer by huggingface. Vocab size: 30522 (?)"""
+    def __init__(self, device="cuda", vq_interface=True, max_length=77):
+        super().__init__()
+        from transformers import BertTokenizerFast  # TODO: add to reuquirements
+        self.tokenizer = BertTokenizerFast.from_pretrained("bert-base-uncased")
+        self.device = device
+        self.vq_interface = vq_interface
+        self.max_length = max_length
+
+    def forward(self, text):
+        batch_encoding = self.tokenizer(text, truncation=True, max_length=self.max_length, return_length=True,
+                                        return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
+        tokens = batch_encoding["input_ids"].to(self.device)
+        return tokens
+
+    @torch.no_grad()
+    def encode(self, text):
+        tokens = self(text)
+        if not self.vq_interface:
+            return tokens
+        return None, None, [None, None, tokens]
+
+    def decode(self, text):
+        return text
+
+
+class BERTEmbedder(AbstractEncoder):
+    """Uses the BERT tokenizr model and add some transformer encoder layers"""
+    def __init__(self, n_embed, n_layer, vocab_size=30522, max_seq_len=77,
+                 device="cuda",use_tokenizer=True, embedding_dropout=0.0):
+        super().__init__()
+        self.use_tknz_fn = use_tokenizer
+        if self.use_tknz_fn:
+            self.tknz_fn = BERTTokenizer(vq_interface=False, max_length=max_seq_len)
+        self.device = device
+        self.transformer = TransformerWrapper(num_tokens=vocab_size, max_seq_len=max_seq_len,
+                                              attn_layers=Encoder(dim=n_embed, depth=n_layer),
+                                              emb_dropout=embedding_dropout)
+
+    def forward(self, text, embedding_manager=None):
+        if self.use_tknz_fn:
+            tokens = self.tknz_fn(text)#.to(self.device)
+        else:
+            tokens = text
+        z = self.transformer(tokens, return_embeddings=True, embedding_manager=embedding_manager)
+        return z
+
+    def encode(self, text, **kwargs):
+        # output of length 77
+        return self(text, **kwargs)
+
+class SpatialRescaler(nn.Module):
+    def __init__(self,
+                 n_stages=1,
+                 method='bilinear',
+                 multiplier=0.5,
+                 in_channels=3,
+                 out_channels=None,
+                 bias=False):
+        super().__init__()
+        self.n_stages = n_stages
+        assert self.n_stages >= 0
+        assert method in ['nearest','linear','bilinear','trilinear','bicubic','area']
+        self.multiplier = multiplier
+        self.interpolator = partial(torch.nn.functional.interpolate, mode=method)
+        self.remap_output = out_channels is not None
+        if self.remap_output:
+            print(f'Spatial Rescaler mapping from {in_channels} to {out_channels} channels after resizing.')
+            self.channel_mapper = nn.Conv2d(in_channels,out_channels,1,bias=bias)
+
+    def forward(self,x):
+        for stage in range(self.n_stages):
+            x = self.interpolator(x, scale_factor=self.multiplier)
+
+
+        if self.remap_output:
+            x = self.channel_mapper(x)
+        return x
+
+    def encode(self, x):
+        return self(x)
+
+class FrozenCLIPEmbedder(AbstractEncoder):
+    """Uses the CLIP transformer encoder for text (from Hugging Face)"""
+    def __init__(self, version="openai/clip-vit-large-patch14", device="cuda", max_length=77):
+        super().__init__()
+        self.tokenizer = CLIPTokenizer.from_pretrained(version)
+        self.transformer = CLIPTextModel.from_pretrained(version)
+        self.device = device
+        self.max_length = max_length
+
+        def embedding_forward(
+                self,
+                input_ids = None,
+                position_ids = None,
+                inputs_embeds = None,
+                embedding_manager = None,
+            ) -> torch.Tensor:
+
+                seq_length = input_ids.shape[-1] if input_ids is not None else inputs_embeds.shape[-2]
+
+                if position_ids is None:
+                    position_ids = self.position_ids[:, :seq_length]
+
+                if inputs_embeds is None:
+                    inputs_embeds = self.token_embedding(input_ids)
+
+                if embedding_manager is not None:
+                    inputs_embeds = embedding_manager(input_ids, inputs_embeds)
+
+
+                position_embeddings = self.position_embedding(position_ids)
+                embeddings = inputs_embeds + position_embeddings
+                
+                return embeddings      
+
+        self.transformer.text_model.embeddings.forward = embedding_forward.__get__(self.transformer.text_model.embeddings)
+
+        def encoder_forward(
+            self,
+            inputs_embeds,
+            attention_mask = None,
+            causal_attention_mask = None,
+            output_attentions = None,
+            output_hidden_states = None,
+            return_dict = None,
+        ):
+            output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+            output_hidden_states = (
+                output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+            )
+            return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+            encoder_states = () if output_hidden_states else None
+            all_attentions = () if output_attentions else None
+
+            hidden_states = inputs_embeds
+            for idx, encoder_layer in enumerate(self.layers):
+                if output_hidden_states:
+                    encoder_states = encoder_states + (hidden_states,)
+
+                layer_outputs = encoder_layer(
+                    hidden_states,
+                    attention_mask,
+                    causal_attention_mask,
+                    output_attentions=output_attentions,
+                )
+
+                hidden_states = layer_outputs[0]
+
+                if output_attentions:
+                    all_attentions = all_attentions + (layer_outputs[1],)
+
+            if output_hidden_states:
+                encoder_states = encoder_states + (hidden_states,)
+
+            return hidden_states
+
+        self.transformer.text_model.encoder.forward = encoder_forward.__get__(self.transformer.text_model.encoder)
+
+
+        def text_encoder_forward(
+            self,
+            input_ids = None,
+            attention_mask = None,
+            position_ids = None,
+            output_attentions = None,
+            output_hidden_states = None,
+            return_dict = None,
+            embedding_manager = None,
+        ):
+            output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+            output_hidden_states = (
+                output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+            )
+            return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+            if input_ids is None:
+                raise ValueError("You have to specify either input_ids")
+
+            input_shape = input_ids.size()
+            input_ids = input_ids.view(-1, input_shape[-1])
+
+            hidden_states = self.embeddings(input_ids=input_ids, position_ids=position_ids, embedding_manager=embedding_manager)
+
+            bsz, seq_len = input_shape
+            # CLIP's text model uses causal mask, prepare it here.
+            # https://github.com/openai/CLIP/blob/cfcffb90e69f37bf2ff1e988237a0fbe41f33c04/clip/model.py#L324
+            causal_attention_mask = _build_causal_attention_mask(bsz, seq_len, hidden_states.dtype).to(
+                hidden_states.device
+            )
+
+            # expand attention_mask
+            if attention_mask is not None:
+                # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+                attention_mask = _expand_mask(attention_mask, hidden_states.dtype)
+
+            last_hidden_state = self.encoder(
+                inputs_embeds=hidden_states,
+                attention_mask=attention_mask,
+                causal_attention_mask=causal_attention_mask,
+                output_attentions=output_attentions,
+                output_hidden_states=output_hidden_states,
+                return_dict=return_dict,
+            )
+
+            last_hidden_state = self.final_layer_norm(last_hidden_state)
+
+            return last_hidden_state
+
+        self.transformer.text_model.forward = text_encoder_forward.__get__(self.transformer.text_model)
+
+        def transformer_forward(
+            self,
+            input_ids = None,
+            attention_mask = None,
+            position_ids = None,
+            output_attentions = None,
+            output_hidden_states = None,
+            return_dict = None,
+            embedding_manager = None,
+        ):
+            return self.text_model(
+                input_ids=input_ids,
+                attention_mask=attention_mask,
+                position_ids=position_ids,
+                output_attentions=output_attentions,
+                output_hidden_states=output_hidden_states,
+                return_dict=return_dict,
+                embedding_manager = embedding_manager
+            )
+
+        self.transformer.forward = transformer_forward.__get__(self.transformer)
+
+
+    def freeze(self):
+        self.transformer = self.transformer.eval()
+        for param in self.parameters():
+            param.requires_grad = False
+
+    def forward(self, text, **kwargs):
+        batch_encoding = self.tokenizer(text, truncation=True, max_length=self.max_length, return_length=True,
+                                        return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
+        tokens = batch_encoding["input_ids"].to(self.device)        
+        z = self.transformer(input_ids=tokens, **kwargs)
+
+        return z
+
+    def encode(self, text, **kwargs):
+        return self(text, **kwargs)
+
+
+class FrozenCLIPTextEmbedder(nn.Module):
+    """
+    Uses the CLIP transformer encoder for text.
+    """
+    def __init__(self, version='ViT-L/14', device="cuda", max_length=77, n_repeat=1, normalize=True):
+        super().__init__()
+        self.model, _ = clip.load(version, jit=False, device="cpu")
+        self.device = device
+        self.max_length = max_length
+        self.n_repeat = n_repeat
+        self.normalize = normalize
+
+    def freeze(self):
+        self.model = self.model.eval()
+        for param in self.parameters():
+            param.requires_grad = False
+
+    def forward(self, text):
+        tokens = clip.tokenize(text).to(self.device)
+        z = self.model.encode_text(tokens)
+        if self.normalize:
+            z = z / torch.linalg.norm(z, dim=1, keepdim=True)
+        return z
+
+    def encode(self, text):
+        z = self(text)
+        if z.ndim==2:
+            z = z[:, None, :]
+        z = repeat(z, 'b 1 d -> b k d', k=self.n_repeat)
+        return z
+
+
+class FrozenClipImageEmbedder(nn.Module):
+    """
+        Uses the CLIP image encoder.
+        """
+    def __init__(
+            self,
+            model,
+            jit=False,
+            device='cuda' if torch.cuda.is_available() else 'cpu',
+            antialias=False,
+        ):
+        super().__init__()
+        self.model, _ = clip.load(name=model, device=device, jit=jit)
+
+        self.antialias = antialias
+
+        self.register_buffer('mean', torch.Tensor([0.48145466, 0.4578275, 0.40821073]), persistent=False)
+        self.register_buffer('std', torch.Tensor([0.26862954, 0.26130258, 0.27577711]), persistent=False)
+
+    def preprocess(self, x):
+        # normalize to [0,1]
+        x = kornia.geometry.resize(x, (224, 224),
+                                   interpolation='bicubic',align_corners=True,
+                                   antialias=self.antialias)
+        x = (x + 1.) / 2.
+        # renormalize according to clip
+        x = kornia.enhance.normalize(x, self.mean, self.std)
+        return x
+
+    def forward(self, x):
+        # x is assumed to be in range [-1,1]
+        return self.model.encode_image(self.preprocess(x))
+
+
+if __name__ == "__main__":
+    from ldm.util import count_params
+    model = FrozenCLIPEmbedder()
+    count_params(model, verbose=True)

ldm/modules/encoders/modules_bak.py

@@ -0,0 +1,496 @@
+import torch
+import torch.nn as nn
+from functools import partial
+import clip
+from einops import rearrange, repeat
+from transformers import CLIPTokenizer, CLIPTextModel
+import kornia
+
+from ldm.modules.x_transformer import Encoder, TransformerWrapper  # TODO: can we directly rely on lucidrains code and simply add this as a reuirement? --> test
+
+def _expand_mask(mask, dtype, tgt_len = None):
+    """
+    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
+    """
+    bsz, src_len = mask.size()
+    tgt_len = tgt_len if tgt_len is not None else src_len
+
+    expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)
+
+    inverted_mask = 1.0 - expanded_mask
+
+    return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)
+
+def _build_causal_attention_mask(bsz, seq_len, dtype):
+        # lazily create causal attention mask, with full attention between the vision tokens
+        # pytorch uses additive attention mask; fill with -inf
+        mask = torch.empty(bsz, seq_len, seq_len, dtype=dtype)
+        mask.fill_(torch.tensor(torch.finfo(dtype).min))
+        mask.triu_(1)  # zero out the lower diagonal
+        mask = mask.unsqueeze(1)  # expand mask
+        return mask
+
+class AbstractEncoder(nn.Module):
+    def __init__(self):
+        super().__init__()
+
+    def encode(self, *args, **kwargs):
+        raise NotImplementedError
+
+
+
+class ClassEmbedder(nn.Module):
+    def __init__(self, embed_dim, n_classes=1000, key='class'):
+        super().__init__()
+        self.key = key
+        self.embedding = nn.Embedding(n_classes, embed_dim)
+
+    def forward(self, batch, key=None):
+        if key is None:
+            key = self.key
+        # this is for use in crossattn
+        c = batch[key][:, None]
+        c = self.embedding(c)
+        return c
+
+
+class TransformerEmbedder(AbstractEncoder):
+    """Some transformer encoder layers"""
+    def __init__(self, n_embed, n_layer, vocab_size, max_seq_len=77, device="cuda"):
+        super().__init__()
+        self.device = device
+        self.transformer = TransformerWrapper(num_tokens=vocab_size, max_seq_len=max_seq_len,
+                                              attn_layers=Encoder(dim=n_embed, depth=n_layer))
+
+    def forward(self, tokens):
+        tokens = tokens.to(self.device)  # meh
+        z = self.transformer(tokens, return_embeddings=True)
+        return z
+
+    def encode(self, x):
+        return self(x)
+
+
+class BERTTokenizer(AbstractEncoder):
+    """ Uses a pretrained BERT tokenizer by huggingface. Vocab size: 30522 (?)"""
+    def __init__(self, device="cuda", vq_interface=True, max_length=77):
+        super().__init__()
+        from transformers import BertTokenizerFast  # TODO: add to reuquirements
+        self.tokenizer = BertTokenizerFast.from_pretrained("bert-base-uncased")
+        self.device = device
+        self.vq_interface = vq_interface
+        self.max_length = max_length
+
+    def forward(self, text):
+        batch_encoding = self.tokenizer(text, truncation=True, max_length=self.max_length, return_length=True,
+                                        return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
+        tokens = batch_encoding["input_ids"].to(self.device)
+        return tokens
+
+    @torch.no_grad()
+    def encode(self, text):
+        tokens = self(text)
+        if not self.vq_interface:
+            return tokens
+        return None, None, [None, None, tokens]
+
+    def decode(self, text):
+        return text
+
+
+class BERTEmbedder(AbstractEncoder):
+    """Uses the BERT tokenizr model and add some transformer encoder layers"""
+    def __init__(self, n_embed, n_layer, vocab_size=30522, max_seq_len=77,
+                 device="cuda",use_tokenizer=True, embedding_dropout=0.0):
+        super().__init__()
+        self.use_tknz_fn = use_tokenizer
+        if self.use_tknz_fn:
+            self.tknz_fn = BERTTokenizer(vq_interface=False, max_length=max_seq_len)
+        self.device = device
+        self.transformer = TransformerWrapper(num_tokens=vocab_size, max_seq_len=max_seq_len,
+                                              attn_layers=Encoder(dim=n_embed, depth=n_layer),
+                                              emb_dropout=embedding_dropout)
+
+    def forward(self, text, embedding_manager=None):
+        if self.use_tknz_fn:
+            tokens = self.tknz_fn(text)#.to(self.device)
+        else:
+            tokens = text
+        z = self.transformer(tokens, return_embeddings=True, embedding_manager=embedding_manager)
+        return z
+
+    def encode(self, text, **kwargs):
+        # output of length 77
+        return self(text, **kwargs)
+
+class SpatialRescaler(nn.Module):
+    def __init__(self,
+                 n_stages=1,
+                 method='bilinear',
+                 multiplier=0.5,
+                 in_channels=3,
+                 out_channels=None,
+                 bias=False):
+        super().__init__()
+        self.n_stages = n_stages
+        assert self.n_stages >= 0
+        assert method in ['nearest','linear','bilinear','trilinear','bicubic','area']
+        self.multiplier = multiplier
+        self.interpolator = partial(torch.nn.functional.interpolate, mode=method)
+        self.remap_output = out_channels is not None
+        if self.remap_output:
+            print(f'Spatial Rescaler mapping from {in_channels} to {out_channels} channels after resizing.')
+            self.channel_mapper = nn.Conv2d(in_channels,out_channels,1,bias=bias)
+
+    def forward(self,x):
+        for stage in range(self.n_stages):
+            x = self.interpolator(x, scale_factor=self.multiplier)
+
+
+        if self.remap_output:
+            x = self.channel_mapper(x)
+        return x
+
+    def encode(self, x):
+        return self(x)
+
+class FrozenCLIPEmbedder(AbstractEncoder):
+    """Uses the CLIP transformer encoder for text (from Hugging Face)"""
+    def __init__(self, version="openai/clip-vit-large-patch14", device="cuda", max_length=77):
+        super().__init__()
+        self.tokenizer = CLIPTokenizer.from_pretrained(version)
+        self.transformer = CLIPTextModel.from_pretrained(version)
+        self.device = device
+        self.max_length = max_length
+        self.freeze()
+
+        def embedding_forward(
+                self,
+                input_ids = None,
+                position_ids = None,
+                inputs_embeds = None,
+                embedding_manager = None,
+            ) -> torch.Tensor:
+
+                seq_length = input_ids.shape[-1] if input_ids is not None else inputs_embeds.shape[-2]
+
+                if position_ids is None:
+                    position_ids = self.position_ids[:, :seq_length]
+
+                if inputs_embeds is None:
+                    inputs_embeds = self.token_embedding(input_ids)
+
+                if embedding_manager is not None:
+                    inputs_embeds = embedding_manager(input_ids, inputs_embeds)
+
+
+                position_embeddings = self.position_embedding(position_ids)
+                embeddings = inputs_embeds + position_embeddings
+                
+                return embeddings      
+
+        self.transformer.text_model.embeddings.forward = embedding_forward.__get__(self.transformer.text_model.embeddings)
+
+        def encoder_forward(
+            self,
+            inputs_embeds,
+            attention_mask = None,
+            causal_attention_mask = None,
+            output_attentions = None,
+            output_hidden_states = None,
+            return_dict = None,
+        ):
+            output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+            output_hidden_states = (
+                output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+            )
+            return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+            encoder_states = () if output_hidden_states else None
+            all_attentions = () if output_attentions else None
+
+            hidden_states = inputs_embeds
+            for idx, encoder_layer in enumerate(self.layers):
+                if output_hidden_states:
+                    encoder_states = encoder_states + (hidden_states,)
+
+                layer_outputs = encoder_layer(
+                    hidden_states,
+                    attention_mask,
+                    causal_attention_mask,
+                    output_attentions=output_attentions,
+                )
+
+                hidden_states = layer_outputs[0]
+
+                if output_attentions:
+                    all_attentions = all_attentions + (layer_outputs[1],)
+
+            if output_hidden_states:
+                encoder_states = encoder_states + (hidden_states,)
+
+            return hidden_states
+
+        self.transformer.text_model.encoder.forward = encoder_forward.__get__(self.transformer.text_model.encoder)
+
+
+        def text_encoder_forward(
+            self,
+            input_ids = None,
+            attention_mask = None,
+            position_ids = None,
+            output_attentions = None,
+            output_hidden_states = None,
+            return_dict = None,
+            embedding_manager = None,
+        ):
+            output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+            output_hidden_states = (
+                output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+            )
+            return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+            if input_ids is None:
+                raise ValueError("You have to specify either input_ids")
+
+            input_shape = input_ids.size()
+            input_ids = input_ids.view(-1, input_shape[-1])
+
+            hidden_states = self.embeddings(input_ids=input_ids, position_ids=position_ids, embedding_manager=embedding_manager)
+
+            bsz, seq_len = input_shape
+            # CLIP's text model uses causal mask, prepare it here.
+            # https://github.com/openai/CLIP/blob/cfcffb90e69f37bf2ff1e988237a0fbe41f33c04/clip/model.py#L324
+            causal_attention_mask = _build_causal_attention_mask(bsz, seq_len, hidden_states.dtype).to(
+                hidden_states.device
+            )
+
+            # expand attention_mask
+            if attention_mask is not None:
+                # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+                attention_mask = _expand_mask(attention_mask, hidden_states.dtype)
+
+            last_hidden_state = self.encoder(
+                inputs_embeds=hidden_states,
+                attention_mask=attention_mask,
+                causal_attention_mask=causal_attention_mask,
+                output_attentions=output_attentions,
+                output_hidden_states=output_hidden_states,
+                return_dict=return_dict,
+            )
+
+            last_hidden_state = self.final_layer_norm(last_hidden_state)
+
+            return last_hidden_state
+
+        self.transformer.text_model.forward = text_encoder_forward.__get__(self.transformer.text_model)
+
+        def transformer_forward(
+            self,
+            input_ids = None,
+            attention_mask = None,
+            position_ids = None,
+            output_attentions = None,
+            output_hidden_states = None,
+            return_dict = None,
+            embedding_manager = None,
+        ):
+            return self.text_model(
+                input_ids=input_ids,
+                attention_mask=attention_mask,
+                position_ids=position_ids,
+                output_attentions=output_attentions,
+                output_hidden_states=output_hidden_states,
+                return_dict=return_dict,
+                embedding_manager = embedding_manager
+            )
+
+        self.transformer.forward = transformer_forward.__get__(self.transformer)
+
+
+    # def update_embedding_func(self, embedding_manager):
+    #     text_model = self.transformer.text_model
+    #     # text_model.old_embeddings = text_model.embeddings
+
+    #     # def new_embeddings(
+    #     #         input_ids = None,
+    #     #         position_ids = None,
+    #     #         inputs_embeds = None,
+    #     #     ) -> torch.Tensor:
+
+    #     #         seq_length = input_ids.shape[-1] if input_ids is not None else inputs_embeds.shape[-2]
+
+    #     #         if position_ids is None:
+    #     #             position_ids = text_model.old_embeddings.position_ids[:, :seq_length]
+
+    #     #         if inputs_embeds is None:
+    #     #             inputs_embeds = text_model.old_embeddings.token_embedding(input_ids)
+
+                    
+    #     #         inputs_embeds = embedding_manager(input_ids, inputs_embeds)
+
+    #     #         position_embeddings = text_model.old_embeddings.position_embedding(position_ids)
+    #     #         embeddings = inputs_embeds + position_embeddings
+            
+    #     #         return embeddings  
+
+    #     # del text_model.embeddings
+    #     # text_model.embeddings = new_embeddings
+
+    #     # class NewEmbeddings(torch.nn.Module):
+
+    #     #     def __init__(self, orig_embedder):
+    #     #         super().__init__()
+    #     #         self.orig_embedder = orig_embedder
+            
+    #     #     def forward(
+    #     #         self,
+    #     #         input_ids = None,
+    #     #         position_ids = None,
+    #     #         inputs_embeds = None,
+    #     #     ) -> torch.Tensor:
+
+    #     #         seq_length = input_ids.shape[-1] if input_ids is not None else inputs_embeds.shape[-2]
+
+    #     #         if position_ids is None:
+    #     #             position_ids = self.orig_embedder.position_ids[:, :seq_length]
+
+    #     #         if inputs_embeds is None:
+    #     #             inputs_embeds = self.orig_embedder.token_embedding(input_ids)
+                    
+    #     #         inputs_embeds = embedding_manager(input_ids, inputs_embeds)
+
+    #     #         position_embeddings = self.orig_embedder.position_embedding(position_ids)
+    #     #         embeddings = inputs_embeds + position_embeddings
+            
+    #     #         return embeddings      
+
+    #     # # self.new_embeddings = 
+    #     # # text_model.embeddings = new_embeddings.__call__.__get__(text_model)
+    #     # text_model.embeddings = NewEmbeddings(text_model.embeddings)
+        
+    #     class NewEmbeddings(torch.nn.Module):
+
+    #         def __init__(self, orig_embedder, embedding_manager):
+    #             super().__init__()
+    #             self.embedding_manager = embedding_manager
+    #             self.orig_embedder     = orig_embedder
+            
+    #         def forward(
+    #             self,
+    #             input_ids = None,
+    #             position_ids = None,
+    #             inputs_embeds = None,
+    #         ) -> torch.Tensor:
+
+    #             seq_length = input_ids.shape[-1] if input_ids is not None else inputs_embeds.shape[-2]
+
+    #             if position_ids is None:
+    #                 position_ids = self.orig_embedder.position_ids[:, :seq_length]
+
+    #             if inputs_embeds is None:
+    #                 inputs_embeds = self.orig_embedder.token_embedding(input_ids)
+                
+    #             # init_embeds = inputs_embeds.clone()
+    #             inputs_embeds = self.embedding_manager(input_ids, inputs_embeds)
+
+    #             # print(inputs_embeds - init_embeds)
+    #             # print((inputs_embeds - init_embeds).max())
+    #             # exit(0)
+
+    #             position_embeddings = self.orig_embedder.position_embedding(position_ids)
+    #             embeddings = inputs_embeds + position_embeddings
+                
+    #             return embeddings      
+
+    #     # self.new_embeddings = 
+    #     # text_model.embeddings = new_embeddings.__call__.__get__(text_model)
+    #     text_model.embeddings = NewEmbeddings(text_model.embeddings, embedding_manager)
+
+    def freeze(self):
+        self.transformer = self.transformer.eval()
+        for param in self.parameters():
+            param.requires_grad = False
+
+    def forward(self, text, **kwargs):
+        batch_encoding = self.tokenizer(text, truncation=True, max_length=self.max_length, return_length=True,
+                                        return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
+        tokens = batch_encoding["input_ids"].to(self.device)        
+        z = self.transformer(input_ids=tokens, **kwargs)
+
+        return z
+
+    def encode(self, text, **kwargs):
+        return self(text, **kwargs)
+
+
+class FrozenCLIPTextEmbedder(nn.Module):
+    """
+    Uses the CLIP transformer encoder for text.
+    """
+    def __init__(self, version='ViT-L/14', device="cuda", max_length=77, n_repeat=1, normalize=True):
+        super().__init__()
+        self.model, _ = clip.load(version, jit=False, device="cpu")
+        self.device = device
+        self.max_length = max_length
+        self.n_repeat = n_repeat
+        self.normalize = normalize
+
+    def freeze(self):
+        self.model = self.model.eval()
+        for param in self.parameters():
+            param.requires_grad = False
+
+    def forward(self, text):
+        tokens = clip.tokenize(text).to(self.device)
+        z = self.model.encode_text(tokens)
+        if self.normalize:
+            z = z / torch.linalg.norm(z, dim=1, keepdim=True)
+        return z
+
+    def encode(self, text):
+        z = self(text)
+        if z.ndim==2:
+            z = z[:, None, :]
+        z = repeat(z, 'b 1 d -> b k d', k=self.n_repeat)
+        return z
+
+
+class FrozenClipImageEmbedder(nn.Module):
+    """
+        Uses the CLIP image encoder.
+        """
+    def __init__(
+            self,
+            model,
+            jit=False,
+            device='cuda' if torch.cuda.is_available() else 'cpu',
+            antialias=False,
+        ):
+        super().__init__()
+        self.model, _ = clip.load(name=model, device=device, jit=jit)
+
+        self.antialias = antialias
+
+        self.register_buffer('mean', torch.Tensor([0.48145466, 0.4578275, 0.40821073]), persistent=False)
+        self.register_buffer('std', torch.Tensor([0.26862954, 0.26130258, 0.27577711]), persistent=False)
+
+    def preprocess(self, x):
+        # normalize to [0,1]
+        x = kornia.geometry.resize(x, (224, 224),
+                                   interpolation='bicubic',align_corners=True,
+                                   antialias=self.antialias)
+        x = (x + 1.) / 2.
+        # renormalize according to clip
+        x = kornia.enhance.normalize(x, self.mean, self.std)
+        return x
+
+    def forward(self, x):
+        # x is assumed to be in range [-1,1]
+        return self.model.encode_image(self.preprocess(x))
+
+
+if __name__ == "__main__":
+    from ldm.util import count_params
+    model = FrozenCLIPEmbedder()
+    count_params(model, verbose=True)

ldm/modules/image_degradation/__init__.py

@@ -0,0 +1,2 @@
+from ldm.modules.image_degradation.bsrgan import degradation_bsrgan_variant as degradation_fn_bsr
+from ldm.modules.image_degradation.bsrgan_light import degradation_bsrgan_variant as degradation_fn_bsr_light

ldm/modules/image_degradation/bsrgan.py

@@ -0,0 +1,730 @@
+# -*- coding: utf-8 -*-
+"""
+# --------------------------------------------
+# Super-Resolution
+# --------------------------------------------
+#
+# Kai Zhang (cskaizhang@gmail.com)
+# https://github.com/cszn
+# From 2019/03--2021/08
+# --------------------------------------------
+"""
+
+import numpy as np
+import cv2
+import torch
+
+from functools import partial
+import random
+from scipy import ndimage
+import scipy
+import scipy.stats as ss
+from scipy.interpolate import interp2d
+from scipy.linalg import orth
+import albumentations
+
+import ldm.modules.image_degradation.utils_image as util
+
+
+def modcrop_np(img, sf):
+    '''
+    Args:
+        img: numpy image, WxH or WxHxC
+        sf: scale factor
+    Return:
+        cropped image
+    '''
+    w, h = img.shape[:2]
+    im = np.copy(img)
+    return im[:w - w % sf, :h - h % sf, ...]
+
+
+"""
+# --------------------------------------------
+# anisotropic Gaussian kernels
+# --------------------------------------------
+"""
+
+
+def analytic_kernel(k):
+    """Calculate the X4 kernel from the X2 kernel (for proof see appendix in paper)"""
+    k_size = k.shape[0]
+    # Calculate the big kernels size
+    big_k = np.zeros((3 * k_size - 2, 3 * k_size - 2))
+    # Loop over the small kernel to fill the big one
+    for r in range(k_size):
+        for c in range(k_size):
+            big_k[2 * r:2 * r + k_size, 2 * c:2 * c + k_size] += k[r, c] * k
+    # Crop the edges of the big kernel to ignore very small values and increase run time of SR
+    crop = k_size // 2
+    cropped_big_k = big_k[crop:-crop, crop:-crop]
+    # Normalize to 1
+    return cropped_big_k / cropped_big_k.sum()
+
+
+def anisotropic_Gaussian(ksize=15, theta=np.pi, l1=6, l2=6):
+    """ generate an anisotropic Gaussian kernel
+    Args:
+        ksize : e.g., 15, kernel size
+        theta : [0,  pi], rotation angle range
+        l1    : [0.1,50], scaling of eigenvalues
+        l2    : [0.1,l1], scaling of eigenvalues
+        If l1 = l2, will get an isotropic Gaussian kernel.
+    Returns:
+        k     : kernel
+    """
+
+    v = np.dot(np.array([[np.cos(theta), -np.sin(theta)], [np.sin(theta), np.cos(theta)]]), np.array([1., 0.]))
+    V = np.array([[v[0], v[1]], [v[1], -v[0]]])
+    D = np.array([[l1, 0], [0, l2]])
+    Sigma = np.dot(np.dot(V, D), np.linalg.inv(V))
+    k = gm_blur_kernel(mean=[0, 0], cov=Sigma, size=ksize)
+
+    return k
+
+
+def gm_blur_kernel(mean, cov, size=15):
+    center = size / 2.0 + 0.5
+    k = np.zeros([size, size])
+    for y in range(size):
+        for x in range(size):
+            cy = y - center + 1
+            cx = x - center + 1
+            k[y, x] = ss.multivariate_normal.pdf([cx, cy], mean=mean, cov=cov)
+
+    k = k / np.sum(k)
+    return k
+
+
+def shift_pixel(x, sf, upper_left=True):
+    """shift pixel for super-resolution with different scale factors
+    Args:
+        x: WxHxC or WxH
+        sf: scale factor
+        upper_left: shift direction
+    """
+    h, w = x.shape[:2]
+    shift = (sf - 1) * 0.5
+    xv, yv = np.arange(0, w, 1.0), np.arange(0, h, 1.0)
+    if upper_left:
+        x1 = xv + shift
+        y1 = yv + shift
+    else:
+        x1 = xv - shift
+        y1 = yv - shift
+
+    x1 = np.clip(x1, 0, w - 1)
+    y1 = np.clip(y1, 0, h - 1)
+
+    if x.ndim == 2:
+        x = interp2d(xv, yv, x)(x1, y1)
+    if x.ndim == 3:
+        for i in range(x.shape[-1]):
+            x[:, :, i] = interp2d(xv, yv, x[:, :, i])(x1, y1)
+
+    return x
+
+
+def blur(x, k):
+    '''
+    x: image, NxcxHxW
+    k: kernel, Nx1xhxw
+    '''
+    n, c = x.shape[:2]
+    p1, p2 = (k.shape[-2] - 1) // 2, (k.shape[-1] - 1) // 2
+    x = torch.nn.functional.pad(x, pad=(p1, p2, p1, p2), mode='replicate')
+    k = k.repeat(1, c, 1, 1)
+    k = k.view(-1, 1, k.shape[2], k.shape[3])
+    x = x.view(1, -1, x.shape[2], x.shape[3])
+    x = torch.nn.functional.conv2d(x, k, bias=None, stride=1, padding=0, groups=n * c)
+    x = x.view(n, c, x.shape[2], x.shape[3])
+
+    return x
+
+
+def gen_kernel(k_size=np.array([15, 15]), scale_factor=np.array([4, 4]), min_var=0.6, max_var=10., noise_level=0):
+    """"
+    # modified version of https://github.com/assafshocher/BlindSR_dataset_generator
+    # Kai Zhang
+    # min_var = 0.175 * sf  # variance of the gaussian kernel will be sampled between min_var and max_var
+    # max_var = 2.5 * sf
+    """
+    # Set random eigen-vals (lambdas) and angle (theta) for COV matrix
+    lambda_1 = min_var + np.random.rand() * (max_var - min_var)
+    lambda_2 = min_var + np.random.rand() * (max_var - min_var)
+    theta = np.random.rand() * np.pi  # random theta
+    noise = -noise_level + np.random.rand(*k_size) * noise_level * 2
+
+    # Set COV matrix using Lambdas and Theta
+    LAMBDA = np.diag([lambda_1, lambda_2])
+    Q = np.array([[np.cos(theta), -np.sin(theta)],
+                  [np.sin(theta), np.cos(theta)]])
+    SIGMA = Q @ LAMBDA @ Q.T
+    INV_SIGMA = np.linalg.inv(SIGMA)[None, None, :, :]
+
+    # Set expectation position (shifting kernel for aligned image)
+    MU = k_size // 2 - 0.5 * (scale_factor - 1)  # - 0.5 * (scale_factor - k_size % 2)
+    MU = MU[None, None, :, None]
+
+    # Create meshgrid for Gaussian
+    [X, Y] = np.meshgrid(range(k_size[0]), range(k_size[1]))
+    Z = np.stack([X, Y], 2)[:, :, :, None]
+
+    # Calcualte Gaussian for every pixel of the kernel
+    ZZ = Z - MU
+    ZZ_t = ZZ.transpose(0, 1, 3, 2)
+    raw_kernel = np.exp(-0.5 * np.squeeze(ZZ_t @ INV_SIGMA @ ZZ)) * (1 + noise)
+
+    # shift the kernel so it will be centered
+    # raw_kernel_centered = kernel_shift(raw_kernel, scale_factor)
+
+    # Normalize the kernel and return
+    # kernel = raw_kernel_centered / np.sum(raw_kernel_centered)
+    kernel = raw_kernel / np.sum(raw_kernel)
+    return kernel
+
+
+def fspecial_gaussian(hsize, sigma):
+    hsize = [hsize, hsize]
+    siz = [(hsize[0] - 1.0) / 2.0, (hsize[1] - 1.0) / 2.0]
+    std = sigma
+    [x, y] = np.meshgrid(np.arange(-siz[1], siz[1] + 1), np.arange(-siz[0], siz[0] + 1))
+    arg = -(x * x + y * y) / (2 * std * std)
+    h = np.exp(arg)
+    h[h < scipy.finfo(float).eps * h.max()] = 0
+    sumh = h.sum()
+    if sumh != 0:
+        h = h / sumh
+    return h
+
+
+def fspecial_laplacian(alpha):
+    alpha = max([0, min([alpha, 1])])
+    h1 = alpha / (alpha + 1)
+    h2 = (1 - alpha) / (alpha + 1)
+    h = [[h1, h2, h1], [h2, -4 / (alpha + 1), h2], [h1, h2, h1]]
+    h = np.array(h)
+    return h
+
+
+def fspecial(filter_type, *args, **kwargs):
+    '''
+    python code from:
+    https://github.com/ronaldosena/imagens-medicas-2/blob/40171a6c259edec7827a6693a93955de2bd39e76/Aulas/aula_2_-_uniform_filter/matlab_fspecial.py
+    '''
+    if filter_type == 'gaussian':
+        return fspecial_gaussian(*args, **kwargs)
+    if filter_type == 'laplacian':
+        return fspecial_laplacian(*args, **kwargs)
+
+
+"""
+# --------------------------------------------
+# degradation models
+# --------------------------------------------
+"""
+
+
+def bicubic_degradation(x, sf=3):
+    '''
+    Args:
+        x: HxWxC image, [0, 1]
+        sf: down-scale factor
+    Return:
+        bicubicly downsampled LR image
+    '''
+    x = util.imresize_np(x, scale=1 / sf)
+    return x
+
+
+def srmd_degradation(x, k, sf=3):
+    ''' blur + bicubic downsampling
+    Args:
+        x: HxWxC image, [0, 1]
+        k: hxw, double
+        sf: down-scale factor
+    Return:
+        downsampled LR image
+    Reference:
+        @inproceedings{zhang2018learning,
+          title={Learning a single convolutional super-resolution network for multiple degradations},
+          author={Zhang, Kai and Zuo, Wangmeng and Zhang, Lei},
+          booktitle={IEEE Conference on Computer Vision and Pattern Recognition},
+          pages={3262--3271},
+          year={2018}
+        }
+    '''
+    x = ndimage.filters.convolve(x, np.expand_dims(k, axis=2), mode='wrap')  # 'nearest' | 'mirror'
+    x = bicubic_degradation(x, sf=sf)
+    return x
+
+
+def dpsr_degradation(x, k, sf=3):
+    ''' bicubic downsampling + blur
+    Args:
+        x: HxWxC image, [0, 1]
+        k: hxw, double
+        sf: down-scale factor
+    Return:
+        downsampled LR image
+    Reference:
+        @inproceedings{zhang2019deep,
+          title={Deep Plug-and-Play Super-Resolution for Arbitrary Blur Kernels},
+          author={Zhang, Kai and Zuo, Wangmeng and Zhang, Lei},
+          booktitle={IEEE Conference on Computer Vision and Pattern Recognition},
+          pages={1671--1681},
+          year={2019}
+        }
+    '''
+    x = bicubic_degradation(x, sf=sf)
+    x = ndimage.filters.convolve(x, np.expand_dims(k, axis=2), mode='wrap')
+    return x
+
+
+def classical_degradation(x, k, sf=3):
+    ''' blur + downsampling
+    Args:
+        x: HxWxC image, [0, 1]/[0, 255]
+        k: hxw, double
+        sf: down-scale factor
+    Return:
+        downsampled LR image
+    '''
+    x = ndimage.filters.convolve(x, np.expand_dims(k, axis=2), mode='wrap')
+    # x = filters.correlate(x, np.expand_dims(np.flip(k), axis=2))
+    st = 0
+    return x[st::sf, st::sf, ...]
+
+
+def add_sharpening(img, weight=0.5, radius=50, threshold=10):
+    """USM sharpening. borrowed from real-ESRGAN
+    Input image: I; Blurry image: B.
+    1. K = I + weight * (I - B)
+    2. Mask = 1 if abs(I - B) > threshold, else: 0
+    3. Blur mask:
+    4. Out = Mask * K + (1 - Mask) * I
+    Args:
+        img (Numpy array): Input image, HWC, BGR; float32, [0, 1].
+        weight (float): Sharp weight. Default: 1.
+        radius (float): Kernel size of Gaussian blur. Default: 50.
+        threshold (int):
+    """
+    if radius % 2 == 0:
+        radius += 1
+    blur = cv2.GaussianBlur(img, (radius, radius), 0)
+    residual = img - blur
+    mask = np.abs(residual) * 255 > threshold
+    mask = mask.astype('float32')
+    soft_mask = cv2.GaussianBlur(mask, (radius, radius), 0)
+
+    K = img + weight * residual
+    K = np.clip(K, 0, 1)
+    return soft_mask * K + (1 - soft_mask) * img
+
+
+def add_blur(img, sf=4):
+    wd2 = 4.0 + sf
+    wd = 2.0 + 0.2 * sf
+    if random.random() < 0.5:
+        l1 = wd2 * random.random()
+        l2 = wd2 * random.random()
+        k = anisotropic_Gaussian(ksize=2 * random.randint(2, 11) + 3, theta=random.random() * np.pi, l1=l1, l2=l2)
+    else:
+        k = fspecial('gaussian', 2 * random.randint(2, 11) + 3, wd * random.random())
+    img = ndimage.filters.convolve(img, np.expand_dims(k, axis=2), mode='mirror')
+
+    return img
+
+
+def add_resize(img, sf=4):
+    rnum = np.random.rand()
+    if rnum > 0.8:  # up
+        sf1 = random.uniform(1, 2)
+    elif rnum < 0.7:  # down
+        sf1 = random.uniform(0.5 / sf, 1)
+    else:
+        sf1 = 1.0
+    img = cv2.resize(img, (int(sf1 * img.shape[1]), int(sf1 * img.shape[0])), interpolation=random.choice([1, 2, 3]))
+    img = np.clip(img, 0.0, 1.0)
+
+    return img
+
+
+# def add_Gaussian_noise(img, noise_level1=2, noise_level2=25):
+#     noise_level = random.randint(noise_level1, noise_level2)
+#     rnum = np.random.rand()
+#     if rnum > 0.6:  # add color Gaussian noise
+#         img += np.random.normal(0, noise_level / 255.0, img.shape).astype(np.float32)
+#     elif rnum < 0.4:  # add grayscale Gaussian noise
+#         img += np.random.normal(0, noise_level / 255.0, (*img.shape[:2], 1)).astype(np.float32)
+#     else:  # add  noise
+#         L = noise_level2 / 255.
+#         D = np.diag(np.random.rand(3))
+#         U = orth(np.random.rand(3, 3))
+#         conv = np.dot(np.dot(np.transpose(U), D), U)
+#         img += np.random.multivariate_normal([0, 0, 0], np.abs(L ** 2 * conv), img.shape[:2]).astype(np.float32)
+#     img = np.clip(img, 0.0, 1.0)
+#     return img
+
+def add_Gaussian_noise(img, noise_level1=2, noise_level2=25):
+    noise_level = random.randint(noise_level1, noise_level2)
+    rnum = np.random.rand()
+    if rnum > 0.6:  # add color Gaussian noise
+        img = img + np.random.normal(0, noise_level / 255.0, img.shape).astype(np.float32)
+    elif rnum < 0.4:  # add grayscale Gaussian noise
+        img = img + np.random.normal(0, noise_level / 255.0, (*img.shape[:2], 1)).astype(np.float32)
+    else:  # add  noise
+        L = noise_level2 / 255.
+        D = np.diag(np.random.rand(3))
+        U = orth(np.random.rand(3, 3))
+        conv = np.dot(np.dot(np.transpose(U), D), U)
+        img = img + np.random.multivariate_normal([0, 0, 0], np.abs(L ** 2 * conv), img.shape[:2]).astype(np.float32)
+    img = np.clip(img, 0.0, 1.0)
+    return img
+
+
+def add_speckle_noise(img, noise_level1=2, noise_level2=25):
+    noise_level = random.randint(noise_level1, noise_level2)
+    img = np.clip(img, 0.0, 1.0)
+    rnum = random.random()
+    if rnum > 0.6:
+        img += img * np.random.normal(0, noise_level / 255.0, img.shape).astype(np.float32)
+    elif rnum < 0.4:
+        img += img * np.random.normal(0, noise_level / 255.0, (*img.shape[:2], 1)).astype(np.float32)
+    else:
+        L = noise_level2 / 255.
+        D = np.diag(np.random.rand(3))
+        U = orth(np.random.rand(3, 3))
+        conv = np.dot(np.dot(np.transpose(U), D), U)
+        img += img * np.random.multivariate_normal([0, 0, 0], np.abs(L ** 2 * conv), img.shape[:2]).astype(np.float32)
+    img = np.clip(img, 0.0, 1.0)
+    return img
+
+
+def add_Poisson_noise(img):
+    img = np.clip((img * 255.0).round(), 0, 255) / 255.
+    vals = 10 ** (2 * random.random() + 2.0)  # [2, 4]
+    if random.random() < 0.5:
+        img = np.random.poisson(img * vals).astype(np.float32) / vals
+    else:
+        img_gray = np.dot(img[..., :3], [0.299, 0.587, 0.114])
+        img_gray = np.clip((img_gray * 255.0).round(), 0, 255) / 255.
+        noise_gray = np.random.poisson(img_gray * vals).astype(np.float32) / vals - img_gray
+        img += noise_gray[:, :, np.newaxis]
+    img = np.clip(img, 0.0, 1.0)
+    return img
+
+
+def add_JPEG_noise(img):
+    quality_factor = random.randint(30, 95)
+    img = cv2.cvtColor(util.single2uint(img), cv2.COLOR_RGB2BGR)
+    result, encimg = cv2.imencode('.jpg', img, [int(cv2.IMWRITE_JPEG_QUALITY), quality_factor])
+    img = cv2.imdecode(encimg, 1)
+    img = cv2.cvtColor(util.uint2single(img), cv2.COLOR_BGR2RGB)
+    return img
+
+
+def random_crop(lq, hq, sf=4, lq_patchsize=64):
+    h, w = lq.shape[:2]
+    rnd_h = random.randint(0, h - lq_patchsize)
+    rnd_w = random.randint(0, w - lq_patchsize)
+    lq = lq[rnd_h:rnd_h + lq_patchsize, rnd_w:rnd_w + lq_patchsize, :]
+
+    rnd_h_H, rnd_w_H = int(rnd_h * sf), int(rnd_w * sf)
+    hq = hq[rnd_h_H:rnd_h_H + lq_patchsize * sf, rnd_w_H:rnd_w_H + lq_patchsize * sf, :]
+    return lq, hq
+
+
+def degradation_bsrgan(img, sf=4, lq_patchsize=72, isp_model=None):
+    """
+    This is the degradation model of BSRGAN from the paper
+    "Designing a Practical Degradation Model for Deep Blind Image Super-Resolution"
+    ----------
+    img: HXWXC, [0, 1], its size should be large than (lq_patchsizexsf)x(lq_patchsizexsf)
+    sf: scale factor
+    isp_model: camera ISP model
+    Returns
+    -------
+    img: low-quality patch, size: lq_patchsizeXlq_patchsizeXC, range: [0, 1]
+    hq: corresponding high-quality patch, size: (lq_patchsizexsf)X(lq_patchsizexsf)XC, range: [0, 1]
+    """
+    isp_prob, jpeg_prob, scale2_prob = 0.25, 0.9, 0.25
+    sf_ori = sf
+
+    h1, w1 = img.shape[:2]
+    img = img.copy()[:w1 - w1 % sf, :h1 - h1 % sf, ...]  # mod crop
+    h, w = img.shape[:2]
+
+    if h < lq_patchsize * sf or w < lq_patchsize * sf:
+        raise ValueError(f'img size ({h1}X{w1}) is too small!')
+
+    hq = img.copy()
+
+    if sf == 4 and random.random() < scale2_prob:  # downsample1
+        if np.random.rand() < 0.5:
+            img = cv2.resize(img, (int(1 / 2 * img.shape[1]), int(1 / 2 * img.shape[0])),
+                             interpolation=random.choice([1, 2, 3]))
+        else:
+            img = util.imresize_np(img, 1 / 2, True)
+        img = np.clip(img, 0.0, 1.0)
+        sf = 2
+
+    shuffle_order = random.sample(range(7), 7)
+    idx1, idx2 = shuffle_order.index(2), shuffle_order.index(3)
+    if idx1 > idx2:  # keep downsample3 last
+        shuffle_order[idx1], shuffle_order[idx2] = shuffle_order[idx2], shuffle_order[idx1]
+
+    for i in shuffle_order:
+
+        if i == 0:
+            img = add_blur(img, sf=sf)
+
+        elif i == 1:
+            img = add_blur(img, sf=sf)
+
+        elif i == 2:
+            a, b = img.shape[1], img.shape[0]
+            # downsample2
+            if random.random() < 0.75:
+                sf1 = random.uniform(1, 2 * sf)
+                img = cv2.resize(img, (int(1 / sf1 * img.shape[1]), int(1 / sf1 * img.shape[0])),
+                                 interpolation=random.choice([1, 2, 3]))
+            else:
+                k = fspecial('gaussian', 25, random.uniform(0.1, 0.6 * sf))
+                k_shifted = shift_pixel(k, sf)
+                k_shifted = k_shifted / k_shifted.sum()  # blur with shifted kernel
+                img = ndimage.filters.convolve(img, np.expand_dims(k_shifted, axis=2), mode='mirror')
+                img = img[0::sf, 0::sf, ...]  # nearest downsampling
+            img = np.clip(img, 0.0, 1.0)
+
+        elif i == 3:
+            # downsample3
+            img = cv2.resize(img, (int(1 / sf * a), int(1 / sf * b)), interpolation=random.choice([1, 2, 3]))
+            img = np.clip(img, 0.0, 1.0)
+
+        elif i == 4:
+            # add Gaussian noise
+            img = add_Gaussian_noise(img, noise_level1=2, noise_level2=25)
+
+        elif i == 5:
+            # add JPEG noise
+            if random.random() < jpeg_prob:
+                img = add_JPEG_noise(img)
+
+        elif i == 6:
+            # add processed camera sensor noise
+            if random.random() < isp_prob and isp_model is not None:
+                with torch.no_grad():
+                    img, hq = isp_model.forward(img.copy(), hq)
+
+    # add final JPEG compression noise
+    img = add_JPEG_noise(img)
+
+    # random crop
+    img, hq = random_crop(img, hq, sf_ori, lq_patchsize)
+
+    return img, hq
+
+
+# todo no isp_model?
+def degradation_bsrgan_variant(image, sf=4, isp_model=None):
+    """
+    This is the degradation model of BSRGAN from the paper
+    "Designing a Practical Degradation Model for Deep Blind Image Super-Resolution"
+    ----------
+    sf: scale factor
+    isp_model: camera ISP model
+    Returns
+    -------
+    img: low-quality patch, size: lq_patchsizeXlq_patchsizeXC, range: [0, 1]
+    hq: corresponding high-quality patch, size: (lq_patchsizexsf)X(lq_patchsizexsf)XC, range: [0, 1]
+    """
+    image = util.uint2single(image)
+    isp_prob, jpeg_prob, scale2_prob = 0.25, 0.9, 0.25
+    sf_ori = sf
+
+    h1, w1 = image.shape[:2]
+    image = image.copy()[:w1 - w1 % sf, :h1 - h1 % sf, ...]  # mod crop
+    h, w = image.shape[:2]
+
+    hq = image.copy()
+
+    if sf == 4 and random.random() < scale2_prob:  # downsample1
+        if np.random.rand() < 0.5:
+            image = cv2.resize(image, (int(1 / 2 * image.shape[1]), int(1 / 2 * image.shape[0])),
+                               interpolation=random.choice([1, 2, 3]))
+        else:
+            image = util.imresize_np(image, 1 / 2, True)
+        image = np.clip(image, 0.0, 1.0)
+        sf = 2
+
+    shuffle_order = random.sample(range(7), 7)
+    idx1, idx2 = shuffle_order.index(2), shuffle_order.index(3)
+    if idx1 > idx2:  # keep downsample3 last
+        shuffle_order[idx1], shuffle_order[idx2] = shuffle_order[idx2], shuffle_order[idx1]
+
+    for i in shuffle_order:
+
+        if i == 0:
+            image = add_blur(image, sf=sf)
+
+        elif i == 1:
+            image = add_blur(image, sf=sf)
+
+        elif i == 2:
+            a, b = image.shape[1], image.shape[0]
+            # downsample2
+            if random.random() < 0.75:
+                sf1 = random.uniform(1, 2 * sf)
+                image = cv2.resize(image, (int(1 / sf1 * image.shape[1]), int(1 / sf1 * image.shape[0])),
+                                   interpolation=random.choice([1, 2, 3]))
+            else:
+                k = fspecial('gaussian', 25, random.uniform(0.1, 0.6 * sf))
+                k_shifted = shift_pixel(k, sf)
+                k_shifted = k_shifted / k_shifted.sum()  # blur with shifted kernel
+                image = ndimage.filters.convolve(image, np.expand_dims(k_shifted, axis=2), mode='mirror')
+                image = image[0::sf, 0::sf, ...]  # nearest downsampling
+            image = np.clip(image, 0.0, 1.0)
+
+        elif i == 3:
+            # downsample3
+            image = cv2.resize(image, (int(1 / sf * a), int(1 / sf * b)), interpolation=random.choice([1, 2, 3]))
+            image = np.clip(image, 0.0, 1.0)
+
+        elif i == 4:
+            # add Gaussian noise
+            image = add_Gaussian_noise(image, noise_level1=2, noise_level2=25)
+
+        elif i == 5:
+            # add JPEG noise
+            if random.random() < jpeg_prob:
+                image = add_JPEG_noise(image)
+
+        # elif i == 6:
+        #     # add processed camera sensor noise
+        #     if random.random() < isp_prob and isp_model is not None:
+        #         with torch.no_grad():
+        #             img, hq = isp_model.forward(img.copy(), hq)
+
+    # add final JPEG compression noise
+    image = add_JPEG_noise(image)
+    image = util.single2uint(image)
+    example = {"image":image}
+    return example
+
+
+# TODO incase there is a pickle error one needs to replace a += x with a = a + x in add_speckle_noise etc...
+def degradation_bsrgan_plus(img, sf=4, shuffle_prob=0.5, use_sharp=True, lq_patchsize=64, isp_model=None):
+    """
+    This is an extended degradation model by combining
+    the degradation models of BSRGAN and Real-ESRGAN
+    ----------
+    img: HXWXC, [0, 1], its size should be large than (lq_patchsizexsf)x(lq_patchsizexsf)
+    sf: scale factor
+    use_shuffle: the degradation shuffle
+    use_sharp: sharpening the img
+    Returns
+    -------
+    img: low-quality patch, size: lq_patchsizeXlq_patchsizeXC, range: [0, 1]
+    hq: corresponding high-quality patch, size: (lq_patchsizexsf)X(lq_patchsizexsf)XC, range: [0, 1]
+    """
+
+    h1, w1 = img.shape[:2]
+    img = img.copy()[:w1 - w1 % sf, :h1 - h1 % sf, ...]  # mod crop
+    h, w = img.shape[:2]
+
+    if h < lq_patchsize * sf or w < lq_patchsize * sf:
+        raise ValueError(f'img size ({h1}X{w1}) is too small!')
+
+    if use_sharp:
+        img = add_sharpening(img)
+    hq = img.copy()
+
+    if random.random() < shuffle_prob:
+        shuffle_order = random.sample(range(13), 13)
+    else:
+        shuffle_order = list(range(13))
+        # local shuffle for noise, JPEG is always the last one
+        shuffle_order[2:6] = random.sample(shuffle_order[2:6], len(range(2, 6)))
+        shuffle_order[9:13] = random.sample(shuffle_order[9:13], len(range(9, 13)))
+
+    poisson_prob, speckle_prob, isp_prob = 0.1, 0.1, 0.1
+
+    for i in shuffle_order:
+        if i == 0:
+            img = add_blur(img, sf=sf)
+        elif i == 1:
+            img = add_resize(img, sf=sf)
+        elif i == 2:
+            img = add_Gaussian_noise(img, noise_level1=2, noise_level2=25)
+        elif i == 3:
+            if random.random() < poisson_prob:
+                img = add_Poisson_noise(img)
+        elif i == 4:
+            if random.random() < speckle_prob:
+                img = add_speckle_noise(img)
+        elif i == 5:
+            if random.random() < isp_prob and isp_model is not None:
+                with torch.no_grad():
+                    img, hq = isp_model.forward(img.copy(), hq)
+        elif i == 6:
+            img = add_JPEG_noise(img)
+        elif i == 7:
+            img = add_blur(img, sf=sf)
+        elif i == 8:
+            img = add_resize(img, sf=sf)
+        elif i == 9:
+            img = add_Gaussian_noise(img, noise_level1=2, noise_level2=25)
+        elif i == 10:
+            if random.random() < poisson_prob:
+                img = add_Poisson_noise(img)
+        elif i == 11:
+            if random.random() < speckle_prob:
+                img = add_speckle_noise(img)
+        elif i == 12:
+            if random.random() < isp_prob and isp_model is not None:
+                with torch.no_grad():
+                    img, hq = isp_model.forward(img.copy(), hq)
+        else:
+            print('check the shuffle!')
+
+    # resize to desired size
+    img = cv2.resize(img, (int(1 / sf * hq.shape[1]), int(1 / sf * hq.shape[0])),
+                     interpolation=random.choice([1, 2, 3]))
+
+    # add final JPEG compression noise
+    img = add_JPEG_noise(img)
+
+    # random crop
+    img, hq = random_crop(img, hq, sf, lq_patchsize)
+
+    return img, hq
+
+
+if __name__ == '__main__':
+	print("hey")
+	img = util.imread_uint('utils/test.png', 3)
+	print(img)
+	img = util.uint2single(img)
+	print(img)
+	img = img[:448, :448]
+	h = img.shape[0] // 4
+	print("resizing to", h)
+	sf = 4
+	deg_fn = partial(degradation_bsrgan_variant, sf=sf)
+	for i in range(20):
+		print(i)
+		img_lq = deg_fn(img)
+		print(img_lq)
+		img_lq_bicubic = albumentations.SmallestMaxSize(max_size=h, interpolation=cv2.INTER_CUBIC)(image=img)["image"]
+		print(img_lq.shape)
+		print("bicubic", img_lq_bicubic.shape)
+		print(img_hq.shape)
+		lq_nearest = cv2.resize(util.single2uint(img_lq), (int(sf * img_lq.shape[1]), int(sf * img_lq.shape[0])),
+		                        interpolation=0)
+		lq_bicubic_nearest = cv2.resize(util.single2uint(img_lq_bicubic), (int(sf * img_lq.shape[1]), int(sf * img_lq.shape[0])),
+		                        interpolation=0)
+		img_concat = np.concatenate([lq_bicubic_nearest, lq_nearest, util.single2uint(img_hq)], axis=1)
+		util.imsave(img_concat, str(i) + '.png')
+
+

ldm/modules/image_degradation/bsrgan_light.py

@@ -0,0 +1,650 @@
+# -*- coding: utf-8 -*-
+import numpy as np
+import cv2
+import torch
+
+from functools import partial
+import random
+from scipy import ndimage
+import scipy
+import scipy.stats as ss
+from scipy.interpolate import interp2d
+from scipy.linalg import orth
+import albumentations
+
+import ldm.modules.image_degradation.utils_image as util
+
+"""
+# --------------------------------------------
+# Super-Resolution
+# --------------------------------------------
+#
+# Kai Zhang (cskaizhang@gmail.com)
+# https://github.com/cszn
+# From 2019/03--2021/08
+# --------------------------------------------
+"""
+
+
+def modcrop_np(img, sf):
+    '''
+    Args:
+        img: numpy image, WxH or WxHxC
+        sf: scale factor
+    Return:
+        cropped image
+    '''
+    w, h = img.shape[:2]
+    im = np.copy(img)
+    return im[:w - w % sf, :h - h % sf, ...]
+
+
+"""
+# --------------------------------------------
+# anisotropic Gaussian kernels
+# --------------------------------------------
+"""
+
+
+def analytic_kernel(k):
+    """Calculate the X4 kernel from the X2 kernel (for proof see appendix in paper)"""
+    k_size = k.shape[0]
+    # Calculate the big kernels size
+    big_k = np.zeros((3 * k_size - 2, 3 * k_size - 2))
+    # Loop over the small kernel to fill the big one
+    for r in range(k_size):
+        for c in range(k_size):
+            big_k[2 * r:2 * r + k_size, 2 * c:2 * c + k_size] += k[r, c] * k
+    # Crop the edges of the big kernel to ignore very small values and increase run time of SR
+    crop = k_size // 2
+    cropped_big_k = big_k[crop:-crop, crop:-crop]
+    # Normalize to 1
+    return cropped_big_k / cropped_big_k.sum()
+
+
+def anisotropic_Gaussian(ksize=15, theta=np.pi, l1=6, l2=6):
+    """ generate an anisotropic Gaussian kernel
+    Args:
+        ksize : e.g., 15, kernel size
+        theta : [0,  pi], rotation angle range
+        l1    : [0.1,50], scaling of eigenvalues
+        l2    : [0.1,l1], scaling of eigenvalues
+        If l1 = l2, will get an isotropic Gaussian kernel.
+    Returns:
+        k     : kernel
+    """
+
+    v = np.dot(np.array([[np.cos(theta), -np.sin(theta)], [np.sin(theta), np.cos(theta)]]), np.array([1., 0.]))
+    V = np.array([[v[0], v[1]], [v[1], -v[0]]])
+    D = np.array([[l1, 0], [0, l2]])
+    Sigma = np.dot(np.dot(V, D), np.linalg.inv(V))
+    k = gm_blur_kernel(mean=[0, 0], cov=Sigma, size=ksize)
+
+    return k
+
+
+def gm_blur_kernel(mean, cov, size=15):
+    center = size / 2.0 + 0.5
+    k = np.zeros([size, size])
+    for y in range(size):
+        for x in range(size):
+            cy = y - center + 1
+            cx = x - center + 1
+            k[y, x] = ss.multivariate_normal.pdf([cx, cy], mean=mean, cov=cov)
+
+    k = k / np.sum(k)
+    return k
+
+
+def shift_pixel(x, sf, upper_left=True):
+    """shift pixel for super-resolution with different scale factors
+    Args:
+        x: WxHxC or WxH
+        sf: scale factor
+        upper_left: shift direction
+    """
+    h, w = x.shape[:2]
+    shift = (sf - 1) * 0.5
+    xv, yv = np.arange(0, w, 1.0), np.arange(0, h, 1.0)
+    if upper_left:
+        x1 = xv + shift
+        y1 = yv + shift
+    else:
+        x1 = xv - shift
+        y1 = yv - shift
+
+    x1 = np.clip(x1, 0, w - 1)
+    y1 = np.clip(y1, 0, h - 1)
+
+    if x.ndim == 2:
+        x = interp2d(xv, yv, x)(x1, y1)
+    if x.ndim == 3:
+        for i in range(x.shape[-1]):
+            x[:, :, i] = interp2d(xv, yv, x[:, :, i])(x1, y1)
+
+    return x
+
+
+def blur(x, k):
+    '''
+    x: image, NxcxHxW
+    k: kernel, Nx1xhxw
+    '''
+    n, c = x.shape[:2]
+    p1, p2 = (k.shape[-2] - 1) // 2, (k.shape[-1] - 1) // 2
+    x = torch.nn.functional.pad(x, pad=(p1, p2, p1, p2), mode='replicate')
+    k = k.repeat(1, c, 1, 1)
+    k = k.view(-1, 1, k.shape[2], k.shape[3])
+    x = x.view(1, -1, x.shape[2], x.shape[3])
+    x = torch.nn.functional.conv2d(x, k, bias=None, stride=1, padding=0, groups=n * c)
+    x = x.view(n, c, x.shape[2], x.shape[3])
+
+    return x
+
+
+def gen_kernel(k_size=np.array([15, 15]), scale_factor=np.array([4, 4]), min_var=0.6, max_var=10., noise_level=0):
+    """"
+    # modified version of https://github.com/assafshocher/BlindSR_dataset_generator
+    # Kai Zhang
+    # min_var = 0.175 * sf  # variance of the gaussian kernel will be sampled between min_var and max_var
+    # max_var = 2.5 * sf
+    """
+    # Set random eigen-vals (lambdas) and angle (theta) for COV matrix
+    lambda_1 = min_var + np.random.rand() * (max_var - min_var)
+    lambda_2 = min_var + np.random.rand() * (max_var - min_var)
+    theta = np.random.rand() * np.pi  # random theta
+    noise = -noise_level + np.random.rand(*k_size) * noise_level * 2
+
+    # Set COV matrix using Lambdas and Theta
+    LAMBDA = np.diag([lambda_1, lambda_2])
+    Q = np.array([[np.cos(theta), -np.sin(theta)],
+                  [np.sin(theta), np.cos(theta)]])
+    SIGMA = Q @ LAMBDA @ Q.T
+    INV_SIGMA = np.linalg.inv(SIGMA)[None, None, :, :]
+
+    # Set expectation position (shifting kernel for aligned image)
+    MU = k_size // 2 - 0.5 * (scale_factor - 1)  # - 0.5 * (scale_factor - k_size % 2)
+    MU = MU[None, None, :, None]
+
+    # Create meshgrid for Gaussian
+    [X, Y] = np.meshgrid(range(k_size[0]), range(k_size[1]))
+    Z = np.stack([X, Y], 2)[:, :, :, None]
+
+    # Calcualte Gaussian for every pixel of the kernel
+    ZZ = Z - MU
+    ZZ_t = ZZ.transpose(0, 1, 3, 2)
+    raw_kernel = np.exp(-0.5 * np.squeeze(ZZ_t @ INV_SIGMA @ ZZ)) * (1 + noise)
+
+    # shift the kernel so it will be centered
+    # raw_kernel_centered = kernel_shift(raw_kernel, scale_factor)
+
+    # Normalize the kernel and return
+    # kernel = raw_kernel_centered / np.sum(raw_kernel_centered)
+    kernel = raw_kernel / np.sum(raw_kernel)
+    return kernel
+
+
+def fspecial_gaussian(hsize, sigma):
+    hsize = [hsize, hsize]
+    siz = [(hsize[0] - 1.0) / 2.0, (hsize[1] - 1.0) / 2.0]
+    std = sigma
+    [x, y] = np.meshgrid(np.arange(-siz[1], siz[1] + 1), np.arange(-siz[0], siz[0] + 1))
+    arg = -(x * x + y * y) / (2 * std * std)
+    h = np.exp(arg)
+    h[h < scipy.finfo(float).eps * h.max()] = 0
+    sumh = h.sum()
+    if sumh != 0:
+        h = h / sumh
+    return h
+
+
+def fspecial_laplacian(alpha):
+    alpha = max([0, min([alpha, 1])])
+    h1 = alpha / (alpha + 1)
+    h2 = (1 - alpha) / (alpha + 1)
+    h = [[h1, h2, h1], [h2, -4 / (alpha + 1), h2], [h1, h2, h1]]
+    h = np.array(h)
+    return h
+
+
+def fspecial(filter_type, *args, **kwargs):
+    '''
+    python code from:
+    https://github.com/ronaldosena/imagens-medicas-2/blob/40171a6c259edec7827a6693a93955de2bd39e76/Aulas/aula_2_-_uniform_filter/matlab_fspecial.py
+    '''
+    if filter_type == 'gaussian':
+        return fspecial_gaussian(*args, **kwargs)
+    if filter_type == 'laplacian':
+        return fspecial_laplacian(*args, **kwargs)
+
+
+"""
+# --------------------------------------------
+# degradation models
+# --------------------------------------------
+"""
+
+
+def bicubic_degradation(x, sf=3):
+    '''
+    Args:
+        x: HxWxC image, [0, 1]
+        sf: down-scale factor
+    Return:
+        bicubicly downsampled LR image
+    '''
+    x = util.imresize_np(x, scale=1 / sf)
+    return x
+
+
+def srmd_degradation(x, k, sf=3):
+    ''' blur + bicubic downsampling
+    Args:
+        x: HxWxC image, [0, 1]
+        k: hxw, double
+        sf: down-scale factor
+    Return:
+        downsampled LR image
+    Reference:
+        @inproceedings{zhang2018learning,
+          title={Learning a single convolutional super-resolution network for multiple degradations},
+          author={Zhang, Kai and Zuo, Wangmeng and Zhang, Lei},
+          booktitle={IEEE Conference on Computer Vision and Pattern Recognition},
+          pages={3262--3271},
+          year={2018}
+        }
+    '''
+    x = ndimage.filters.convolve(x, np.expand_dims(k, axis=2), mode='wrap')  # 'nearest' | 'mirror'
+    x = bicubic_degradation(x, sf=sf)
+    return x
+
+
+def dpsr_degradation(x, k, sf=3):
+    ''' bicubic downsampling + blur
+    Args:
+        x: HxWxC image, [0, 1]
+        k: hxw, double
+        sf: down-scale factor
+    Return:
+        downsampled LR image
+    Reference:
+        @inproceedings{zhang2019deep,
+          title={Deep Plug-and-Play Super-Resolution for Arbitrary Blur Kernels},
+          author={Zhang, Kai and Zuo, Wangmeng and Zhang, Lei},
+          booktitle={IEEE Conference on Computer Vision and Pattern Recognition},
+          pages={1671--1681},
+          year={2019}
+        }
+    '''
+    x = bicubic_degradation(x, sf=sf)
+    x = ndimage.filters.convolve(x, np.expand_dims(k, axis=2), mode='wrap')
+    return x
+
+
+def classical_degradation(x, k, sf=3):
+    ''' blur + downsampling
+    Args:
+        x: HxWxC image, [0, 1]/[0, 255]
+        k: hxw, double
+        sf: down-scale factor
+    Return:
+        downsampled LR image
+    '''
+    x = ndimage.filters.convolve(x, np.expand_dims(k, axis=2), mode='wrap')
+    # x = filters.correlate(x, np.expand_dims(np.flip(k), axis=2))
+    st = 0
+    return x[st::sf, st::sf, ...]
+
+
+def add_sharpening(img, weight=0.5, radius=50, threshold=10):
+    """USM sharpening. borrowed from real-ESRGAN
+    Input image: I; Blurry image: B.
+    1. K = I + weight * (I - B)
+    2. Mask = 1 if abs(I - B) > threshold, else: 0
+    3. Blur mask:
+    4. Out = Mask * K + (1 - Mask) * I
+    Args:
+        img (Numpy array): Input image, HWC, BGR; float32, [0, 1].
+        weight (float): Sharp weight. Default: 1.
+        radius (float): Kernel size of Gaussian blur. Default: 50.
+        threshold (int):
+    """
+    if radius % 2 == 0:
+        radius += 1
+    blur = cv2.GaussianBlur(img, (radius, radius), 0)
+    residual = img - blur
+    mask = np.abs(residual) * 255 > threshold
+    mask = mask.astype('float32')
+    soft_mask = cv2.GaussianBlur(mask, (radius, radius), 0)
+
+    K = img + weight * residual
+    K = np.clip(K, 0, 1)
+    return soft_mask * K + (1 - soft_mask) * img
+
+
+def add_blur(img, sf=4):
+    wd2 = 4.0 + sf
+    wd = 2.0 + 0.2 * sf
+
+    wd2 = wd2/4
+    wd = wd/4
+
+    if random.random() < 0.5:
+        l1 = wd2 * random.random()
+        l2 = wd2 * random.random()
+        k = anisotropic_Gaussian(ksize=random.randint(2, 11) + 3, theta=random.random() * np.pi, l1=l1, l2=l2)
+    else:
+        k = fspecial('gaussian', random.randint(2, 4) + 3, wd * random.random())
+    img = ndimage.filters.convolve(img, np.expand_dims(k, axis=2), mode='mirror')
+
+    return img
+
+
+def add_resize(img, sf=4):
+    rnum = np.random.rand()
+    if rnum > 0.8:  # up
+        sf1 = random.uniform(1, 2)
+    elif rnum < 0.7:  # down
+        sf1 = random.uniform(0.5 / sf, 1)
+    else:
+        sf1 = 1.0
+    img = cv2.resize(img, (int(sf1 * img.shape[1]), int(sf1 * img.shape[0])), interpolation=random.choice([1, 2, 3]))
+    img = np.clip(img, 0.0, 1.0)
+
+    return img
+
+
+# def add_Gaussian_noise(img, noise_level1=2, noise_level2=25):
+#     noise_level = random.randint(noise_level1, noise_level2)
+#     rnum = np.random.rand()
+#     if rnum > 0.6:  # add color Gaussian noise
+#         img += np.random.normal(0, noise_level / 255.0, img.shape).astype(np.float32)
+#     elif rnum < 0.4:  # add grayscale Gaussian noise
+#         img += np.random.normal(0, noise_level / 255.0, (*img.shape[:2], 1)).astype(np.float32)
+#     else:  # add  noise
+#         L = noise_level2 / 255.
+#         D = np.diag(np.random.rand(3))
+#         U = orth(np.random.rand(3, 3))
+#         conv = np.dot(np.dot(np.transpose(U), D), U)
+#         img += np.random.multivariate_normal([0, 0, 0], np.abs(L ** 2 * conv), img.shape[:2]).astype(np.float32)
+#     img = np.clip(img, 0.0, 1.0)
+#     return img
+
+def add_Gaussian_noise(img, noise_level1=2, noise_level2=25):
+    noise_level = random.randint(noise_level1, noise_level2)
+    rnum = np.random.rand()
+    if rnum > 0.6:  # add color Gaussian noise
+        img = img + np.random.normal(0, noise_level / 255.0, img.shape).astype(np.float32)
+    elif rnum < 0.4:  # add grayscale Gaussian noise
+        img = img + np.random.normal(0, noise_level / 255.0, (*img.shape[:2], 1)).astype(np.float32)
+    else:  # add  noise
+        L = noise_level2 / 255.
+        D = np.diag(np.random.rand(3))
+        U = orth(np.random.rand(3, 3))
+        conv = np.dot(np.dot(np.transpose(U), D), U)
+        img = img + np.random.multivariate_normal([0, 0, 0], np.abs(L ** 2 * conv), img.shape[:2]).astype(np.float32)
+    img = np.clip(img, 0.0, 1.0)
+    return img
+
+
+def add_speckle_noise(img, noise_level1=2, noise_level2=25):
+    noise_level = random.randint(noise_level1, noise_level2)
+    img = np.clip(img, 0.0, 1.0)
+    rnum = random.random()
+    if rnum > 0.6:
+        img += img * np.random.normal(0, noise_level / 255.0, img.shape).astype(np.float32)
+    elif rnum < 0.4:
+        img += img * np.random.normal(0, noise_level / 255.0, (*img.shape[:2], 1)).astype(np.float32)
+    else:
+        L = noise_level2 / 255.
+        D = np.diag(np.random.rand(3))
+        U = orth(np.random.rand(3, 3))
+        conv = np.dot(np.dot(np.transpose(U), D), U)
+        img += img * np.random.multivariate_normal([0, 0, 0], np.abs(L ** 2 * conv), img.shape[:2]).astype(np.float32)
+    img = np.clip(img, 0.0, 1.0)
+    return img
+
+
+def add_Poisson_noise(img):
+    img = np.clip((img * 255.0).round(), 0, 255) / 255.
+    vals = 10 ** (2 * random.random() + 2.0)  # [2, 4]
+    if random.random() < 0.5:
+        img = np.random.poisson(img * vals).astype(np.float32) / vals
+    else:
+        img_gray = np.dot(img[..., :3], [0.299, 0.587, 0.114])
+        img_gray = np.clip((img_gray * 255.0).round(), 0, 255) / 255.
+        noise_gray = np.random.poisson(img_gray * vals).astype(np.float32) / vals - img_gray
+        img += noise_gray[:, :, np.newaxis]
+    img = np.clip(img, 0.0, 1.0)
+    return img
+
+
+def add_JPEG_noise(img):
+    quality_factor = random.randint(80, 95)
+    img = cv2.cvtColor(util.single2uint(img), cv2.COLOR_RGB2BGR)
+    result, encimg = cv2.imencode('.jpg', img, [int(cv2.IMWRITE_JPEG_QUALITY), quality_factor])
+    img = cv2.imdecode(encimg, 1)
+    img = cv2.cvtColor(util.uint2single(img), cv2.COLOR_BGR2RGB)
+    return img
+
+
+def random_crop(lq, hq, sf=4, lq_patchsize=64):
+    h, w = lq.shape[:2]
+    rnd_h = random.randint(0, h - lq_patchsize)
+    rnd_w = random.randint(0, w - lq_patchsize)
+    lq = lq[rnd_h:rnd_h + lq_patchsize, rnd_w:rnd_w + lq_patchsize, :]
+
+    rnd_h_H, rnd_w_H = int(rnd_h * sf), int(rnd_w * sf)
+    hq = hq[rnd_h_H:rnd_h_H + lq_patchsize * sf, rnd_w_H:rnd_w_H + lq_patchsize * sf, :]
+    return lq, hq
+
+
+def degradation_bsrgan(img, sf=4, lq_patchsize=72, isp_model=None):
+    """
+    This is the degradation model of BSRGAN from the paper
+    "Designing a Practical Degradation Model for Deep Blind Image Super-Resolution"
+    ----------
+    img: HXWXC, [0, 1], its size should be large than (lq_patchsizexsf)x(lq_patchsizexsf)
+    sf: scale factor
+    isp_model: camera ISP model
+    Returns
+    -------
+    img: low-quality patch, size: lq_patchsizeXlq_patchsizeXC, range: [0, 1]
+    hq: corresponding high-quality patch, size: (lq_patchsizexsf)X(lq_patchsizexsf)XC, range: [0, 1]
+    """
+    isp_prob, jpeg_prob, scale2_prob = 0.25, 0.9, 0.25
+    sf_ori = sf
+
+    h1, w1 = img.shape[:2]
+    img = img.copy()[:w1 - w1 % sf, :h1 - h1 % sf, ...]  # mod crop
+    h, w = img.shape[:2]
+
+    if h < lq_patchsize * sf or w < lq_patchsize * sf:
+        raise ValueError(f'img size ({h1}X{w1}) is too small!')
+
+    hq = img.copy()
+
+    if sf == 4 and random.random() < scale2_prob:  # downsample1
+        if np.random.rand() < 0.5:
+            img = cv2.resize(img, (int(1 / 2 * img.shape[1]), int(1 / 2 * img.shape[0])),
+                             interpolation=random.choice([1, 2, 3]))
+        else:
+            img = util.imresize_np(img, 1 / 2, True)
+        img = np.clip(img, 0.0, 1.0)
+        sf = 2
+
+    shuffle_order = random.sample(range(7), 7)
+    idx1, idx2 = shuffle_order.index(2), shuffle_order.index(3)
+    if idx1 > idx2:  # keep downsample3 last
+        shuffle_order[idx1], shuffle_order[idx2] = shuffle_order[idx2], shuffle_order[idx1]
+
+    for i in shuffle_order:
+
+        if i == 0:
+            img = add_blur(img, sf=sf)
+
+        elif i == 1:
+            img = add_blur(img, sf=sf)
+
+        elif i == 2:
+            a, b = img.shape[1], img.shape[0]
+            # downsample2
+            if random.random() < 0.75:
+                sf1 = random.uniform(1, 2 * sf)
+                img = cv2.resize(img, (int(1 / sf1 * img.shape[1]), int(1 / sf1 * img.shape[0])),
+                                 interpolation=random.choice([1, 2, 3]))
+            else:
+                k = fspecial('gaussian', 25, random.uniform(0.1, 0.6 * sf))
+                k_shifted = shift_pixel(k, sf)
+                k_shifted = k_shifted / k_shifted.sum()  # blur with shifted kernel
+                img = ndimage.filters.convolve(img, np.expand_dims(k_shifted, axis=2), mode='mirror')
+                img = img[0::sf, 0::sf, ...]  # nearest downsampling
+            img = np.clip(img, 0.0, 1.0)
+
+        elif i == 3:
+            # downsample3
+            img = cv2.resize(img, (int(1 / sf * a), int(1 / sf * b)), interpolation=random.choice([1, 2, 3]))
+            img = np.clip(img, 0.0, 1.0)
+
+        elif i == 4:
+            # add Gaussian noise
+            img = add_Gaussian_noise(img, noise_level1=2, noise_level2=8)
+
+        elif i == 5:
+            # add JPEG noise
+            if random.random() < jpeg_prob:
+                img = add_JPEG_noise(img)
+
+        elif i == 6:
+            # add processed camera sensor noise
+            if random.random() < isp_prob and isp_model is not None:
+                with torch.no_grad():
+                    img, hq = isp_model.forward(img.copy(), hq)
+
+    # add final JPEG compression noise
+    img = add_JPEG_noise(img)
+
+    # random crop
+    img, hq = random_crop(img, hq, sf_ori, lq_patchsize)
+
+    return img, hq
+
+
+# todo no isp_model?
+def degradation_bsrgan_variant(image, sf=4, isp_model=None):
+    """
+    This is the degradation model of BSRGAN from the paper
+    "Designing a Practical Degradation Model for Deep Blind Image Super-Resolution"
+    ----------
+    sf: scale factor
+    isp_model: camera ISP model
+    Returns
+    -------
+    img: low-quality patch, size: lq_patchsizeXlq_patchsizeXC, range: [0, 1]
+    hq: corresponding high-quality patch, size: (lq_patchsizexsf)X(lq_patchsizexsf)XC, range: [0, 1]
+    """
+    image = util.uint2single(image)
+    isp_prob, jpeg_prob, scale2_prob = 0.25, 0.9, 0.25
+    sf_ori = sf
+
+    h1, w1 = image.shape[:2]
+    image = image.copy()[:w1 - w1 % sf, :h1 - h1 % sf, ...]  # mod crop
+    h, w = image.shape[:2]
+
+    hq = image.copy()
+
+    if sf == 4 and random.random() < scale2_prob:  # downsample1
+        if np.random.rand() < 0.5:
+            image = cv2.resize(image, (int(1 / 2 * image.shape[1]), int(1 / 2 * image.shape[0])),
+                               interpolation=random.choice([1, 2, 3]))
+        else:
+            image = util.imresize_np(image, 1 / 2, True)
+        image = np.clip(image, 0.0, 1.0)
+        sf = 2
+
+    shuffle_order = random.sample(range(7), 7)
+    idx1, idx2 = shuffle_order.index(2), shuffle_order.index(3)
+    if idx1 > idx2:  # keep downsample3 last
+        shuffle_order[idx1], shuffle_order[idx2] = shuffle_order[idx2], shuffle_order[idx1]
+
+    for i in shuffle_order:
+
+        if i == 0:
+            image = add_blur(image, sf=sf)
+
+        # elif i == 1:
+        #     image = add_blur(image, sf=sf)
+
+        if i == 0:
+            pass
+
+        elif i == 2:
+            a, b = image.shape[1], image.shape[0]
+            # downsample2
+            if random.random() < 0.8:
+                sf1 = random.uniform(1, 2 * sf)
+                image = cv2.resize(image, (int(1 / sf1 * image.shape[1]), int(1 / sf1 * image.shape[0])),
+                                   interpolation=random.choice([1, 2, 3]))
+            else:
+                k = fspecial('gaussian', 25, random.uniform(0.1, 0.6 * sf))
+                k_shifted = shift_pixel(k, sf)
+                k_shifted = k_shifted / k_shifted.sum()  # blur with shifted kernel
+                image = ndimage.filters.convolve(image, np.expand_dims(k_shifted, axis=2), mode='mirror')
+                image = image[0::sf, 0::sf, ...]  # nearest downsampling
+
+            image = np.clip(image, 0.0, 1.0)
+
+        elif i == 3:
+            # downsample3
+            image = cv2.resize(image, (int(1 / sf * a), int(1 / sf * b)), interpolation=random.choice([1, 2, 3]))
+            image = np.clip(image, 0.0, 1.0)
+
+        elif i == 4:
+            # add Gaussian noise
+            image = add_Gaussian_noise(image, noise_level1=1, noise_level2=2)
+
+        elif i == 5:
+            # add JPEG noise
+            if random.random() < jpeg_prob:
+                image = add_JPEG_noise(image)
+        #
+        # elif i == 6:
+        #     # add processed camera sensor noise
+        #     if random.random() < isp_prob and isp_model is not None:
+        #         with torch.no_grad():
+        #             img, hq = isp_model.forward(img.copy(), hq)
+
+    # add final JPEG compression noise
+    image = add_JPEG_noise(image)
+    image = util.single2uint(image)
+    example = {"image": image}
+    return example
+
+
+
+
+if __name__ == '__main__':
+    print("hey")
+    img = util.imread_uint('utils/test.png', 3)
+    img = img[:448, :448]
+    h = img.shape[0] // 4
+    print("resizing to", h)
+    sf = 4
+    deg_fn = partial(degradation_bsrgan_variant, sf=sf)
+    for i in range(20):
+        print(i)
+        img_hq = img
+        img_lq = deg_fn(img)["image"]
+        img_hq, img_lq = util.uint2single(img_hq), util.uint2single(img_lq)
+        print(img_lq)
+        img_lq_bicubic = albumentations.SmallestMaxSize(max_size=h, interpolation=cv2.INTER_CUBIC)(image=img_hq)["image"]
+        print(img_lq.shape)
+        print("bicubic", img_lq_bicubic.shape)
+        print(img_hq.shape)
+        lq_nearest = cv2.resize(util.single2uint(img_lq), (int(sf * img_lq.shape[1]), int(sf * img_lq.shape[0])),
+                                interpolation=0)
+        lq_bicubic_nearest = cv2.resize(util.single2uint(img_lq_bicubic),
+                                        (int(sf * img_lq.shape[1]), int(sf * img_lq.shape[0])),
+                                        interpolation=0)
+        img_concat = np.concatenate([lq_bicubic_nearest, lq_nearest, util.single2uint(img_hq)], axis=1)
+        util.imsave(img_concat, str(i) + '.png')

ldm/modules/image_degradation/utils_image.py

@@ -0,0 +1,916 @@
+import os
+import math
+import random
+import numpy as np
+import torch
+import cv2
+from torchvision.utils import make_grid
+from datetime import datetime
+#import matplotlib.pyplot as plt   # TODO: check with Dominik, also bsrgan.py vs bsrgan_light.py
+
+
+os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"
+
+
+'''
+# --------------------------------------------
+# Kai Zhang (github: https://github.com/cszn)
+# 03/Mar/2019
+# --------------------------------------------
+# https://github.com/twhui/SRGAN-pyTorch
+# https://github.com/xinntao/BasicSR
+# --------------------------------------------
+'''
+
+
+IMG_EXTENSIONS = ['.jpg', '.JPG', '.jpeg', '.JPEG', '.png', '.PNG', '.ppm', '.PPM', '.bmp', '.BMP', '.tif']
+
+
+def is_image_file(filename):
+    return any(filename.endswith(extension) for extension in IMG_EXTENSIONS)
+
+
+def get_timestamp():
+    return datetime.now().strftime('%y%m%d-%H%M%S')
+
+
+def imshow(x, title=None, cbar=False, figsize=None):
+    plt.figure(figsize=figsize)
+    plt.imshow(np.squeeze(x), interpolation='nearest', cmap='gray')
+    if title:
+        plt.title(title)
+    if cbar:
+        plt.colorbar()
+    plt.show()
+
+
+def surf(Z, cmap='rainbow', figsize=None):
+    plt.figure(figsize=figsize)
+    ax3 = plt.axes(projection='3d')
+
+    w, h = Z.shape[:2]
+    xx = np.arange(0,w,1)
+    yy = np.arange(0,h,1)
+    X, Y = np.meshgrid(xx, yy)
+    ax3.plot_surface(X,Y,Z,cmap=cmap)
+    #ax3.contour(X,Y,Z, zdim='z',offset=-2，cmap=cmap)
+    plt.show()
+
+
+'''
+# --------------------------------------------
+# get image pathes
+# --------------------------------------------
+'''
+
+
+def get_image_paths(dataroot):
+    paths = None  # return None if dataroot is None
+    if dataroot is not None:
+        paths = sorted(_get_paths_from_images(dataroot))
+    return paths
+
+
+def _get_paths_from_images(path):
+    assert os.path.isdir(path), '{:s} is not a valid directory'.format(path)
+    images = []
+    for dirpath, _, fnames in sorted(os.walk(path)):
+        for fname in sorted(fnames):
+            if is_image_file(fname):
+                img_path = os.path.join(dirpath, fname)
+                images.append(img_path)
+    assert images, '{:s} has no valid image file'.format(path)
+    return images
+
+
+'''
+# --------------------------------------------
+# split large images into small images 
+# --------------------------------------------
+'''
+
+
+def patches_from_image(img, p_size=512, p_overlap=64, p_max=800):
+    w, h = img.shape[:2]
+    patches = []
+    if w > p_max and h > p_max:
+        w1 = list(np.arange(0, w-p_size, p_size-p_overlap, dtype=np.int))
+        h1 = list(np.arange(0, h-p_size, p_size-p_overlap, dtype=np.int))
+        w1.append(w-p_size)
+        h1.append(h-p_size)
+#        print(w1)
+#        print(h1)
+        for i in w1:
+            for j in h1:
+                patches.append(img[i:i+p_size, j:j+p_size,:])
+    else:
+        patches.append(img)
+
+    return patches
+
+
+def imssave(imgs, img_path):
+    """
+    imgs: list, N images of size WxHxC
+    """
+    img_name, ext = os.path.splitext(os.path.basename(img_path))
+
+    for i, img in enumerate(imgs):
+        if img.ndim == 3:
+            img = img[:, :, [2, 1, 0]]
+        new_path = os.path.join(os.path.dirname(img_path), img_name+str('_s{:04d}'.format(i))+'.png')
+        cv2.imwrite(new_path, img)
+
+
+def split_imageset(original_dataroot, taget_dataroot, n_channels=3, p_size=800, p_overlap=96, p_max=1000):
+    """
+    split the large images from original_dataroot into small overlapped images with size (p_size)x(p_size),
+    and save them into taget_dataroot; only the images with larger size than (p_max)x(p_max)
+    will be splitted.
+    Args:
+        original_dataroot:
+        taget_dataroot:
+        p_size: size of small images
+        p_overlap: patch size in training is a good choice
+        p_max: images with smaller size than (p_max)x(p_max) keep unchanged.
+    """
+    paths = get_image_paths(original_dataroot)
+    for img_path in paths:
+        # img_name, ext = os.path.splitext(os.path.basename(img_path))
+        img = imread_uint(img_path, n_channels=n_channels)
+        patches = patches_from_image(img, p_size, p_overlap, p_max)
+        imssave(patches, os.path.join(taget_dataroot,os.path.basename(img_path)))
+        #if original_dataroot == taget_dataroot:
+        #del img_path
+
+'''
+# --------------------------------------------
+# makedir
+# --------------------------------------------
+'''
+
+
+def mkdir(path):
+    if not os.path.exists(path):
+        os.makedirs(path)
+
+
+def mkdirs(paths):
+    if isinstance(paths, str):
+        mkdir(paths)
+    else:
+        for path in paths:
+            mkdir(path)
+
+
+def mkdir_and_rename(path):
+    if os.path.exists(path):
+        new_name = path + '_archived_' + get_timestamp()
+        print('Path already exists. Rename it to [{:s}]'.format(new_name))
+        os.rename(path, new_name)
+    os.makedirs(path)
+
+
+'''
+# --------------------------------------------
+# read image from path
+# opencv is fast, but read BGR numpy image
+# --------------------------------------------
+'''
+
+
+# --------------------------------------------
+# get uint8 image of size HxWxn_channles (RGB)
+# --------------------------------------------
+def imread_uint(path, n_channels=3):
+    #  input: path
+    # output: HxWx3(RGB or GGG), or HxWx1 (G)
+    if n_channels == 1:
+        img = cv2.imread(path, 0)  # cv2.IMREAD_GRAYSCALE
+        img = np.expand_dims(img, axis=2)  # HxWx1
+    elif n_channels == 3:
+        img = cv2.imread(path, cv2.IMREAD_UNCHANGED)  # BGR or G
+        if img.ndim == 2:
+            img = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)  # GGG
+        else:
+            img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)  # RGB
+    return img
+
+
+# --------------------------------------------
+# matlab's imwrite
+# --------------------------------------------
+def imsave(img, img_path):
+    img = np.squeeze(img)
+    if img.ndim == 3:
+        img = img[:, :, [2, 1, 0]]
+    cv2.imwrite(img_path, img)
+
+def imwrite(img, img_path):
+    img = np.squeeze(img)
+    if img.ndim == 3:
+        img = img[:, :, [2, 1, 0]]
+    cv2.imwrite(img_path, img)
+
+
+
+# --------------------------------------------
+# get single image of size HxWxn_channles (BGR)
+# --------------------------------------------
+def read_img(path):
+    # read image by cv2
+    # return: Numpy float32, HWC, BGR, [0,1]
+    img = cv2.imread(path, cv2.IMREAD_UNCHANGED)  # cv2.IMREAD_GRAYSCALE
+    img = img.astype(np.float32) / 255.
+    if img.ndim == 2:
+        img = np.expand_dims(img, axis=2)
+    # some images have 4 channels
+    if img.shape[2] > 3:
+        img = img[:, :, :3]
+    return img
+
+
+'''
+# --------------------------------------------
+# image format conversion
+# --------------------------------------------
+# numpy(single) <--->  numpy(unit)
+# numpy(single) <--->  tensor
+# numpy(unit)   <--->  tensor
+# --------------------------------------------
+'''
+
+
+# --------------------------------------------
+# numpy(single) [0, 1] <--->  numpy(unit)
+# --------------------------------------------
+
+
+def uint2single(img):
+
+    return np.float32(img/255.)
+
+
+def single2uint(img):
+
+    return np.uint8((img.clip(0, 1)*255.).round())
+
+
+def uint162single(img):
+
+    return np.float32(img/65535.)
+
+
+def single2uint16(img):
+
+    return np.uint16((img.clip(0, 1)*65535.).round())
+
+
+# --------------------------------------------
+# numpy(unit) (HxWxC or HxW) <--->  tensor
+# --------------------------------------------
+
+
+# convert uint to 4-dimensional torch tensor
+def uint2tensor4(img):
+    if img.ndim == 2:
+        img = np.expand_dims(img, axis=2)
+    return torch.from_numpy(np.ascontiguousarray(img)).permute(2, 0, 1).float().div(255.).unsqueeze(0)
+
+
+# convert uint to 3-dimensional torch tensor
+def uint2tensor3(img):
+    if img.ndim == 2:
+        img = np.expand_dims(img, axis=2)
+    return torch.from_numpy(np.ascontiguousarray(img)).permute(2, 0, 1).float().div(255.)
+
+
+# convert 2/3/4-dimensional torch tensor to uint
+def tensor2uint(img):
+    img = img.data.squeeze().float().clamp_(0, 1).cpu().numpy()
+    if img.ndim == 3:
+        img = np.transpose(img, (1, 2, 0))
+    return np.uint8((img*255.0).round())
+
+
+# --------------------------------------------
+# numpy(single) (HxWxC) <--->  tensor
+# --------------------------------------------
+
+
+# convert single (HxWxC) to 3-dimensional torch tensor
+def single2tensor3(img):
+    return torch.from_numpy(np.ascontiguousarray(img)).permute(2, 0, 1).float()
+
+
+# convert single (HxWxC) to 4-dimensional torch tensor
+def single2tensor4(img):
+    return torch.from_numpy(np.ascontiguousarray(img)).permute(2, 0, 1).float().unsqueeze(0)
+
+
+# convert torch tensor to single
+def tensor2single(img):
+    img = img.data.squeeze().float().cpu().numpy()
+    if img.ndim == 3:
+        img = np.transpose(img, (1, 2, 0))
+
+    return img
+
+# convert torch tensor to single
+def tensor2single3(img):
+    img = img.data.squeeze().float().cpu().numpy()
+    if img.ndim == 3:
+        img = np.transpose(img, (1, 2, 0))
+    elif img.ndim == 2:
+        img = np.expand_dims(img, axis=2)
+    return img
+
+
+def single2tensor5(img):
+    return torch.from_numpy(np.ascontiguousarray(img)).permute(2, 0, 1, 3).float().unsqueeze(0)
+
+
+def single32tensor5(img):
+    return torch.from_numpy(np.ascontiguousarray(img)).float().unsqueeze(0).unsqueeze(0)
+
+
+def single42tensor4(img):
+    return torch.from_numpy(np.ascontiguousarray(img)).permute(2, 0, 1, 3).float()
+
+
+# from skimage.io import imread, imsave
+def tensor2img(tensor, out_type=np.uint8, min_max=(0, 1)):
+    '''
+    Converts a torch Tensor into an image Numpy array of BGR channel order
+    Input: 4D(B,(3/1),H,W), 3D(C,H,W), or 2D(H,W), any range, RGB channel order
+    Output: 3D(H,W,C) or 2D(H,W), [0,255], np.uint8 (default)
+    '''
+    tensor = tensor.squeeze().float().cpu().clamp_(*min_max)  # squeeze first, then clamp
+    tensor = (tensor - min_max[0]) / (min_max[1] - min_max[0])  # to range [0,1]
+    n_dim = tensor.dim()
+    if n_dim == 4:
+        n_img = len(tensor)
+        img_np = make_grid(tensor, nrow=int(math.sqrt(n_img)), normalize=False).numpy()
+        img_np = np.transpose(img_np[[2, 1, 0], :, :], (1, 2, 0))  # HWC, BGR
+    elif n_dim == 3:
+        img_np = tensor.numpy()
+        img_np = np.transpose(img_np[[2, 1, 0], :, :], (1, 2, 0))  # HWC, BGR
+    elif n_dim == 2:
+        img_np = tensor.numpy()
+    else:
+        raise TypeError(
+            'Only support 4D, 3D and 2D tensor. But received with dimension: {:d}'.format(n_dim))
+    if out_type == np.uint8:
+        img_np = (img_np * 255.0).round()
+        # Important. Unlike matlab, numpy.unit8() WILL NOT round by default.
+    return img_np.astype(out_type)
+
+
+'''
+# --------------------------------------------
+# Augmentation, flipe and/or rotate
+# --------------------------------------------
+# The following two are enough.
+# (1) augmet_img: numpy image of WxHxC or WxH
+# (2) augment_img_tensor4: tensor image 1xCxWxH
+# --------------------------------------------
+'''
+
+
+def augment_img(img, mode=0):
+    '''Kai Zhang (github: https://github.com/cszn)
+    '''
+    if mode == 0:
+        return img
+    elif mode == 1:
+        return np.flipud(np.rot90(img))
+    elif mode == 2:
+        return np.flipud(img)
+    elif mode == 3:
+        return np.rot90(img, k=3)
+    elif mode == 4:
+        return np.flipud(np.rot90(img, k=2))
+    elif mode == 5:
+        return np.rot90(img)
+    elif mode == 6:
+        return np.rot90(img, k=2)
+    elif mode == 7:
+        return np.flipud(np.rot90(img, k=3))
+
+
+def augment_img_tensor4(img, mode=0):
+    '''Kai Zhang (github: https://github.com/cszn)
+    '''
+    if mode == 0:
+        return img
+    elif mode == 1:
+        return img.rot90(1, [2, 3]).flip([2])
+    elif mode == 2:
+        return img.flip([2])
+    elif mode == 3:
+        return img.rot90(3, [2, 3])
+    elif mode == 4:
+        return img.rot90(2, [2, 3]).flip([2])
+    elif mode == 5:
+        return img.rot90(1, [2, 3])
+    elif mode == 6:
+        return img.rot90(2, [2, 3])
+    elif mode == 7:
+        return img.rot90(3, [2, 3]).flip([2])
+
+
+def augment_img_tensor(img, mode=0):
+    '''Kai Zhang (github: https://github.com/cszn)
+    '''
+    img_size = img.size()
+    img_np = img.data.cpu().numpy()
+    if len(img_size) == 3:
+        img_np = np.transpose(img_np, (1, 2, 0))
+    elif len(img_size) == 4:
+        img_np = np.transpose(img_np, (2, 3, 1, 0))
+    img_np = augment_img(img_np, mode=mode)
+    img_tensor = torch.from_numpy(np.ascontiguousarray(img_np))
+    if len(img_size) == 3:
+        img_tensor = img_tensor.permute(2, 0, 1)
+    elif len(img_size) == 4:
+        img_tensor = img_tensor.permute(3, 2, 0, 1)
+
+    return img_tensor.type_as(img)
+
+
+def augment_img_np3(img, mode=0):
+    if mode == 0:
+        return img
+    elif mode == 1:
+        return img.transpose(1, 0, 2)
+    elif mode == 2:
+        return img[::-1, :, :]
+    elif mode == 3:
+        img = img[::-1, :, :]
+        img = img.transpose(1, 0, 2)
+        return img
+    elif mode == 4:
+        return img[:, ::-1, :]
+    elif mode == 5:
+        img = img[:, ::-1, :]
+        img = img.transpose(1, 0, 2)
+        return img
+    elif mode == 6:
+        img = img[:, ::-1, :]
+        img = img[::-1, :, :]
+        return img
+    elif mode == 7:
+        img = img[:, ::-1, :]
+        img = img[::-1, :, :]
+        img = img.transpose(1, 0, 2)
+        return img
+
+
+def augment_imgs(img_list, hflip=True, rot=True):
+    # horizontal flip OR rotate
+    hflip = hflip and random.random() < 0.5
+    vflip = rot and random.random() < 0.5
+    rot90 = rot and random.random() < 0.5
+
+    def _augment(img):
+        if hflip:
+            img = img[:, ::-1, :]
+        if vflip:
+            img = img[::-1, :, :]
+        if rot90:
+            img = img.transpose(1, 0, 2)
+        return img
+
+    return [_augment(img) for img in img_list]
+
+
+'''
+# --------------------------------------------
+# modcrop and shave
+# --------------------------------------------
+'''
+
+
+def modcrop(img_in, scale):
+    # img_in: Numpy, HWC or HW
+    img = np.copy(img_in)
+    if img.ndim == 2:
+        H, W = img.shape
+        H_r, W_r = H % scale, W % scale
+        img = img[:H - H_r, :W - W_r]
+    elif img.ndim == 3:
+        H, W, C = img.shape
+        H_r, W_r = H % scale, W % scale
+        img = img[:H - H_r, :W - W_r, :]
+    else:
+        raise ValueError('Wrong img ndim: [{:d}].'.format(img.ndim))
+    return img
+
+
+def shave(img_in, border=0):
+    # img_in: Numpy, HWC or HW
+    img = np.copy(img_in)
+    h, w = img.shape[:2]
+    img = img[border:h-border, border:w-border]
+    return img
+
+
+'''
+# --------------------------------------------
+# image processing process on numpy image
+# channel_convert(in_c, tar_type, img_list):
+# rgb2ycbcr(img, only_y=True):
+# bgr2ycbcr(img, only_y=True):
+# ycbcr2rgb(img):
+# --------------------------------------------
+'''
+
+
+def rgb2ycbcr(img, only_y=True):
+    '''same as matlab rgb2ycbcr
+    only_y: only return Y channel
+    Input:
+        uint8, [0, 255]
+        float, [0, 1]
+    '''
+    in_img_type = img.dtype
+    img.astype(np.float32)
+    if in_img_type != np.uint8:
+        img *= 255.
+    # convert
+    if only_y:
+        rlt = np.dot(img, [65.481, 128.553, 24.966]) / 255.0 + 16.0
+    else:
+        rlt = np.matmul(img, [[65.481, -37.797, 112.0], [128.553, -74.203, -93.786],
+                              [24.966, 112.0, -18.214]]) / 255.0 + [16, 128, 128]
+    if in_img_type == np.uint8:
+        rlt = rlt.round()
+    else:
+        rlt /= 255.
+    return rlt.astype(in_img_type)
+
+
+def ycbcr2rgb(img):
+    '''same as matlab ycbcr2rgb
+    Input:
+        uint8, [0, 255]
+        float, [0, 1]
+    '''
+    in_img_type = img.dtype
+    img.astype(np.float32)
+    if in_img_type != np.uint8:
+        img *= 255.
+    # convert
+    rlt = np.matmul(img, [[0.00456621, 0.00456621, 0.00456621], [0, -0.00153632, 0.00791071],
+                          [0.00625893, -0.00318811, 0]]) * 255.0 + [-222.921, 135.576, -276.836]
+    if in_img_type == np.uint8:
+        rlt = rlt.round()
+    else:
+        rlt /= 255.
+    return rlt.astype(in_img_type)
+
+
+def bgr2ycbcr(img, only_y=True):
+    '''bgr version of rgb2ycbcr
+    only_y: only return Y channel
+    Input:
+        uint8, [0, 255]
+        float, [0, 1]
+    '''
+    in_img_type = img.dtype
+    img.astype(np.float32)
+    if in_img_type != np.uint8:
+        img *= 255.
+    # convert
+    if only_y:
+        rlt = np.dot(img, [24.966, 128.553, 65.481]) / 255.0 + 16.0
+    else:
+        rlt = np.matmul(img, [[24.966, 112.0, -18.214], [128.553, -74.203, -93.786],
+                              [65.481, -37.797, 112.0]]) / 255.0 + [16, 128, 128]
+    if in_img_type == np.uint8:
+        rlt = rlt.round()
+    else:
+        rlt /= 255.
+    return rlt.astype(in_img_type)
+
+
+def channel_convert(in_c, tar_type, img_list):
+    # conversion among BGR, gray and y
+    if in_c == 3 and tar_type == 'gray':  # BGR to gray
+        gray_list = [cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) for img in img_list]
+        return [np.expand_dims(img, axis=2) for img in gray_list]
+    elif in_c == 3 and tar_type == 'y':  # BGR to y
+        y_list = [bgr2ycbcr(img, only_y=True) for img in img_list]
+        return [np.expand_dims(img, axis=2) for img in y_list]
+    elif in_c == 1 and tar_type == 'RGB':  # gray/y to BGR
+        return [cv2.cvtColor(img, cv2.COLOR_GRAY2BGR) for img in img_list]
+    else:
+        return img_list
+
+
+'''
+# --------------------------------------------
+# metric, PSNR and SSIM
+# --------------------------------------------
+'''
+
+
+# --------------------------------------------
+# PSNR
+# --------------------------------------------
+def calculate_psnr(img1, img2, border=0):
+    # img1 and img2 have range [0, 255]
+    #img1 = img1.squeeze()
+    #img2 = img2.squeeze()
+    if not img1.shape == img2.shape:
+        raise ValueError('Input images must have the same dimensions.')
+    h, w = img1.shape[:2]
+    img1 = img1[border:h-border, border:w-border]
+    img2 = img2[border:h-border, border:w-border]
+
+    img1 = img1.astype(np.float64)
+    img2 = img2.astype(np.float64)
+    mse = np.mean((img1 - img2)**2)
+    if mse == 0:
+        return float('inf')
+    return 20 * math.log10(255.0 / math.sqrt(mse))
+
+
+# --------------------------------------------
+# SSIM
+# --------------------------------------------
+def calculate_ssim(img1, img2, border=0):
+    '''calculate SSIM
+    the same outputs as MATLAB's
+    img1, img2: [0, 255]
+    '''
+    #img1 = img1.squeeze()
+    #img2 = img2.squeeze()
+    if not img1.shape == img2.shape:
+        raise ValueError('Input images must have the same dimensions.')
+    h, w = img1.shape[:2]
+    img1 = img1[border:h-border, border:w-border]
+    img2 = img2[border:h-border, border:w-border]
+
+    if img1.ndim == 2:
+        return ssim(img1, img2)
+    elif img1.ndim == 3:
+        if img1.shape[2] == 3:
+            ssims = []
+            for i in range(3):
+                ssims.append(ssim(img1[:,:,i], img2[:,:,i]))
+            return np.array(ssims).mean()
+        elif img1.shape[2] == 1:
+            return ssim(np.squeeze(img1), np.squeeze(img2))
+    else:
+        raise ValueError('Wrong input image dimensions.')
+
+
+def ssim(img1, img2):
+    C1 = (0.01 * 255)**2
+    C2 = (0.03 * 255)**2
+
+    img1 = img1.astype(np.float64)
+    img2 = img2.astype(np.float64)
+    kernel = cv2.getGaussianKernel(11, 1.5)
+    window = np.outer(kernel, kernel.transpose())
+
+    mu1 = cv2.filter2D(img1, -1, window)[5:-5, 5:-5]  # valid
+    mu2 = cv2.filter2D(img2, -1, window)[5:-5, 5:-5]
+    mu1_sq = mu1**2
+    mu2_sq = mu2**2
+    mu1_mu2 = mu1 * mu2
+    sigma1_sq = cv2.filter2D(img1**2, -1, window)[5:-5, 5:-5] - mu1_sq
+    sigma2_sq = cv2.filter2D(img2**2, -1, window)[5:-5, 5:-5] - mu2_sq
+    sigma12 = cv2.filter2D(img1 * img2, -1, window)[5:-5, 5:-5] - mu1_mu2
+
+    ssim_map = ((2 * mu1_mu2 + C1) * (2 * sigma12 + C2)) / ((mu1_sq + mu2_sq + C1) *
+                                                            (sigma1_sq + sigma2_sq + C2))
+    return ssim_map.mean()
+
+
+'''
+# --------------------------------------------
+# matlab's bicubic imresize (numpy and torch) [0, 1]
+# --------------------------------------------
+'''
+
+
+# matlab 'imresize' function, now only support 'bicubic'
+def cubic(x):
+    absx = torch.abs(x)
+    absx2 = absx**2
+    absx3 = absx**3
+    return (1.5*absx3 - 2.5*absx2 + 1) * ((absx <= 1).type_as(absx)) + \
+        (-0.5*absx3 + 2.5*absx2 - 4*absx + 2) * (((absx > 1)*(absx <= 2)).type_as(absx))
+
+
+def calculate_weights_indices(in_length, out_length, scale, kernel, kernel_width, antialiasing):
+    if (scale < 1) and (antialiasing):
+        # Use a modified kernel to simultaneously interpolate and antialias- larger kernel width
+        kernel_width = kernel_width / scale
+
+    # Output-space coordinates
+    x = torch.linspace(1, out_length, out_length)
+
+    # Input-space coordinates. Calculate the inverse mapping such that 0.5
+    # in output space maps to 0.5 in input space, and 0.5+scale in output
+    # space maps to 1.5 in input space.
+    u = x / scale + 0.5 * (1 - 1 / scale)
+
+    # What is the left-most pixel that can be involved in the computation?
+    left = torch.floor(u - kernel_width / 2)
+
+    # What is the maximum number of pixels that can be involved in the
+    # computation?  Note: it's OK to use an extra pixel here; if the
+    # corresponding weights are all zero, it will be eliminated at the end
+    # of this function.
+    P = math.ceil(kernel_width) + 2
+
+    # The indices of the input pixels involved in computing the k-th output
+    # pixel are in row k of the indices matrix.
+    indices = left.view(out_length, 1).expand(out_length, P) + torch.linspace(0, P - 1, P).view(
+        1, P).expand(out_length, P)
+
+    # The weights used to compute the k-th output pixel are in row k of the
+    # weights matrix.
+    distance_to_center = u.view(out_length, 1).expand(out_length, P) - indices
+    # apply cubic kernel
+    if (scale < 1) and (antialiasing):
+        weights = scale * cubic(distance_to_center * scale)
+    else:
+        weights = cubic(distance_to_center)
+    # Normalize the weights matrix so that each row sums to 1.
+    weights_sum = torch.sum(weights, 1).view(out_length, 1)
+    weights = weights / weights_sum.expand(out_length, P)
+
+    # If a column in weights is all zero, get rid of it. only consider the first and last column.
+    weights_zero_tmp = torch.sum((weights == 0), 0)
+    if not math.isclose(weights_zero_tmp[0], 0, rel_tol=1e-6):
+        indices = indices.narrow(1, 1, P - 2)
+        weights = weights.narrow(1, 1, P - 2)
+    if not math.isclose(weights_zero_tmp[-1], 0, rel_tol=1e-6):
+        indices = indices.narrow(1, 0, P - 2)
+        weights = weights.narrow(1, 0, P - 2)
+    weights = weights.contiguous()
+    indices = indices.contiguous()
+    sym_len_s = -indices.min() + 1
+    sym_len_e = indices.max() - in_length
+    indices = indices + sym_len_s - 1
+    return weights, indices, int(sym_len_s), int(sym_len_e)
+
+
+# --------------------------------------------
+# imresize for tensor image [0, 1]
+# --------------------------------------------
+def imresize(img, scale, antialiasing=True):
+    # Now the scale should be the same for H and W
+    # input: img: pytorch tensor, CHW or HW [0,1]
+    # output: CHW or HW [0,1] w/o round
+    need_squeeze = True if img.dim() == 2 else False
+    if need_squeeze:
+        img.unsqueeze_(0)
+    in_C, in_H, in_W = img.size()
+    out_C, out_H, out_W = in_C, math.ceil(in_H * scale), math.ceil(in_W * scale)
+    kernel_width = 4
+    kernel = 'cubic'
+
+    # Return the desired dimension order for performing the resize.  The
+    # strategy is to perform the resize first along the dimension with the
+    # smallest scale factor.
+    # Now we do not support this.
+
+    # get weights and indices
+    weights_H, indices_H, sym_len_Hs, sym_len_He = calculate_weights_indices(
+        in_H, out_H, scale, kernel, kernel_width, antialiasing)
+    weights_W, indices_W, sym_len_Ws, sym_len_We = calculate_weights_indices(
+        in_W, out_W, scale, kernel, kernel_width, antialiasing)
+    # process H dimension
+    # symmetric copying
+    img_aug = torch.FloatTensor(in_C, in_H + sym_len_Hs + sym_len_He, in_W)
+    img_aug.narrow(1, sym_len_Hs, in_H).copy_(img)
+
+    sym_patch = img[:, :sym_len_Hs, :]
+    inv_idx = torch.arange(sym_patch.size(1) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(1, inv_idx)
+    img_aug.narrow(1, 0, sym_len_Hs).copy_(sym_patch_inv)
+
+    sym_patch = img[:, -sym_len_He:, :]
+    inv_idx = torch.arange(sym_patch.size(1) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(1, inv_idx)
+    img_aug.narrow(1, sym_len_Hs + in_H, sym_len_He).copy_(sym_patch_inv)
+
+    out_1 = torch.FloatTensor(in_C, out_H, in_W)
+    kernel_width = weights_H.size(1)
+    for i in range(out_H):
+        idx = int(indices_H[i][0])
+        for j in range(out_C):
+            out_1[j, i, :] = img_aug[j, idx:idx + kernel_width, :].transpose(0, 1).mv(weights_H[i])
+
+    # process W dimension
+    # symmetric copying
+    out_1_aug = torch.FloatTensor(in_C, out_H, in_W + sym_len_Ws + sym_len_We)
+    out_1_aug.narrow(2, sym_len_Ws, in_W).copy_(out_1)
+
+    sym_patch = out_1[:, :, :sym_len_Ws]
+    inv_idx = torch.arange(sym_patch.size(2) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(2, inv_idx)
+    out_1_aug.narrow(2, 0, sym_len_Ws).copy_(sym_patch_inv)
+
+    sym_patch = out_1[:, :, -sym_len_We:]
+    inv_idx = torch.arange(sym_patch.size(2) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(2, inv_idx)
+    out_1_aug.narrow(2, sym_len_Ws + in_W, sym_len_We).copy_(sym_patch_inv)
+
+    out_2 = torch.FloatTensor(in_C, out_H, out_W)
+    kernel_width = weights_W.size(1)
+    for i in range(out_W):
+        idx = int(indices_W[i][0])
+        for j in range(out_C):
+            out_2[j, :, i] = out_1_aug[j, :, idx:idx + kernel_width].mv(weights_W[i])
+    if need_squeeze:
+        out_2.squeeze_()
+    return out_2
+
+
+# --------------------------------------------
+# imresize for numpy image [0, 1]
+# --------------------------------------------
+def imresize_np(img, scale, antialiasing=True):
+    # Now the scale should be the same for H and W
+    # input: img: Numpy, HWC or HW [0,1]
+    # output: HWC or HW [0,1] w/o round
+    img = torch.from_numpy(img)
+    need_squeeze = True if img.dim() == 2 else False
+    if need_squeeze:
+        img.unsqueeze_(2)
+
+    in_H, in_W, in_C = img.size()
+    out_C, out_H, out_W = in_C, math.ceil(in_H * scale), math.ceil(in_W * scale)
+    kernel_width = 4
+    kernel = 'cubic'
+
+    # Return the desired dimension order for performing the resize.  The
+    # strategy is to perform the resize first along the dimension with the
+    # smallest scale factor.
+    # Now we do not support this.
+
+    # get weights and indices
+    weights_H, indices_H, sym_len_Hs, sym_len_He = calculate_weights_indices(
+        in_H, out_H, scale, kernel, kernel_width, antialiasing)
+    weights_W, indices_W, sym_len_Ws, sym_len_We = calculate_weights_indices(
+        in_W, out_W, scale, kernel, kernel_width, antialiasing)
+    # process H dimension
+    # symmetric copying
+    img_aug = torch.FloatTensor(in_H + sym_len_Hs + sym_len_He, in_W, in_C)
+    img_aug.narrow(0, sym_len_Hs, in_H).copy_(img)
+
+    sym_patch = img[:sym_len_Hs, :, :]
+    inv_idx = torch.arange(sym_patch.size(0) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(0, inv_idx)
+    img_aug.narrow(0, 0, sym_len_Hs).copy_(sym_patch_inv)
+
+    sym_patch = img[-sym_len_He:, :, :]
+    inv_idx = torch.arange(sym_patch.size(0) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(0, inv_idx)
+    img_aug.narrow(0, sym_len_Hs + in_H, sym_len_He).copy_(sym_patch_inv)
+
+    out_1 = torch.FloatTensor(out_H, in_W, in_C)
+    kernel_width = weights_H.size(1)
+    for i in range(out_H):
+        idx = int(indices_H[i][0])
+        for j in range(out_C):
+            out_1[i, :, j] = img_aug[idx:idx + kernel_width, :, j].transpose(0, 1).mv(weights_H[i])
+
+    # process W dimension
+    # symmetric copying
+    out_1_aug = torch.FloatTensor(out_H, in_W + sym_len_Ws + sym_len_We, in_C)
+    out_1_aug.narrow(1, sym_len_Ws, in_W).copy_(out_1)
+
+    sym_patch = out_1[:, :sym_len_Ws, :]
+    inv_idx = torch.arange(sym_patch.size(1) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(1, inv_idx)
+    out_1_aug.narrow(1, 0, sym_len_Ws).copy_(sym_patch_inv)
+
+    sym_patch = out_1[:, -sym_len_We:, :]
+    inv_idx = torch.arange(sym_patch.size(1) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(1, inv_idx)
+    out_1_aug.narrow(1, sym_len_Ws + in_W, sym_len_We).copy_(sym_patch_inv)
+
+    out_2 = torch.FloatTensor(out_H, out_W, in_C)
+    kernel_width = weights_W.size(1)
+    for i in range(out_W):
+        idx = int(indices_W[i][0])
+        for j in range(out_C):
+            out_2[:, i, j] = out_1_aug[:, idx:idx + kernel_width, j].mv(weights_W[i])
+    if need_squeeze:
+        out_2.squeeze_()
+
+    return out_2.numpy()
+
+
+if __name__ == '__main__':
+    print('---')
+#    img = imread_uint('test.bmp', 3)
+#    img = uint2single(img)
+#    img_bicubic = imresize_np(img, 1/4)

ldm/modules/losses/__init__.py

@@ -0,0 +1 @@
+from ldm.modules.losses.contperceptual import LPIPSWithDiscriminator

ldm/modules/losses/contperceptual.py

@@ -0,0 +1,111 @@
+import torch
+import torch.nn as nn
+
+from taming.modules.losses.vqperceptual import *  # TODO: taming dependency yes/no?
+
+
+class LPIPSWithDiscriminator(nn.Module):
+    def __init__(self, disc_start, logvar_init=0.0, kl_weight=1.0, pixelloss_weight=1.0,
+                 disc_num_layers=3, disc_in_channels=3, disc_factor=1.0, disc_weight=1.0,
+                 perceptual_weight=1.0, use_actnorm=False, disc_conditional=False,
+                 disc_loss="hinge"):
+
+        super().__init__()
+        assert disc_loss in ["hinge", "vanilla"]
+        self.kl_weight = kl_weight
+        self.pixel_weight = pixelloss_weight
+        self.perceptual_loss = LPIPS().eval()
+        self.perceptual_weight = perceptual_weight
+        # output log variance
+        self.logvar = nn.Parameter(torch.ones(size=()) * logvar_init)
+
+        self.discriminator = NLayerDiscriminator(input_nc=disc_in_channels,
+                                                 n_layers=disc_num_layers,
+                                                 use_actnorm=use_actnorm
+                                                 ).apply(weights_init)
+        self.discriminator_iter_start = disc_start
+        self.disc_loss = hinge_d_loss if disc_loss == "hinge" else vanilla_d_loss
+        self.disc_factor = disc_factor
+        self.discriminator_weight = disc_weight
+        self.disc_conditional = disc_conditional
+
+    def calculate_adaptive_weight(self, nll_loss, g_loss, last_layer=None):
+        if last_layer is not None:
+            nll_grads = torch.autograd.grad(nll_loss, last_layer, retain_graph=True)[0]
+            g_grads = torch.autograd.grad(g_loss, last_layer, retain_graph=True)[0]
+        else:
+            nll_grads = torch.autograd.grad(nll_loss, self.last_layer[0], retain_graph=True)[0]
+            g_grads = torch.autograd.grad(g_loss, self.last_layer[0], retain_graph=True)[0]
+
+        d_weight = torch.norm(nll_grads) / (torch.norm(g_grads) + 1e-4)
+        d_weight = torch.clamp(d_weight, 0.0, 1e4).detach()
+        d_weight = d_weight * self.discriminator_weight
+        return d_weight
+
+    def forward(self, inputs, reconstructions, posteriors, optimizer_idx,
+                global_step, last_layer=None, cond=None, split="train",
+                weights=None):
+        rec_loss = torch.abs(inputs.contiguous() - reconstructions.contiguous())
+        if self.perceptual_weight > 0:
+            p_loss = self.perceptual_loss(inputs.contiguous(), reconstructions.contiguous())
+            rec_loss = rec_loss + self.perceptual_weight * p_loss
+
+        nll_loss = rec_loss / torch.exp(self.logvar) + self.logvar
+        weighted_nll_loss = nll_loss
+        if weights is not None:
+            weighted_nll_loss = weights*nll_loss
+        weighted_nll_loss = torch.sum(weighted_nll_loss) / weighted_nll_loss.shape[0]
+        nll_loss = torch.sum(nll_loss) / nll_loss.shape[0]
+        kl_loss = posteriors.kl()
+        kl_loss = torch.sum(kl_loss) / kl_loss.shape[0]
+
+        # now the GAN part
+        if optimizer_idx == 0:
+            # generator update
+            if cond is None:
+                assert not self.disc_conditional
+                logits_fake = self.discriminator(reconstructions.contiguous())
+            else:
+                assert self.disc_conditional
+                logits_fake = self.discriminator(torch.cat((reconstructions.contiguous(), cond), dim=1))
+            g_loss = -torch.mean(logits_fake)
+
+            if self.disc_factor > 0.0:
+                try:
+                    d_weight = self.calculate_adaptive_weight(nll_loss, g_loss, last_layer=last_layer)
+                except RuntimeError:
+                    assert not self.training
+                    d_weight = torch.tensor(0.0)
+            else:
+                d_weight = torch.tensor(0.0)
+
+            disc_factor = adopt_weight(self.disc_factor, global_step, threshold=self.discriminator_iter_start)
+            loss = weighted_nll_loss + self.kl_weight * kl_loss + d_weight * disc_factor * g_loss
+
+            log = {"{}/total_loss".format(split): loss.clone().detach().mean(), "{}/logvar".format(split): self.logvar.detach(),
+                   "{}/kl_loss".format(split): kl_loss.detach().mean(), "{}/nll_loss".format(split): nll_loss.detach().mean(),
+                   "{}/rec_loss".format(split): rec_loss.detach().mean(),
+                   "{}/d_weight".format(split): d_weight.detach(),
+                   "{}/disc_factor".format(split): torch.tensor(disc_factor),
+                   "{}/g_loss".format(split): g_loss.detach().mean(),
+                   }
+            return loss, log
+
+        if optimizer_idx == 1:
+            # second pass for discriminator update
+            if cond is None:
+                logits_real = self.discriminator(inputs.contiguous().detach())
+                logits_fake = self.discriminator(reconstructions.contiguous().detach())
+            else:
+                logits_real = self.discriminator(torch.cat((inputs.contiguous().detach(), cond), dim=1))
+                logits_fake = self.discriminator(torch.cat((reconstructions.contiguous().detach(), cond), dim=1))
+
+            disc_factor = adopt_weight(self.disc_factor, global_step, threshold=self.discriminator_iter_start)
+            d_loss = disc_factor * self.disc_loss(logits_real, logits_fake)
+
+            log = {"{}/disc_loss".format(split): d_loss.clone().detach().mean(),
+                   "{}/logits_real".format(split): logits_real.detach().mean(),
+                   "{}/logits_fake".format(split): logits_fake.detach().mean()
+                   }
+            return d_loss, log
+

ldm/modules/losses/vqperceptual.py

@@ -0,0 +1,167 @@
+import torch
+from torch import nn
+import torch.nn.functional as F
+from einops import repeat
+
+from taming.modules.discriminator.model import NLayerDiscriminator, weights_init
+from taming.modules.losses.lpips import LPIPS
+from taming.modules.losses.vqperceptual import hinge_d_loss, vanilla_d_loss
+
+
+def hinge_d_loss_with_exemplar_weights(logits_real, logits_fake, weights):
+    assert weights.shape[0] == logits_real.shape[0] == logits_fake.shape[0]
+    loss_real = torch.mean(F.relu(1. - logits_real), dim=[1,2,3])
+    loss_fake = torch.mean(F.relu(1. + logits_fake), dim=[1,2,3])
+    loss_real = (weights * loss_real).sum() / weights.sum()
+    loss_fake = (weights * loss_fake).sum() / weights.sum()
+    d_loss = 0.5 * (loss_real + loss_fake)
+    return d_loss
+
+def adopt_weight(weight, global_step, threshold=0, value=0.):
+    if global_step < threshold:
+        weight = value
+    return weight
+
+
+def measure_perplexity(predicted_indices, n_embed):
+    # src: https://github.com/karpathy/deep-vector-quantization/blob/main/model.py
+    # eval cluster perplexity. when perplexity == num_embeddings then all clusters are used exactly equally
+    encodings = F.one_hot(predicted_indices, n_embed).float().reshape(-1, n_embed)
+    avg_probs = encodings.mean(0)
+    perplexity = (-(avg_probs * torch.log(avg_probs + 1e-10)).sum()).exp()
+    cluster_use = torch.sum(avg_probs > 0)
+    return perplexity, cluster_use
+
+def l1(x, y):
+    return torch.abs(x-y)
+
+
+def l2(x, y):
+    return torch.pow((x-y), 2)
+
+
+class VQLPIPSWithDiscriminator(nn.Module):
+    def __init__(self, disc_start, codebook_weight=1.0, pixelloss_weight=1.0,
+                 disc_num_layers=3, disc_in_channels=3, disc_factor=1.0, disc_weight=1.0,
+                 perceptual_weight=1.0, use_actnorm=False, disc_conditional=False,
+                 disc_ndf=64, disc_loss="hinge", n_classes=None, perceptual_loss="lpips",
+                 pixel_loss="l1"):
+        super().__init__()
+        assert disc_loss in ["hinge", "vanilla"]
+        assert perceptual_loss in ["lpips", "clips", "dists"]
+        assert pixel_loss in ["l1", "l2"]
+        self.codebook_weight = codebook_weight
+        self.pixel_weight = pixelloss_weight
+        if perceptual_loss == "lpips":
+            print(f"{self.__class__.__name__}: Running with LPIPS.")
+            self.perceptual_loss = LPIPS().eval()
+        else:
+            raise ValueError(f"Unknown perceptual loss: >> {perceptual_loss} <<")
+        self.perceptual_weight = perceptual_weight
+
+        if pixel_loss == "l1":
+            self.pixel_loss = l1
+        else:
+            self.pixel_loss = l2
+
+        self.discriminator = NLayerDiscriminator(input_nc=disc_in_channels,
+                                                 n_layers=disc_num_layers,
+                                                 use_actnorm=use_actnorm,
+                                                 ndf=disc_ndf
+                                                 ).apply(weights_init)
+        self.discriminator_iter_start = disc_start
+        if disc_loss == "hinge":
+            self.disc_loss = hinge_d_loss
+        elif disc_loss == "vanilla":
+            self.disc_loss = vanilla_d_loss
+        else:
+            raise ValueError(f"Unknown GAN loss '{disc_loss}'.")
+        print(f"VQLPIPSWithDiscriminator running with {disc_loss} loss.")
+        self.disc_factor = disc_factor
+        self.discriminator_weight = disc_weight
+        self.disc_conditional = disc_conditional
+        self.n_classes = n_classes
+
+    def calculate_adaptive_weight(self, nll_loss, g_loss, last_layer=None):
+        if last_layer is not None:
+            nll_grads = torch.autograd.grad(nll_loss, last_layer, retain_graph=True)[0]
+            g_grads = torch.autograd.grad(g_loss, last_layer, retain_graph=True)[0]
+        else:
+            nll_grads = torch.autograd.grad(nll_loss, self.last_layer[0], retain_graph=True)[0]
+            g_grads = torch.autograd.grad(g_loss, self.last_layer[0], retain_graph=True)[0]
+
+        d_weight = torch.norm(nll_grads) / (torch.norm(g_grads) + 1e-4)
+        d_weight = torch.clamp(d_weight, 0.0, 1e4).detach()
+        d_weight = d_weight * self.discriminator_weight
+        return d_weight
+
+    def forward(self, codebook_loss, inputs, reconstructions, optimizer_idx,
+                global_step, last_layer=None, cond=None, split="train", predicted_indices=None):
+        if not exists(codebook_loss):
+            codebook_loss = torch.tensor([0.]).to(inputs.device)
+        #rec_loss = torch.abs(inputs.contiguous() - reconstructions.contiguous())
+        rec_loss = self.pixel_loss(inputs.contiguous(), reconstructions.contiguous())
+        if self.perceptual_weight > 0:
+            p_loss = self.perceptual_loss(inputs.contiguous(), reconstructions.contiguous())
+            rec_loss = rec_loss + self.perceptual_weight * p_loss
+        else:
+            p_loss = torch.tensor([0.0])
+
+        nll_loss = rec_loss
+        #nll_loss = torch.sum(nll_loss) / nll_loss.shape[0]
+        nll_loss = torch.mean(nll_loss)
+
+        # now the GAN part
+        if optimizer_idx == 0:
+            # generator update
+            if cond is None:
+                assert not self.disc_conditional
+                logits_fake = self.discriminator(reconstructions.contiguous())
+            else:
+                assert self.disc_conditional
+                logits_fake = self.discriminator(torch.cat((reconstructions.contiguous(), cond), dim=1))
+            g_loss = -torch.mean(logits_fake)
+
+            try:
+                d_weight = self.calculate_adaptive_weight(nll_loss, g_loss, last_layer=last_layer)
+            except RuntimeError:
+                assert not self.training
+                d_weight = torch.tensor(0.0)
+
+            disc_factor = adopt_weight(self.disc_factor, global_step, threshold=self.discriminator_iter_start)
+            loss = nll_loss + d_weight * disc_factor * g_loss + self.codebook_weight * codebook_loss.mean()
+
+            log = {"{}/total_loss".format(split): loss.clone().detach().mean(),
+                   "{}/quant_loss".format(split): codebook_loss.detach().mean(),
+                   "{}/nll_loss".format(split): nll_loss.detach().mean(),
+                   "{}/rec_loss".format(split): rec_loss.detach().mean(),
+                   "{}/p_loss".format(split): p_loss.detach().mean(),
+                   "{}/d_weight".format(split): d_weight.detach(),
+                   "{}/disc_factor".format(split): torch.tensor(disc_factor),
+                   "{}/g_loss".format(split): g_loss.detach().mean(),
+                   }
+            if predicted_indices is not None:
+                assert self.n_classes is not None
+                with torch.no_grad():
+                    perplexity, cluster_usage = measure_perplexity(predicted_indices, self.n_classes)
+                log[f"{split}/perplexity"] = perplexity
+                log[f"{split}/cluster_usage"] = cluster_usage
+            return loss, log
+
+        if optimizer_idx == 1:
+            # second pass for discriminator update
+            if cond is None:
+                logits_real = self.discriminator(inputs.contiguous().detach())
+                logits_fake = self.discriminator(reconstructions.contiguous().detach())
+            else:
+                logits_real = self.discriminator(torch.cat((inputs.contiguous().detach(), cond), dim=1))
+                logits_fake = self.discriminator(torch.cat((reconstructions.contiguous().detach(), cond), dim=1))
+
+            disc_factor = adopt_weight(self.disc_factor, global_step, threshold=self.discriminator_iter_start)
+            d_loss = disc_factor * self.disc_loss(logits_real, logits_fake)
+
+            log = {"{}/disc_loss".format(split): d_loss.clone().detach().mean(),
+                   "{}/logits_real".format(split): logits_real.detach().mean(),
+                   "{}/logits_fake".format(split): logits_fake.detach().mean()
+                   }
+            return d_loss, log

ldm/modules/x_transformer.py

@@ -0,0 +1,650 @@
+"""shout-out to https://github.com/lucidrains/x-transformers/tree/main/x_transformers"""
+import torch
+from torch import nn, einsum
+import torch.nn.functional as F
+from functools import partial
+from inspect import isfunction
+from collections import namedtuple
+from einops import rearrange, repeat, reduce
+
+# constants
+
+DEFAULT_DIM_HEAD = 64
+
+Intermediates = namedtuple('Intermediates', [
+    'pre_softmax_attn',
+    'post_softmax_attn'
+])
+
+LayerIntermediates = namedtuple('Intermediates', [
+    'hiddens',
+    'attn_intermediates'
+])
+
+
+class AbsolutePositionalEmbedding(nn.Module):
+    def __init__(self, dim, max_seq_len):
+        super().__init__()
+        self.emb = nn.Embedding(max_seq_len, dim)
+        self.init_()
+
+    def init_(self):
+        nn.init.normal_(self.emb.weight, std=0.02)
+
+    def forward(self, x):
+        n = torch.arange(x.shape[1], device=x.device)
+        return self.emb(n)[None, :, :]
+
+
+class FixedPositionalEmbedding(nn.Module):
+    def __init__(self, dim):
+        super().__init__()
+        inv_freq = 1. / (10000 ** (torch.arange(0, dim, 2).float() / dim))
+        self.register_buffer('inv_freq', inv_freq)
+
+    def forward(self, x, seq_dim=1, offset=0):
+        t = torch.arange(x.shape[seq_dim], device=x.device).type_as(self.inv_freq) + offset
+        sinusoid_inp = torch.einsum('i , j -> i j', t, self.inv_freq)
+        emb = torch.cat((sinusoid_inp.sin(), sinusoid_inp.cos()), dim=-1)
+        return emb[None, :, :]
+
+
+# helpers
+
+def exists(val):
+    return val is not None
+
+
+def default(val, d):
+    if exists(val):
+        return val
+    return d() if isfunction(d) else d
+
+
+def always(val):
+    def inner(*args, **kwargs):
+        return val
+    return inner
+
+
+def not_equals(val):
+    def inner(x):
+        return x != val
+    return inner
+
+
+def equals(val):
+    def inner(x):
+        return x == val
+    return inner
+
+
+def max_neg_value(tensor):
+    return -torch.finfo(tensor.dtype).max
+
+
+# keyword argument helpers
+
+def pick_and_pop(keys, d):
+    values = list(map(lambda key: d.pop(key), keys))
+    return dict(zip(keys, values))
+
+
+def group_dict_by_key(cond, d):
+    return_val = [dict(), dict()]
+    for key in d.keys():
+        match = bool(cond(key))
+        ind = int(not match)
+        return_val[ind][key] = d[key]
+    return (*return_val,)
+
+
+def string_begins_with(prefix, str):
+    return str.startswith(prefix)
+
+
+def group_by_key_prefix(prefix, d):
+    return group_dict_by_key(partial(string_begins_with, prefix), d)
+
+
+def groupby_prefix_and_trim(prefix, d):
+    kwargs_with_prefix, kwargs = group_dict_by_key(partial(string_begins_with, prefix), d)
+    kwargs_without_prefix = dict(map(lambda x: (x[0][len(prefix):], x[1]), tuple(kwargs_with_prefix.items())))
+    return kwargs_without_prefix, kwargs
+
+
+# classes
+class Scale(nn.Module):
+    def __init__(self, value, fn):
+        super().__init__()
+        self.value = value
+        self.fn = fn
+
+    def forward(self, x, **kwargs):
+        x, *rest = self.fn(x, **kwargs)
+        return (x * self.value, *rest)
+
+
+class Rezero(nn.Module):
+    def __init__(self, fn):
+        super().__init__()
+        self.fn = fn
+        self.g = nn.Parameter(torch.zeros(1))
+
+    def forward(self, x, **kwargs):
+        x, *rest = self.fn(x, **kwargs)
+        return (x * self.g, *rest)
+
+
+class ScaleNorm(nn.Module):
+    def __init__(self, dim, eps=1e-5):
+        super().__init__()
+        self.scale = dim ** -0.5
+        self.eps = eps
+        self.g = nn.Parameter(torch.ones(1))
+
+    def forward(self, x):
+        norm = torch.norm(x, dim=-1, keepdim=True) * self.scale
+        return x / norm.clamp(min=self.eps) * self.g
+
+
+class RMSNorm(nn.Module):
+    def __init__(self, dim, eps=1e-8):
+        super().__init__()
+        self.scale = dim ** -0.5
+        self.eps = eps
+        self.g = nn.Parameter(torch.ones(dim))
+
+    def forward(self, x):
+        norm = torch.norm(x, dim=-1, keepdim=True) * self.scale
+        return x / norm.clamp(min=self.eps) * self.g
+
+
+class Residual(nn.Module):
+    def forward(self, x, residual):
+        return x + residual
+
+
+class GRUGating(nn.Module):
+    def __init__(self, dim):
+        super().__init__()
+        self.gru = nn.GRUCell(dim, dim)
+
+    def forward(self, x, residual):
+        gated_output = self.gru(
+            rearrange(x, 'b n d -> (b n) d'),
+            rearrange(residual, 'b n d -> (b n) d')
+        )
+
+        return gated_output.reshape_as(x)
+
+
+# feedforward
+
+class GEGLU(nn.Module):
+    def __init__(self, dim_in, dim_out):
+        super().__init__()
+        self.proj = nn.Linear(dim_in, dim_out * 2)
+
+    def forward(self, x):
+        x, gate = self.proj(x).chunk(2, dim=-1)
+        return x * F.gelu(gate)
+
+
+class FeedForward(nn.Module):
+    def __init__(self, dim, dim_out=None, mult=4, glu=False, dropout=0.):
+        super().__init__()
+        inner_dim = int(dim * mult)
+        dim_out = default(dim_out, dim)
+        project_in = nn.Sequential(
+            nn.Linear(dim, inner_dim),
+            nn.GELU()
+        ) if not glu else GEGLU(dim, inner_dim)
+
+        self.net = nn.Sequential(
+            project_in,
+            nn.Dropout(dropout),
+            nn.Linear(inner_dim, dim_out)
+        )
+
+    def forward(self, x):
+        return self.net(x)
+
+
+# attention.
+class Attention(nn.Module):
+    def __init__(
+            self,
+            dim,
+            dim_head=DEFAULT_DIM_HEAD,
+            heads=8,
+            causal=False,
+            mask=None,
+            talking_heads=False,
+            sparse_topk=None,
+            use_entmax15=False,
+            num_mem_kv=0,
+            dropout=0.,
+            on_attn=False
+    ):
+        super().__init__()
+        if use_entmax15:
+            raise NotImplementedError("Check out entmax activation instead of softmax activation!")
+        self.scale = dim_head ** -0.5
+        self.heads = heads
+        self.causal = causal
+        self.mask = mask
+
+        inner_dim = dim_head * heads
+
+        self.to_q = nn.Linear(dim, inner_dim, bias=False)
+        self.to_k = nn.Linear(dim, inner_dim, bias=False)
+        self.to_v = nn.Linear(dim, inner_dim, bias=False)
+        self.dropout = nn.Dropout(dropout)
+
+        # talking heads
+        self.talking_heads = talking_heads
+        if talking_heads:
+            self.pre_softmax_proj = nn.Parameter(torch.randn(heads, heads))
+            self.post_softmax_proj = nn.Parameter(torch.randn(heads, heads))
+
+        # explicit topk sparse attention
+        self.sparse_topk = sparse_topk
+
+        # entmax
+        #self.attn_fn = entmax15 if use_entmax15 else F.softmax
+        self.attn_fn = F.softmax
+
+        # add memory key / values
+        self.num_mem_kv = num_mem_kv
+        if num_mem_kv > 0:
+            self.mem_k = nn.Parameter(torch.randn(heads, num_mem_kv, dim_head))
+            self.mem_v = nn.Parameter(torch.randn(heads, num_mem_kv, dim_head))
+
+        # attention on attention
+        self.attn_on_attn = on_attn
+        self.to_out = nn.Sequential(nn.Linear(inner_dim, dim * 2), nn.GLU()) if on_attn else nn.Linear(inner_dim, dim)
+
+    def forward(
+            self,
+            x,
+            context=None,
+            mask=None,
+            context_mask=None,
+            rel_pos=None,
+            sinusoidal_emb=None,
+            prev_attn=None,
+            mem=None
+    ):
+        b, n, _, h, talking_heads, device = *x.shape, self.heads, self.talking_heads, x.device
+        kv_input = default(context, x)
+
+        q_input = x
+        k_input = kv_input
+        v_input = kv_input
+
+        if exists(mem):
+            k_input = torch.cat((mem, k_input), dim=-2)
+            v_input = torch.cat((mem, v_input), dim=-2)
+
+        if exists(sinusoidal_emb):
+            # in shortformer, the query would start at a position offset depending on the past cached memory
+            offset = k_input.shape[-2] - q_input.shape[-2]
+            q_input = q_input + sinusoidal_emb(q_input, offset=offset)
+            k_input = k_input + sinusoidal_emb(k_input)
+
+        q = self.to_q(q_input)
+        k = self.to_k(k_input)
+        v = self.to_v(v_input)
+
+        q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h=h), (q, k, v))
+
+        input_mask = None
+        if any(map(exists, (mask, context_mask))):
+            q_mask = default(mask, lambda: torch.ones((b, n), device=device).bool())
+            k_mask = q_mask if not exists(context) else context_mask
+            k_mask = default(k_mask, lambda: torch.ones((b, k.shape[-2]), device=device).bool())
+            q_mask = rearrange(q_mask, 'b i -> b () i ()')
+            k_mask = rearrange(k_mask, 'b j -> b () () j')
+            input_mask = q_mask * k_mask
+
+        if self.num_mem_kv > 0:
+            mem_k, mem_v = map(lambda t: repeat(t, 'h n d -> b h n d', b=b), (self.mem_k, self.mem_v))
+            k = torch.cat((mem_k, k), dim=-2)
+            v = torch.cat((mem_v, v), dim=-2)
+            if exists(input_mask):
+                input_mask = F.pad(input_mask, (self.num_mem_kv, 0), value=True)
+
+        dots = einsum('b h i d, b h j d -> b h i j', q, k) * self.scale
+        mask_value = max_neg_value(dots)
+
+        if exists(prev_attn):
+            dots = dots + prev_attn
+
+        pre_softmax_attn = dots
+
+        if talking_heads:
+            dots = einsum('b h i j, h k -> b k i j', dots, self.pre_softmax_proj).contiguous()
+
+        if exists(rel_pos):
+            dots = rel_pos(dots)
+
+        if exists(input_mask):
+            dots.masked_fill_(~input_mask, mask_value)
+            del input_mask
+
+        if self.causal:
+            i, j = dots.shape[-2:]
+            r = torch.arange(i, device=device)
+            mask = rearrange(r, 'i -> () () i ()') < rearrange(r, 'j -> () () () j')
+            mask = F.pad(mask, (j - i, 0), value=False)
+            dots.masked_fill_(mask, mask_value)
+            del mask
+
+        if exists(self.sparse_topk) and self.sparse_topk < dots.shape[-1]:
+            top, _ = dots.topk(self.sparse_topk, dim=-1)
+            vk = top[..., -1].unsqueeze(-1).expand_as(dots)
+            mask = dots < vk
+            dots.masked_fill_(mask, mask_value)
+            del mask
+
+        attn = self.attn_fn(dots, dim=-1)
+        post_softmax_attn = attn
+
+        attn = self.dropout(attn)
+
+        if talking_heads:
+            attn = einsum('b h i j, h k -> b k i j', attn, self.post_softmax_proj).contiguous()
+
+        out = einsum('b h i j, b h j d -> b h i d', attn, v)
+        out = rearrange(out, 'b h n d -> b n (h d)')
+
+        intermediates = Intermediates(
+            pre_softmax_attn=pre_softmax_attn,
+            post_softmax_attn=post_softmax_attn
+        )
+
+        return self.to_out(out), intermediates
+
+
+class AttentionLayers(nn.Module):
+    def __init__(
+            self,
+            dim,
+            depth,
+            heads=8,
+            causal=False,
+            cross_attend=False,
+            only_cross=False,
+            use_scalenorm=False,
+            use_rmsnorm=False,
+            use_rezero=False,
+            rel_pos_num_buckets=32,
+            rel_pos_max_distance=128,
+            position_infused_attn=False,
+            custom_layers=None,
+            sandwich_coef=None,
+            par_ratio=None,
+            residual_attn=False,
+            cross_residual_attn=False,
+            macaron=False,
+            pre_norm=True,
+            gate_residual=False,
+            **kwargs
+    ):
+        super().__init__()
+        ff_kwargs, kwargs = groupby_prefix_and_trim('ff_', kwargs)
+        attn_kwargs, _ = groupby_prefix_and_trim('attn_', kwargs)
+
+        dim_head = attn_kwargs.get('dim_head', DEFAULT_DIM_HEAD)
+
+        self.dim = dim
+        self.depth = depth
+        self.layers = nn.ModuleList([])
+
+        self.has_pos_emb = position_infused_attn
+        self.pia_pos_emb = FixedPositionalEmbedding(dim) if position_infused_attn else None
+        self.rotary_pos_emb = always(None)
+
+        assert rel_pos_num_buckets <= rel_pos_max_distance, 'number of relative position buckets must be less than the relative position max distance'
+        self.rel_pos = None
+
+        self.pre_norm = pre_norm
+
+        self.residual_attn = residual_attn
+        self.cross_residual_attn = cross_residual_attn
+
+        norm_class = ScaleNorm if use_scalenorm else nn.LayerNorm
+        norm_class = RMSNorm if use_rmsnorm else norm_class
+        norm_fn = partial(norm_class, dim)
+
+        norm_fn = nn.Identity if use_rezero else norm_fn
+        branch_fn = Rezero if use_rezero else None
+
+        if cross_attend and not only_cross:
+            default_block = ('a', 'c', 'f')
+        elif cross_attend and only_cross:
+            default_block = ('c', 'f')
+        else:
+            default_block = ('a', 'f')
+
+        if macaron:
+            default_block = ('f',) + default_block
+
+        if exists(custom_layers):
+            layer_types = custom_layers
+        elif exists(par_ratio):
+            par_depth = depth * len(default_block)
+            assert 1 < par_ratio <= par_depth, 'par ratio out of range'
+            default_block = tuple(filter(not_equals('f'), default_block))
+            par_attn = par_depth // par_ratio
+            depth_cut = par_depth * 2 // 3  # 2 / 3 attention layer cutoff suggested by PAR paper
+            par_width = (depth_cut + depth_cut // par_attn) // par_attn
+            assert len(default_block) <= par_width, 'default block is too large for par_ratio'
+            par_block = default_block + ('f',) * (par_width - len(default_block))
+            par_head = par_block * par_attn
+            layer_types = par_head + ('f',) * (par_depth - len(par_head))
+        elif exists(sandwich_coef):
+            assert sandwich_coef > 0 and sandwich_coef <= depth, 'sandwich coefficient should be less than the depth'
+            layer_types = ('a',) * sandwich_coef + default_block * (depth - sandwich_coef) + ('f',) * sandwich_coef
+        else:
+            layer_types = default_block * depth
+
+        self.layer_types = layer_types
+        self.num_attn_layers = len(list(filter(equals('a'), layer_types)))
+
+        for layer_type in self.layer_types:
+            if layer_type == 'a':
+                layer = Attention(dim, heads=heads, causal=causal, **attn_kwargs)
+            elif layer_type == 'c':
+                layer = Attention(dim, heads=heads, **attn_kwargs)
+            elif layer_type == 'f':
+                layer = FeedForward(dim, **ff_kwargs)
+                layer = layer if not macaron else Scale(0.5, layer)
+            else:
+                raise Exception(f'invalid layer type {layer_type}')
+
+            if isinstance(layer, Attention) and exists(branch_fn):
+                layer = branch_fn(layer)
+
+            if gate_residual:
+                residual_fn = GRUGating(dim)
+            else:
+                residual_fn = Residual()
+
+            self.layers.append(nn.ModuleList([
+                norm_fn(),
+                layer,
+                residual_fn
+            ]))
+
+    def forward(
+            self,
+            x,
+            context=None,
+            mask=None,
+            context_mask=None,
+            mems=None,
+            return_hiddens=False,
+            **kwargs
+    ):
+        hiddens = []
+        intermediates = []
+        prev_attn = None
+        prev_cross_attn = None
+
+        mems = mems.copy() if exists(mems) else [None] * self.num_attn_layers
+
+        for ind, (layer_type, (norm, block, residual_fn)) in enumerate(zip(self.layer_types, self.layers)):
+            is_last = ind == (len(self.layers) - 1)
+
+            if layer_type == 'a':
+                hiddens.append(x)
+                layer_mem = mems.pop(0)
+
+            residual = x
+
+            if self.pre_norm:
+                x = norm(x)
+
+            if layer_type == 'a':
+                out, inter = block(x, mask=mask, sinusoidal_emb=self.pia_pos_emb, rel_pos=self.rel_pos,
+                                   prev_attn=prev_attn, mem=layer_mem)
+            elif layer_type == 'c':
+                out, inter = block(x, context=context, mask=mask, context_mask=context_mask, prev_attn=prev_cross_attn)
+            elif layer_type == 'f':
+                out = block(x)
+
+            x = residual_fn(out, residual)
+
+            if layer_type in ('a', 'c'):
+                intermediates.append(inter)
+
+            if layer_type == 'a' and self.residual_attn:
+                prev_attn = inter.pre_softmax_attn
+            elif layer_type == 'c' and self.cross_residual_attn:
+                prev_cross_attn = inter.pre_softmax_attn
+
+            if not self.pre_norm and not is_last:
+                x = norm(x)
+
+        if return_hiddens:
+            intermediates = LayerIntermediates(
+                hiddens=hiddens,
+                attn_intermediates=intermediates
+            )
+
+            return x, intermediates
+
+        return x
+
+
+class Encoder(AttentionLayers):
+    def __init__(self, **kwargs):
+        assert 'causal' not in kwargs, 'cannot set causality on encoder'
+        super().__init__(causal=False, **kwargs)
+
+
+
+class TransformerWrapper(nn.Module):
+    def __init__(
+            self,
+            *,
+            num_tokens,
+            max_seq_len,
+            attn_layers,
+            emb_dim=None,
+            max_mem_len=0.,
+            emb_dropout=0.,
+            num_memory_tokens=None,
+            tie_embedding=False,
+            use_pos_emb=True
+    ):
+        super().__init__()
+        assert isinstance(attn_layers, AttentionLayers), 'attention layers must be one of Encoder or Decoder'
+
+        dim = attn_layers.dim
+        emb_dim = default(emb_dim, dim)
+
+        self.max_seq_len = max_seq_len
+        self.max_mem_len = max_mem_len
+        self.num_tokens = num_tokens
+
+        self.token_emb = nn.Embedding(num_tokens, emb_dim)
+        self.pos_emb = AbsolutePositionalEmbedding(emb_dim, max_seq_len) if (
+                    use_pos_emb and not attn_layers.has_pos_emb) else always(0)
+        self.emb_dropout = nn.Dropout(emb_dropout)
+
+        self.project_emb = nn.Linear(emb_dim, dim) if emb_dim != dim else nn.Identity()
+        self.attn_layers = attn_layers
+        self.norm = nn.LayerNorm(dim)
+
+        self.init_()
+
+        self.to_logits = nn.Linear(dim, num_tokens) if not tie_embedding else lambda t: t @ self.token_emb.weight.t()
+
+        # memory tokens (like [cls]) from Memory Transformers paper
+        num_memory_tokens = default(num_memory_tokens, 0)
+        self.num_memory_tokens = num_memory_tokens
+        if num_memory_tokens > 0:
+            self.memory_tokens = nn.Parameter(torch.randn(num_memory_tokens, dim))
+
+            # let funnel encoder know number of memory tokens, if specified
+            if hasattr(attn_layers, 'num_memory_tokens'):
+                attn_layers.num_memory_tokens = num_memory_tokens
+
+    def init_(self):
+        nn.init.normal_(self.token_emb.weight, std=0.02)
+
+    def forward(
+            self,
+            x,
+            return_embeddings=False,
+            mask=None,
+            return_mems=False,
+            return_attn=False,
+            mems=None,
+            embedding_manager=None,
+            **kwargs
+    ):
+        b, n, device, num_mem = *x.shape, x.device, self.num_memory_tokens
+
+        embedded_x = self.token_emb(x)
+        
+        if embedding_manager:
+            x = embedding_manager(x, embedded_x)
+        else:
+            x = embedded_x
+
+        x = x + self.pos_emb(x)
+        x = self.emb_dropout(x)
+
+        x = self.project_emb(x)
+
+        if num_mem > 0:
+            mem = repeat(self.memory_tokens, 'n d -> b n d', b=b)
+            x = torch.cat((mem, x), dim=1)
+
+            # auto-handle masking after appending memory tokens
+            if exists(mask):
+                mask = F.pad(mask, (num_mem, 0), value=True)
+
+        x, intermediates = self.attn_layers(x, mask=mask, mems=mems, return_hiddens=True, **kwargs)
+        x = self.norm(x)
+
+        mem, x = x[:, :num_mem], x[:, num_mem:]
+
+        out = self.to_logits(x) if not return_embeddings else x
+
+        if return_mems:
+            hiddens = intermediates.hiddens
+            new_mems = list(map(lambda pair: torch.cat(pair, dim=-2), zip(mems, hiddens))) if exists(mems) else hiddens
+            new_mems = list(map(lambda t: t[..., -self.max_mem_len:, :].detach(), new_mems))
+            return out, new_mems
+
+        if return_attn:
+            attn_maps = list(map(lambda t: t.post_softmax_attn, intermediates.attn_intermediates))
+            return out, attn_maps
+
+        return out
+

ldm/util.py

@@ -0,0 +1,205 @@
+import importlib
+
+import torch
+import numpy as np
+from collections import abc
+from einops import rearrange
+from functools import partial
+
+import multiprocessing as mp
+from threading import Thread
+from queue import Queue
+
+from inspect import isfunction
+from PIL import Image, ImageDraw, ImageFont
+
+
+def log_txt_as_img(wh, xc, size=10):
+    # wh a tuple of (width, height)
+    # xc a list of captions to plot
+    b = len(xc)
+    txts = list()
+    for bi in range(b):
+        txt = Image.new("RGB", wh, color="white")
+        draw = ImageDraw.Draw(txt)
+        # Changed to work on Windows
+        font = ImageFont.load_default()
+        #font = ImageFont.truetype('data/DejaVuSans.ttf', size=size)
+        nc = int(40 * (wh[0] / 256))
+        lines = "\n".join(xc[bi][start:start + nc] for start in range(0, len(xc[bi]), nc))
+
+        try:
+            draw.text((0, 0), lines, fill="black", font=font)
+        except UnicodeEncodeError:
+            print("Cant encode string for logging. Skipping.")
+
+        txt = np.array(txt).transpose(2, 0, 1) / 127.5 - 1.0
+        txts.append(txt)
+    txts = np.stack(txts)
+    txts = torch.tensor(txts)
+    return txts
+
+
+def ismap(x):
+    if not isinstance(x, torch.Tensor):
+        return False
+    return (len(x.shape) == 4) and (x.shape[1] > 3)
+
+
+def isimage(x):
+    if not isinstance(x, torch.Tensor):
+        return False
+    return (len(x.shape) == 4) and (x.shape[1] == 3 or x.shape[1] == 1)
+
+
+def exists(x):
+    return x is not None
+
+
+def default(val, d):
+    if exists(val):
+        return val
+    return d() if isfunction(d) else d
+
+
+def mean_flat(tensor):
+    """
+    https://github.com/openai/guided-diffusion/blob/27c20a8fab9cb472df5d6bdd6c8d11c8f430b924/guided_diffusion/nn.py#L86
+    Take the mean over all non-batch dimensions.
+    """
+    return tensor.mean(dim=list(range(1, len(tensor.shape))))
+
+
+def count_params(model, verbose=False):
+    total_params = sum(p.numel() for p in model.parameters())
+    if verbose:
+        print(f"{model.__class__.__name__} has {total_params * 1.e-6:.2f} M params.")
+    return total_params
+
+
+def instantiate_from_config(config, **kwargs):
+    if not "target" in config:
+        if config == '__is_first_stage__':
+            return None
+        elif config == "__is_unconditional__":
+            return None
+        raise KeyError("Expected key `target` to instantiate.")
+    return get_obj_from_str(config["target"])(**config.get("params", dict()), **kwargs)
+
+
+def get_obj_from_str(string, reload=False):
+    module, cls = string.rsplit(".", 1)
+    if reload:
+        module_imp = importlib.import_module(module)
+        importlib.reload(module_imp)
+    return getattr(importlib.import_module(module, package=None), cls)
+
+
+def _do_parallel_data_prefetch(func, Q, data, idx, idx_to_fn=False):
+    # create dummy dataset instance
+
+    # run prefetching
+    if idx_to_fn:
+        res = func(data, worker_id=idx)
+    else:
+        res = func(data)
+    Q.put([idx, res])
+    Q.put("Done")
+
+
+def parallel_data_prefetch(
+        func: callable, data, n_proc, target_data_type="ndarray", cpu_intensive=True, use_worker_id=False
+):
+    # if target_data_type not in ["ndarray", "list"]:
+    #     raise ValueError(
+    #         "Data, which is passed to parallel_data_prefetch has to be either of type list or ndarray."
+    #     )
+    if isinstance(data, np.ndarray) and target_data_type == "list":
+        raise ValueError("list expected but function got ndarray.")
+    elif isinstance(data, abc.Iterable):
+        if isinstance(data, dict):
+            print(
+                f'WARNING:"data" argument passed to parallel_data_prefetch is a dict: Using only its values and disregarding keys.'
+            )
+            data = list(data.values())
+        if target_data_type == "ndarray":
+            data = np.asarray(data)
+        else:
+            data = list(data)
+    else:
+        raise TypeError(
+            f"The data, that shall be processed parallel has to be either an np.ndarray or an Iterable, but is actually {type(data)}."
+        )
+
+    if cpu_intensive:
+        Q = mp.Queue(1000)
+        proc = mp.Process
+    else:
+        Q = Queue(1000)
+        proc = Thread
+    # spawn processes
+    if target_data_type == "ndarray":
+        arguments = [
+            [func, Q, part, i, use_worker_id]
+            for i, part in enumerate(np.array_split(data, n_proc))
+        ]
+    else:
+        step = (
+            int(len(data) / n_proc + 1)
+            if len(data) % n_proc != 0
+            else int(len(data) / n_proc)
+        )
+        arguments = [
+            [func, Q, part, i, use_worker_id]
+            for i, part in enumerate(
+                [data[i: i + step] for i in range(0, len(data), step)]
+            )
+        ]
+    processes = []
+    for i in range(n_proc):
+        p = proc(target=_do_parallel_data_prefetch, args=arguments[i])
+        processes += [p]
+
+    # start processes
+    print(f"Start prefetching...")
+    import time
+
+    start = time.time()
+    gather_res = [[] for _ in range(n_proc)]
+    try:
+        for p in processes:
+            p.start()
+
+        k = 0
+        while k < n_proc:
+            # get result
+            res = Q.get()
+            if res == "Done":
+                k += 1
+            else:
+                gather_res[res[0]] = res[1]
+
+    except Exception as e:
+        print("Exception: ", e)
+        for p in processes:
+            p.terminate()
+
+        raise e
+    finally:
+        for p in processes:
+            p.join()
+        print(f"Prefetching complete. [{time.time() - start} sec.]")
+
+    if target_data_type == 'ndarray':
+        if not isinstance(gather_res[0], np.ndarray):
+            return np.concatenate([np.asarray(r) for r in gather_res], axis=0)
+
+        # order outputs
+        return np.concatenate(gather_res, axis=0)
+    elif target_data_type == 'list':
+        out = []
+        for r in gather_res:
+            out.extend(r)
+        return out
+    else:
+        return gather_res

main.py

@@ -0,0 +1,855 @@
+import argparse, os, sys, datetime, glob, importlib, csv
+import numpy as np
+import time
+import torch
+
+import torchvision
+import pytorch_lightning as pl
+
+from packaging import version
+from omegaconf import OmegaConf
+from torch.utils.data import random_split, DataLoader, Dataset, Subset
+from functools import partial
+from PIL import Image
+
+from pytorch_lightning import seed_everything
+from pytorch_lightning.trainer import Trainer
+from pytorch_lightning.callbacks import ModelCheckpoint, Callback, LearningRateMonitor
+from pytorch_lightning.utilities.distributed import rank_zero_only
+from pytorch_lightning.utilities import rank_zero_info
+
+from ldm.data.base import Txt2ImgIterableBaseDataset
+from ldm.util import instantiate_from_config
+
+def load_model_from_config(config, ckpt, verbose=False):
+    print(f"Loading model from {ckpt}")
+    pl_sd = torch.load(ckpt, map_location="cpu")
+    sd = pl_sd["state_dict"]
+    config.model.params.ckpt_path = ckpt
+    model = instantiate_from_config(config.model)
+    m, u = model.load_state_dict(sd, strict=False)
+    if len(m) > 0 and verbose:
+        print("missing keys:")
+        print(m)
+    if len(u) > 0 and verbose:
+        print("unexpected keys:")
+        print(u)
+
+    model.cuda()
+    return model
+
+def get_parser(**parser_kwargs):
+    def str2bool(v):
+        if isinstance(v, bool):
+            return v
+        if v.lower() in ("yes", "true", "t", "y", "1"):
+            return True
+        elif v.lower() in ("no", "false", "f", "n", "0"):
+            return False
+        else:
+            raise argparse.ArgumentTypeError("Boolean value expected.")
+
+    parser = argparse.ArgumentParser(**parser_kwargs)
+    parser.add_argument(
+        "-n",
+        "--name",
+        type=str,
+        const=True,
+        default="",
+        nargs="?",
+        help="postfix for logdir",
+    )
+    parser.add_argument(
+        "-r",
+        "--resume",
+        type=str,
+        const=True,
+        default="",
+        nargs="?",
+        help="resume from logdir or checkpoint in logdir",
+    )
+    parser.add_argument(
+        "-b",
+        "--base",
+        nargs="*",
+        metavar="base_config.yaml",
+        help="paths to base configs. Loaded from left-to-right. "
+             "Parameters can be overwritten or added with command-line options of the form `--key value`.",
+        default=list(),
+    )
+    parser.add_argument(
+        "-t",
+        "--train",
+        type=str2bool,
+        const=True,
+        default=False,
+        nargs="?",
+        help="train",
+    )
+    parser.add_argument(
+        "--no-test",
+        type=str2bool,
+        const=True,
+        default=False,
+        nargs="?",
+        help="disable test",
+    )
+    parser.add_argument(
+        "-p",
+        "--project",
+        help="name of new or path to existing project"
+    )
+    parser.add_argument(
+        "-d",
+        "--debug",
+        type=str2bool,
+        nargs="?",
+        const=True,
+        default=False,
+        help="enable post-mortem debugging",
+    )
+    parser.add_argument(
+        "-s",
+        "--seed",
+        type=int,
+        default=23,
+        help="seed for seed_everything",
+    )
+    parser.add_argument(
+        "-f",
+        "--postfix",
+        type=str,
+        default="",
+        help="post-postfix for default name",
+    )
+    parser.add_argument(
+        "-l",
+        "--logdir",
+        type=str,
+        default="logs",
+        help="directory for logging dat shit",
+    )
+    parser.add_argument(
+        "--scale_lr",
+        type=str2bool,
+        nargs="?",
+        const=False,
+        default=False,
+        help="scale base-lr by ngpu * batch_size * n_accumulate",
+    )
+
+    parser.add_argument(
+        "--datadir_in_name", 
+        type=str2bool, 
+        nargs="?", 
+        const=True, 
+        default=True, 
+        help="Prepend the final directory in the data_root to the output directory name")
+
+    parser.add_argument("--actual_resume", 
+        type=str,
+        required=True,
+        help="Path to model to actually resume from")
+
+    parser.add_argument("--data_root", 
+        type=str, 
+        required=True, 
+        help="Path to directory with training images")
+    
+    parser.add_argument("--reg_data_root", 
+        type=str, 
+        required=True, 
+        help="Path to directory with regularization images")
+
+    parser.add_argument("--embedding_manager_ckpt", 
+        type=str, 
+        default="", 
+        help="Initialize embedding manager from a checkpoint")
+
+    parser.add_argument("--class_word", 
+        type=str, 
+        default="dog",
+        help="Placeholder token which will be used to denote the concept in future prompts")
+
+    parser.add_argument("--init_word", 
+        type=str, 
+        help="Word to use as source for initial token embedding")
+
+    return parser
+
+
+def nondefault_trainer_args(opt):
+    parser = argparse.ArgumentParser()
+    parser = Trainer.add_argparse_args(parser)
+    args = parser.parse_args([])
+    return sorted(k for k in vars(args) if getattr(opt, k) != getattr(args, k))
+
+
+class WrappedDataset(Dataset):
+    """Wraps an arbitrary object with __len__ and __getitem__ into a pytorch dataset"""
+
+    def __init__(self, dataset):
+        self.data = dataset
+
+    def __len__(self):
+        return len(self.data)
+
+    def __getitem__(self, idx):
+        return self.data[idx]
+
+
+def worker_init_fn(_):
+    worker_info = torch.utils.data.get_worker_info()
+
+    dataset = worker_info.dataset
+    worker_id = worker_info.id
+
+    if isinstance(dataset, Txt2ImgIterableBaseDataset):
+        split_size = dataset.num_records // worker_info.num_workers
+        # reset num_records to the true number to retain reliable length information
+        dataset.sample_ids = dataset.valid_ids[worker_id * split_size:(worker_id + 1) * split_size]
+        current_id = np.random.choice(len(np.random.get_state()[1]), 1)
+        return np.random.seed(np.random.get_state()[1][current_id] + worker_id)
+    else:
+        return np.random.seed(np.random.get_state()[1][0] + worker_id)
+
+class ConcatDataset(Dataset):
+    def __init__(self, *datasets):
+        self.datasets = datasets
+
+    def __getitem__(self, idx):
+        return tuple(d[idx] for d in self.datasets)
+
+    def __len__(self):
+        return min(len(d) for d in self.datasets)
+    
+class DataModuleFromConfig(pl.LightningDataModule):
+    def __init__(self, batch_size, train=None, reg = None, validation=None, test=None, predict=None,
+                 wrap=False, num_workers=None, shuffle_test_loader=False, use_worker_init_fn=False,
+                 shuffle_val_dataloader=False):
+        super().__init__()
+        self.batch_size = batch_size
+        self.dataset_configs = dict()
+        self.num_workers = num_workers if num_workers is not None else batch_size * 2
+        self.use_worker_init_fn = use_worker_init_fn
+        if train is not None:
+            self.dataset_configs["train"] = train
+        if reg is not None:
+            self.dataset_configs["reg"] = reg
+        
+        self.train_dataloader = self._train_dataloader
+        
+        if validation is not None:
+            self.dataset_configs["validation"] = validation
+            self.val_dataloader = partial(self._val_dataloader, shuffle=shuffle_val_dataloader)
+        if test is not None:
+            self.dataset_configs["test"] = test
+            self.test_dataloader = partial(self._test_dataloader, shuffle=shuffle_test_loader)
+        if predict is not None:
+            self.dataset_configs["predict"] = predict
+            self.predict_dataloader = self._predict_dataloader
+        self.wrap = wrap
+
+    def prepare_data(self):
+        for data_cfg in self.dataset_configs.values():
+            instantiate_from_config(data_cfg)
+
+    def setup(self, stage=None):
+        self.datasets = dict(
+            (k, instantiate_from_config(self.dataset_configs[k]))
+            for k in self.dataset_configs)
+        if self.wrap:
+            for k in self.datasets:
+                self.datasets[k] = WrappedDataset(self.datasets[k])
+
+    def _train_dataloader(self):
+        is_iterable_dataset = isinstance(self.datasets['train'], Txt2ImgIterableBaseDataset)
+        if is_iterable_dataset or self.use_worker_init_fn:
+            init_fn = worker_init_fn
+        else:
+            init_fn = None
+        train_set = self.datasets["train"]
+        reg_set = self.datasets["reg"]
+        concat_dataset = ConcatDataset(train_set, reg_set)
+        return DataLoader(concat_dataset, batch_size=self.batch_size,
+                          num_workers=self.num_workers, shuffle=False if is_iterable_dataset else True,
+                          worker_init_fn=init_fn)
+
+    def _val_dataloader(self, shuffle=False):
+        if isinstance(self.datasets['validation'], Txt2ImgIterableBaseDataset) or self.use_worker_init_fn:
+            init_fn = worker_init_fn
+        else:
+            init_fn = None
+        return DataLoader(self.datasets["validation"],
+                          batch_size=self.batch_size,
+                          num_workers=self.num_workers,
+                          worker_init_fn=init_fn,
+                          shuffle=shuffle)
+
+    def _test_dataloader(self, shuffle=False):
+        is_iterable_dataset = isinstance(self.datasets['train'], Txt2ImgIterableBaseDataset)
+        if is_iterable_dataset or self.use_worker_init_fn:
+            init_fn = worker_init_fn
+        else:
+            init_fn = None
+
+        # do not shuffle dataloader for iterable dataset
+        shuffle = shuffle and (not is_iterable_dataset)
+
+        return DataLoader(self.datasets["test"], batch_size=self.batch_size,
+                          num_workers=self.num_workers, worker_init_fn=init_fn, shuffle=shuffle)
+
+    def _predict_dataloader(self, shuffle=False):
+        if isinstance(self.datasets['predict'], Txt2ImgIterableBaseDataset) or self.use_worker_init_fn:
+            init_fn = worker_init_fn
+        else:
+            init_fn = None
+        return DataLoader(self.datasets["predict"], batch_size=self.batch_size,
+                          num_workers=self.num_workers, worker_init_fn=init_fn)
+
+
+class SetupCallback(Callback):
+    def __init__(self, resume, now, logdir, ckptdir, cfgdir, config, lightning_config):
+        super().__init__()
+        self.resume = resume
+        self.now = now
+        self.logdir = logdir
+        self.ckptdir = ckptdir
+        self.cfgdir = cfgdir
+        self.config = config
+        self.lightning_config = lightning_config
+
+    def on_keyboard_interrupt(self, trainer, pl_module):
+        if trainer.global_rank == 0:
+            print("Summoning checkpoint.")
+            ckpt_path = os.path.join(self.ckptdir, "last.ckpt")
+            trainer.save_checkpoint(ckpt_path)
+
+    def on_pretrain_routine_start(self, trainer, pl_module):
+        if trainer.global_rank == 0:
+            # Create logdirs and save configs
+            os.makedirs(self.logdir, exist_ok=True)
+            os.makedirs(self.ckptdir, exist_ok=True)
+            os.makedirs(self.cfgdir, exist_ok=True)
+
+            if "callbacks" in self.lightning_config:
+                if 'metrics_over_trainsteps_checkpoint' in self.lightning_config['callbacks']:
+                    os.makedirs(os.path.join(self.ckptdir, 'trainstep_checkpoints'), exist_ok=True)
+            print("Project config")
+            print(OmegaConf.to_yaml(self.config))
+            OmegaConf.save(self.config,
+                           os.path.join(self.cfgdir, "{}-project.yaml".format(self.now)))
+
+            print("Lightning config")
+            print(OmegaConf.to_yaml(self.lightning_config))
+            OmegaConf.save(OmegaConf.create({"lightning": self.lightning_config}),
+                           os.path.join(self.cfgdir, "{}-lightning.yaml".format(self.now)))
+
+        else:
+            # ModelCheckpoint callback created log directory --- remove it
+            if not self.resume and os.path.exists(self.logdir):
+                dst, name = os.path.split(self.logdir)
+                dst = os.path.join(dst, "child_runs", name)
+                os.makedirs(os.path.split(dst)[0], exist_ok=True)
+                try:
+                    os.rename(self.logdir, dst)
+                except FileNotFoundError:
+                    pass
+
+
+class ImageLogger(Callback):
+    def __init__(self, batch_frequency, max_images, clamp=True, increase_log_steps=True,
+                 rescale=True, disabled=False, log_on_batch_idx=False, log_first_step=False,
+                 log_images_kwargs=None):
+        super().__init__()
+        self.rescale = rescale
+        self.batch_freq = batch_frequency
+        self.max_images = max_images
+        self.logger_log_images = {
+            pl.loggers.TestTubeLogger: self._testtube,
+        }
+        self.log_steps = [2 ** n for n in range(int(np.log2(self.batch_freq)) + 1)]
+        if not increase_log_steps:
+            self.log_steps = [self.batch_freq]
+        self.clamp = clamp
+        self.disabled = disabled
+        self.log_on_batch_idx = log_on_batch_idx
+        self.log_images_kwargs = log_images_kwargs if log_images_kwargs else {}
+        self.log_first_step = log_first_step
+
+    @rank_zero_only
+    def _testtube(self, pl_module, images, batch_idx, split):
+        for k in images:
+            grid = torchvision.utils.make_grid(images[k])
+            grid = (grid + 1.0) / 2.0  # -1,1 -> 0,1; c,h,w
+
+            tag = f"{split}/{k}"
+            pl_module.logger.experiment.add_image(
+                tag, grid,
+                global_step=pl_module.global_step)
+
+    @rank_zero_only
+    def log_local(self, save_dir, split, images,
+                  global_step, current_epoch, batch_idx):
+        root = os.path.join(save_dir, "images", split)
+        for k in images:
+            grid = torchvision.utils.make_grid(images[k], nrow=4)
+            if self.rescale:
+                grid = (grid + 1.0) / 2.0  # -1,1 -> 0,1; c,h,w
+            grid = grid.transpose(0, 1).transpose(1, 2).squeeze(-1)
+            grid = grid.numpy()
+            grid = (grid * 255).astype(np.uint8)
+            filename = "{}_globalstep-{:05}_epoch-{:01}_batch-{:04}.jpg".format(
+                k,
+                global_step,
+                current_epoch,
+                batch_idx)
+            path = os.path.join(root, filename)
+            os.makedirs(os.path.split(path)[0], exist_ok=True)
+            Image.fromarray(grid).save(path)
+
+    def log_img(self, pl_module, batch, batch_idx, split="train"):
+        check_idx = batch_idx if self.log_on_batch_idx else pl_module.global_step
+        if (self.check_frequency(check_idx) and  # batch_idx % self.batch_freq == 0
+                hasattr(pl_module, "log_images") and
+                callable(pl_module.log_images) and
+                self.max_images > 0):
+            logger = type(pl_module.logger)
+
+            is_train = pl_module.training
+            if is_train:
+                pl_module.eval()
+
+            with torch.no_grad():
+                images = pl_module.log_images(batch, split=split, **self.log_images_kwargs)
+
+            for k in images:
+                N = min(images[k].shape[0], self.max_images)
+                images[k] = images[k][:N]
+                if isinstance(images[k], torch.Tensor):
+                    images[k] = images[k].detach().cpu()
+                    if self.clamp:
+                        images[k] = torch.clamp(images[k], -1., 1.)
+
+            self.log_local(pl_module.logger.save_dir, split, images,
+                           pl_module.global_step, pl_module.current_epoch, batch_idx)
+
+            logger_log_images = self.logger_log_images.get(logger, lambda *args, **kwargs: None)
+            logger_log_images(pl_module, images, pl_module.global_step, split)
+
+            if is_train:
+                pl_module.train()
+
+    def check_frequency(self, check_idx):
+        if ((check_idx % self.batch_freq) == 0 or (check_idx in self.log_steps)) and (
+                check_idx > 0 or self.log_first_step):
+            try:
+                self.log_steps.pop(0)
+            except IndexError as e:
+                print(e)
+                pass
+            return True
+        return False
+
+    def on_train_batch_end(self, trainer, pl_module, outputs, batch, batch_idx, dataloader_idx):
+        if not self.disabled and (pl_module.global_step > 0 or self.log_first_step):
+            self.log_img(pl_module, batch, batch_idx, split="train")
+
+    def on_validation_batch_end(self, trainer, pl_module, outputs, batch, batch_idx, dataloader_idx):
+        if not self.disabled and pl_module.global_step > 0:
+            self.log_img(pl_module, batch, batch_idx, split="val")
+        if hasattr(pl_module, 'calibrate_grad_norm'):
+            if (pl_module.calibrate_grad_norm and batch_idx % 25 == 0) and batch_idx > 0:
+                self.log_gradients(trainer, pl_module, batch_idx=batch_idx)
+
+
+class CUDACallback(Callback):
+    # see https://github.com/SeanNaren/minGPT/blob/master/mingpt/callback.py
+    def on_train_epoch_start(self, trainer, pl_module):
+        # Reset the memory use counter
+        torch.cuda.reset_peak_memory_stats(trainer.root_gpu)
+        torch.cuda.synchronize(trainer.root_gpu)
+        self.start_time = time.time()
+
+    def on_train_epoch_end(self, trainer, pl_module):
+        torch.cuda.synchronize(trainer.root_gpu)
+        max_memory = torch.cuda.max_memory_allocated(trainer.root_gpu) / 2 ** 20
+        epoch_time = time.time() - self.start_time
+
+        try:
+            max_memory = trainer.training_type_plugin.reduce(max_memory)
+            epoch_time = trainer.training_type_plugin.reduce(epoch_time)
+
+            rank_zero_info(f"Average Epoch time: {epoch_time:.2f} seconds")
+            rank_zero_info(f"Average Peak memory {max_memory:.2f}MiB")
+        except AttributeError:
+            pass
+
+class ModeSwapCallback(Callback):
+
+    def __init__(self, swap_step=2000):
+        super().__init__()
+        self.is_frozen = False
+        self.swap_step = swap_step
+
+    def on_train_epoch_start(self, trainer, pl_module):
+        if trainer.global_step < self.swap_step and not self.is_frozen:
+            self.is_frozen = True
+            trainer.optimizers = [pl_module.configure_opt_embedding()]
+
+        if trainer.global_step > self.swap_step and self.is_frozen:
+            self.is_frozen = False
+            trainer.optimizers = [pl_module.configure_opt_model()]
+
+if __name__ == "__main__":
+    # custom parser to specify config files, train, test and debug mode,
+    # postfix, resume.
+    # `--key value` arguments are interpreted as arguments to the trainer.
+    # `nested.key=value` arguments are interpreted as config parameters.
+    # configs are merged from left-to-right followed by command line parameters.
+
+    # model:
+    #   base_learning_rate: float
+    #   target: path to lightning module
+    #   params:
+    #       key: value
+    # data:
+    #   target: main.DataModuleFromConfig
+    #   params:
+    #      batch_size: int
+    #      wrap: bool
+    #      train:
+    #          target: path to train dataset
+    #          params:
+    #              key: value
+    #      validation:
+    #          target: path to validation dataset
+    #          params:
+    #              key: value
+    #      test:
+    #          target: path to test dataset
+    #          params:
+    #              key: value
+    # lightning: (optional, has sane defaults and can be specified on cmdline)
+    #   trainer:
+    #       additional arguments to trainer
+    #   logger:
+    #       logger to instantiate
+    #   modelcheckpoint:
+    #       modelcheckpoint to instantiate
+    #   callbacks:
+    #       callback1:
+    #           target: importpath
+    #           params:
+    #               key: value
+
+    now = datetime.datetime.now().strftime("%Y-%m-%dT%H-%M-%S")
+
+    # add cwd for convenience and to make classes in this file available when
+    # running as `python main.py`
+    # (in particular `main.DataModuleFromConfig`)
+    sys.path.append(os.getcwd())
+
+    parser = get_parser()
+    parser = Trainer.add_argparse_args(parser)
+
+    opt, unknown = parser.parse_known_args()
+    if opt.name and opt.resume:
+        raise ValueError(
+            "-n/--name and -r/--resume cannot be specified both."
+            "If you want to resume training in a new log folder, "
+            "use -n/--name in combination with --resume_from_checkpoint"
+        )
+    if opt.resume:
+        if not os.path.exists(opt.resume):
+            raise ValueError("Cannot find {}".format(opt.resume))
+        if os.path.isfile(opt.resume):
+            paths = opt.resume.split("/")
+            # idx = len(paths)-paths[::-1].index("logs")+1
+            # logdir = "/".join(paths[:idx])
+            logdir = "/".join(paths[:-2])
+            ckpt = opt.resume
+        else:
+            assert os.path.isdir(opt.resume), opt.resume
+            logdir = opt.resume.rstrip("/")
+            ckpt = os.path.join(logdir, "checkpoints", "last.ckpt")
+
+        opt.resume_from_checkpoint = ckpt
+        base_configs = sorted(glob.glob(os.path.join(logdir, "configs/*.yaml")))
+        opt.base = base_configs + opt.base
+        _tmp = logdir.split("/")
+        nowname = _tmp[-1]
+    else:
+        if opt.name:
+            name = "_" + opt.name
+        elif opt.base:
+            cfg_fname = os.path.split(opt.base[0])[-1]
+            cfg_name = os.path.splitext(cfg_fname)[0]
+            name = "_" + cfg_name
+        else:
+            name = ""
+
+        if opt.datadir_in_name:
+            now = os.path.basename(os.path.normpath(opt.data_root)) + now
+            
+        nowname = now + name + opt.postfix
+        logdir = os.path.join(opt.logdir, nowname)
+
+    ckptdir = os.path.join(logdir, "checkpoints")
+    cfgdir = os.path.join(logdir, "configs")
+    seed_everything(opt.seed)
+
+    try:
+        # init and save configs
+        configs = [OmegaConf.load(cfg) for cfg in opt.base]
+        cli = OmegaConf.from_dotlist(unknown)
+        config = OmegaConf.merge(*configs, cli)
+        lightning_config = config.pop("lightning", OmegaConf.create())
+        # merge trainer cli with config
+        trainer_config = lightning_config.get("trainer", OmegaConf.create())
+        for k in nondefault_trainer_args(opt):
+            trainer_config[k] = getattr(opt, k)
+        if not "gpus" in trainer_config:
+            del trainer_config["accelerator"]
+            cpu = True
+        else:
+            gpuinfo = trainer_config["gpus"]
+            print(f"Running on GPUs {gpuinfo}")
+            cpu = False
+        trainer_opt = argparse.Namespace(**trainer_config)
+        lightning_config.trainer = trainer_config
+
+        # model
+
+        # config.model.params.personalization_config.params.init_word = opt.init_word
+        # config.model.params.personalization_config.params.embedding_manager_ckpt = opt.embedding_manager_ckpt
+        # config.model.params.personalization_config.params.placeholder_tokens = opt.placeholder_tokens
+
+        # if opt.init_word:
+        #     config.model.params.personalization_config.params.initializer_words[0] = opt.init_word
+            
+        config.data.params.train.params.placeholder_token = opt.class_word
+        config.data.params.reg.params.placeholder_token = opt.class_word
+        config.data.params.validation.params.placeholder_token = opt.class_word
+
+        if opt.actual_resume:
+            model = load_model_from_config(config, opt.actual_resume)
+        else:
+            model = instantiate_from_config(config.model)
+
+        # trainer and callbacks
+        trainer_kwargs = dict()
+
+        # default logger configs
+        default_logger_cfgs = {
+            "wandb": {
+                "target": "pytorch_lightning.loggers.WandbLogger",
+                "params": {
+                    "name": nowname,
+                    "save_dir": logdir,
+                    "offline": opt.debug,
+                    "id": nowname,
+                }
+            },
+            "testtube": {
+                "target": "pytorch_lightning.loggers.TestTubeLogger",
+                "params": {
+                    "name": "testtube",
+                    "save_dir": logdir,
+                }
+            },
+        }
+        default_logger_cfg = default_logger_cfgs["testtube"]
+        if "logger" in lightning_config:
+            logger_cfg = lightning_config.logger
+        else:
+            logger_cfg = OmegaConf.create()
+        logger_cfg = OmegaConf.merge(default_logger_cfg, logger_cfg)
+        trainer_kwargs["logger"] = instantiate_from_config(logger_cfg)
+
+        # modelcheckpoint - use TrainResult/EvalResult(checkpoint_on=metric) to
+        # specify which metric is used to determine best models
+        default_modelckpt_cfg = {
+            "target": "pytorch_lightning.callbacks.ModelCheckpoint",
+            "params": {
+                "dirpath": ckptdir,
+                "filename": "{epoch:06}",
+                "verbose": True,
+                "save_last": True,
+            }
+        }
+        if hasattr(model, "monitor"):
+            print(f"Monitoring {model.monitor} as checkpoint metric.")
+            default_modelckpt_cfg["params"]["monitor"] = model.monitor
+            default_modelckpt_cfg["params"]["save_top_k"] = 1
+
+        if "modelcheckpoint" in lightning_config:
+            modelckpt_cfg = lightning_config.modelcheckpoint
+        else:
+            modelckpt_cfg =  OmegaConf.create()
+        modelckpt_cfg = OmegaConf.merge(default_modelckpt_cfg, modelckpt_cfg)
+        print(f"Merged modelckpt-cfg: \n{modelckpt_cfg}")
+        if version.parse(pl.__version__) < version.parse('1.4.0'):
+            trainer_kwargs["checkpoint_callback"] = instantiate_from_config(modelckpt_cfg)
+
+        # add callback which sets up log directory
+        default_callbacks_cfg = {
+            "setup_callback": {
+                "target": "main.SetupCallback",
+                "params": {
+                    "resume": opt.resume,
+                    "now": now,
+                    "logdir": logdir,
+                    "ckptdir": ckptdir,
+                    "cfgdir": cfgdir,
+                    "config": config,
+                    "lightning_config": lightning_config,
+                }
+            },
+            "image_logger": {
+                "target": "main.ImageLogger",
+                "params": {
+                    "batch_frequency": 750,
+                    "max_images": 4,
+                    "clamp": True
+                }
+            },
+            "learning_rate_logger": {
+                "target": "main.LearningRateMonitor",
+                "params": {
+                    "logging_interval": "step",
+                    # "log_momentum": True
+                }
+            },
+            "cuda_callback": {
+                "target": "main.CUDACallback"
+            },
+        }
+        if version.parse(pl.__version__) >= version.parse('1.4.0'):
+            default_callbacks_cfg.update({'checkpoint_callback': modelckpt_cfg})
+
+        if "callbacks" in lightning_config:
+            callbacks_cfg = lightning_config.callbacks
+        else:
+            callbacks_cfg = OmegaConf.create()
+
+        if 'metrics_over_trainsteps_checkpoint' in callbacks_cfg:
+            print(
+                'Caution: Saving checkpoints every n train steps without deleting. This might require some free space.')
+            default_metrics_over_trainsteps_ckpt_dict = {
+                'metrics_over_trainsteps_checkpoint':
+                    {"target": 'pytorch_lightning.callbacks.ModelCheckpoint',
+                     'params': {
+                         "dirpath": os.path.join(ckptdir, 'trainstep_checkpoints'),
+                         "filename": "{epoch:06}-{step:09}",
+                         "verbose": True,
+                         'save_top_k': -1,
+                         'every_n_train_steps': 10000,
+                         'save_weights_only': True
+                     }
+                     }
+            }
+            default_callbacks_cfg.update(default_metrics_over_trainsteps_ckpt_dict)
+
+        callbacks_cfg = OmegaConf.merge(default_callbacks_cfg, callbacks_cfg)
+        if 'ignore_keys_callback' in callbacks_cfg and hasattr(trainer_opt, 'resume_from_checkpoint'):
+            callbacks_cfg.ignore_keys_callback.params['ckpt_path'] = trainer_opt.resume_from_checkpoint
+        elif 'ignore_keys_callback' in callbacks_cfg:
+            del callbacks_cfg['ignore_keys_callback']
+
+        trainer_kwargs["callbacks"] = [instantiate_from_config(callbacks_cfg[k]) for k in callbacks_cfg]
+        trainer_kwargs["max_steps"] = trainer_opt.max_steps
+
+        trainer = Trainer.from_argparse_args(trainer_opt, **trainer_kwargs)
+        trainer.logdir = logdir  ###
+
+        # data
+        config.data.params.train.params.data_root = opt.data_root
+        config.data.params.reg.params.data_root = opt.reg_data_root
+        config.data.params.validation.params.data_root = opt.data_root
+        data = instantiate_from_config(config.data)
+
+        data = instantiate_from_config(config.data)
+        # NOTE according to https://pytorch-lightning.readthedocs.io/en/latest/datamodules.html
+        # calling these ourselves should not be necessary but it is.
+        # lightning still takes care of proper multiprocessing though
+        data.prepare_data()
+        data.setup()
+        print("#### Data #####")
+        for k in data.datasets:
+            print(f"{k}, {data.datasets[k].__class__.__name__}, {len(data.datasets[k])}")
+
+        # configure learning rate
+        bs, base_lr = config.data.params.batch_size, config.model.base_learning_rate
+        if not cpu:
+            ngpu = len(lightning_config.trainer.gpus.strip(",").split(','))
+        else:
+            ngpu = 1
+        if 'accumulate_grad_batches' in lightning_config.trainer:
+            accumulate_grad_batches = lightning_config.trainer.accumulate_grad_batches
+        else:
+            accumulate_grad_batches = 1
+        print(f"accumulate_grad_batches = {accumulate_grad_batches}")
+        lightning_config.trainer.accumulate_grad_batches = accumulate_grad_batches
+        if opt.scale_lr:
+            model.learning_rate = accumulate_grad_batches * ngpu * bs * base_lr
+            print(
+                "Setting learning rate to {:.2e} = {} (accumulate_grad_batches) * {} (num_gpus) * {} (batchsize) * {:.2e} (base_lr)".format(
+                    model.learning_rate, accumulate_grad_batches, ngpu, bs, base_lr))
+        else:
+            model.learning_rate = base_lr
+            print("++++ NOT USING LR SCALING ++++")
+            print(f"Setting learning rate to {model.learning_rate:.2e}")
+
+
+        # allow checkpointing via USR1
+        def melk(*args, **kwargs):
+            # run all checkpoint hooks
+            if trainer.global_rank == 0:
+                print("Here comes the checkpoint...")
+                ckpt_path = os.path.join(ckptdir, "last.ckpt")
+                trainer.save_checkpoint(ckpt_path)
+
+
+        def divein(*args, **kwargs):
+            if trainer.global_rank == 0:
+                import pudb;
+                pudb.set_trace()
+
+
+        import signal
+
+
+        # Changed to work with windows
+        signal.signal(signal.SIGTERM, melk)
+        #signal.signal(signal.SIGUSR1, melk)
+        signal.signal(signal.SIGTERM, divein)
+        #signal.signal(signal.SIGUSR2, divein)
+
+        # run
+        if opt.train:
+            try:
+                trainer.fit(model, data)
+            except Exception:
+                melk()
+                raise
+        if not opt.no_test and not trainer.interrupted:
+            trainer.test(model, data)
+    except Exception:
+        if opt.debug and trainer.global_rank == 0:
+            try:
+                import pudb as debugger
+            except ImportError:
+                import pdb as debugger
+            debugger.post_mortem()
+        raise
+    finally:
+        # move newly created debug project to debug_runs
+        if opt.debug and not opt.resume and trainer.global_rank == 0:
+            dst, name = os.path.split(logdir)
+            dst = os.path.join(dst, "debug_runs", name)
+            os.makedirs(os.path.split(dst)[0], exist_ok=True)
+            os.rename(logdir, dst)
+        if trainer.global_rank == 0:
+            print("Another one bites the dust...")
+            print(trainer.profiler.summary())

merge_embeddings.py

@@ -0,0 +1,111 @@
+from ldm.modules.encoders.modules import FrozenCLIPEmbedder, BERTEmbedder
+from ldm.modules.embedding_manager import EmbeddingManager
+
+import argparse, os
+from functools import partial
+
+import torch
+
+def get_placeholder_loop(placeholder_string, embedder, is_sd):
+    
+    new_placeholder   = None
+    
+    while True:
+        if new_placeholder is None:
+            new_placeholder = input(f"Placeholder string {placeholder_string} was already used. Please enter a replacement string: ")
+        else:
+            new_placeholder = input(f"Placeholder string '{new_placeholder}' maps to more than a single token. Please enter another string: ")
+
+        token = get_clip_token_for_string(embedder.tokenizer, new_placeholder) if is_sd else get_bert_token_for_string(embedder.tknz_fn, new_placeholder)
+
+        if token is not None:
+            return new_placeholder, token
+            
+def get_clip_token_for_string(tokenizer, string):
+    batch_encoding = tokenizer(string, truncation=True, max_length=77, return_length=True,
+                               return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
+    tokens = batch_encoding["input_ids"]
+
+    if torch.count_nonzero(tokens - 49407) == 2:
+        return tokens[0, 1]
+    
+    return None
+
+def get_bert_token_for_string(tokenizer, string):
+    token = tokenizer(string)
+    if torch.count_nonzero(token) == 3:
+        return token[0, 1]
+
+    return None
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--manager_ckpts", 
+        type=str, 
+        nargs="+", 
+        required=True,
+        help="Paths to a set of embedding managers to be merged."
+    )
+
+    parser.add_argument(
+        "--output_path",
+        type=str,
+        required=True,
+        help="Output path for the merged manager",
+    )
+
+    parser.add_argument(
+        "-sd", "--stable_diffusion",
+        action="store_true",
+        help="Flag to denote that we are merging stable diffusion embeddings"
+    )
+
+    args = parser.parse_args()
+
+    if args.stable_diffusion:
+        embedder = FrozenCLIPEmbedder().cuda()
+    else:
+        embedder = BERTEmbedder(n_embed=1280, n_layer=32).cuda()
+
+    EmbeddingManager = partial(EmbeddingManager, embedder, ["*"])
+
+    string_to_token_dict = {}    
+    string_to_param_dict = torch.nn.ParameterDict()
+
+    placeholder_to_src = {}
+
+    for manager_ckpt in args.manager_ckpts:
+        print(f"Parsing {manager_ckpt}...")
+
+        manager = EmbeddingManager()
+        manager.load(manager_ckpt)
+
+        for placeholder_string in manager.string_to_token_dict:
+            if not placeholder_string in string_to_token_dict:
+                string_to_token_dict[placeholder_string] = manager.string_to_token_dict[placeholder_string]
+                string_to_param_dict[placeholder_string] = manager.string_to_param_dict[placeholder_string]
+
+                placeholder_to_src[placeholder_string] = manager_ckpt
+            else:
+                new_placeholder, new_token = get_placeholder_loop(placeholder_string, embedder, is_sd=args.stable_diffusion)
+                string_to_token_dict[new_placeholder] = new_token
+                string_to_param_dict[new_placeholder] = manager.string_to_param_dict[placeholder_string]
+
+                placeholder_to_src[new_placeholder] = manager_ckpt
+
+    print("Saving combined manager...")
+    merged_manager = EmbeddingManager()
+    merged_manager.string_to_param_dict = string_to_param_dict
+    merged_manager.string_to_token_dict = string_to_token_dict
+    merged_manager.save(args.output_path)
+
+    print("Managers merged. Final list of placeholders: ")
+    print(placeholder_to_src)
+
+
+
+                

models/first_stage_models/kl-f16/config.yaml

@@ -0,0 +1,44 @@
+model:
+  base_learning_rate: 4.5e-06
+  target: ldm.models.autoencoder.AutoencoderKL
+  params:
+    monitor: val/rec_loss
+    embed_dim: 16
+    lossconfig:
+      target: ldm.modules.losses.LPIPSWithDiscriminator
+      params:
+        disc_start: 50001
+        kl_weight: 1.0e-06
+        disc_weight: 0.5
+    ddconfig:
+      double_z: true
+      z_channels: 16
+      resolution: 256
+      in_channels: 3
+      out_ch: 3
+      ch: 128
+      ch_mult:
+      - 1
+      - 1
+      - 2
+      - 2
+      - 4
+      num_res_blocks: 2
+      attn_resolutions:
+      - 16
+      dropout: 0.0
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 6
+    wrap: true
+    train:
+      target: ldm.data.openimages.FullOpenImagesTrain
+      params:
+        size: 384
+        crop_size: 256
+    validation:
+      target: ldm.data.openimages.FullOpenImagesValidation
+      params:
+        size: 384
+        crop_size: 256

models/first_stage_models/kl-f32/config.yaml

@@ -0,0 +1,46 @@
+model:
+  base_learning_rate: 4.5e-06
+  target: ldm.models.autoencoder.AutoencoderKL
+  params:
+    monitor: val/rec_loss
+    embed_dim: 64
+    lossconfig:
+      target: ldm.modules.losses.LPIPSWithDiscriminator
+      params:
+        disc_start: 50001
+        kl_weight: 1.0e-06
+        disc_weight: 0.5
+    ddconfig:
+      double_z: true
+      z_channels: 64
+      resolution: 256
+      in_channels: 3
+      out_ch: 3
+      ch: 128
+      ch_mult:
+      - 1
+      - 1
+      - 2
+      - 2
+      - 4
+      - 4
+      num_res_blocks: 2
+      attn_resolutions:
+      - 16
+      - 8
+      dropout: 0.0
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 6
+    wrap: true
+    train:
+      target: ldm.data.openimages.FullOpenImagesTrain
+      params:
+        size: 384
+        crop_size: 256
+    validation:
+      target: ldm.data.openimages.FullOpenImagesValidation
+      params:
+        size: 384
+        crop_size: 256

models/first_stage_models/kl-f4/config.yaml

@@ -0,0 +1,41 @@
+model:
+  base_learning_rate: 4.5e-06
+  target: ldm.models.autoencoder.AutoencoderKL
+  params:
+    monitor: val/rec_loss
+    embed_dim: 3
+    lossconfig:
+      target: ldm.modules.losses.LPIPSWithDiscriminator
+      params:
+        disc_start: 50001
+        kl_weight: 1.0e-06
+        disc_weight: 0.5
+    ddconfig:
+      double_z: true
+      z_channels: 3
+      resolution: 256
+      in_channels: 3
+      out_ch: 3
+      ch: 128
+      ch_mult:
+      - 1
+      - 2
+      - 4
+      num_res_blocks: 2
+      attn_resolutions: []
+      dropout: 0.0
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 10
+    wrap: true
+    train:
+      target: ldm.data.openimages.FullOpenImagesTrain
+      params:
+        size: 384
+        crop_size: 256
+    validation:
+      target: ldm.data.openimages.FullOpenImagesValidation
+      params:
+        size: 384
+        crop_size: 256

models/first_stage_models/kl-f8/config.yaml

@@ -0,0 +1,42 @@
+model:
+  base_learning_rate: 4.5e-06
+  target: ldm.models.autoencoder.AutoencoderKL
+  params:
+    monitor: val/rec_loss
+    embed_dim: 4
+    lossconfig:
+      target: ldm.modules.losses.LPIPSWithDiscriminator
+      params:
+        disc_start: 50001
+        kl_weight: 1.0e-06
+        disc_weight: 0.5
+    ddconfig:
+      double_z: true
+      z_channels: 4
+      resolution: 256
+      in_channels: 3
+      out_ch: 3
+      ch: 128
+      ch_mult:
+      - 1
+      - 2
+      - 4
+      - 4
+      num_res_blocks: 2
+      attn_resolutions: []
+      dropout: 0.0
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 4
+    wrap: true
+    train:
+      target: ldm.data.openimages.FullOpenImagesTrain
+      params:
+        size: 384
+        crop_size: 256
+    validation:
+      target: ldm.data.openimages.FullOpenImagesValidation
+      params:
+        size: 384
+        crop_size: 256

models/first_stage_models/vq-f16/config.yaml

@@ -0,0 +1,49 @@
+model:
+  base_learning_rate: 4.5e-06
+  target: ldm.models.autoencoder.VQModel
+  params:
+    embed_dim: 8
+    n_embed: 16384
+    ddconfig:
+      double_z: false
+      z_channels: 8
+      resolution: 256
+      in_channels: 3
+      out_ch: 3
+      ch: 128
+      ch_mult:
+      - 1
+      - 1
+      - 2
+      - 2
+      - 4
+      num_res_blocks: 2
+      attn_resolutions:
+      - 16
+      dropout: 0.0
+    lossconfig:
+      target: taming.modules.losses.vqperceptual.VQLPIPSWithDiscriminator
+      params:
+        disc_conditional: false
+        disc_in_channels: 3
+        disc_start: 250001
+        disc_weight: 0.75
+        disc_num_layers: 2
+        codebook_weight: 1.0
+
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 14
+    num_workers: 20
+    wrap: true
+    train:
+      target: ldm.data.openimages.FullOpenImagesTrain
+      params:
+        size: 384
+        crop_size: 256
+    validation:
+      target: ldm.data.openimages.FullOpenImagesValidation
+      params:
+        size: 384
+        crop_size: 256

models/first_stage_models/vq-f4-noattn/config.yaml

@@ -0,0 +1,46 @@
+model:
+  base_learning_rate: 4.5e-06
+  target: ldm.models.autoencoder.VQModel
+  params:
+    embed_dim: 3
+    n_embed: 8192
+    monitor: val/rec_loss
+
+    ddconfig:
+      attn_type: none
+      double_z: false
+      z_channels: 3
+      resolution: 256
+      in_channels: 3
+      out_ch: 3
+      ch: 128
+      ch_mult:
+      - 1
+      - 2
+      - 4
+      num_res_blocks: 2
+      attn_resolutions: []
+      dropout: 0.0
+    lossconfig:
+      target: taming.modules.losses.vqperceptual.VQLPIPSWithDiscriminator
+      params:
+        disc_conditional: false
+        disc_in_channels: 3
+        disc_start: 11
+        disc_weight: 0.75
+        codebook_weight: 1.0
+
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 8
+    num_workers: 12
+    wrap: true
+    train:
+      target: ldm.data.openimages.FullOpenImagesTrain
+      params:
+        crop_size: 256
+    validation:
+      target: ldm.data.openimages.FullOpenImagesValidation
+      params:
+        crop_size: 256

models/first_stage_models/vq-f4/config.yaml

@@ -0,0 +1,45 @@
+model:
+  base_learning_rate: 4.5e-06
+  target: ldm.models.autoencoder.VQModel
+  params:
+    embed_dim: 3
+    n_embed: 8192
+    monitor: val/rec_loss
+
+    ddconfig:
+      double_z: false
+      z_channels: 3
+      resolution: 256
+      in_channels: 3
+      out_ch: 3
+      ch: 128
+      ch_mult:
+      - 1
+      - 2
+      - 4
+      num_res_blocks: 2
+      attn_resolutions: []
+      dropout: 0.0
+    lossconfig:
+      target: taming.modules.losses.vqperceptual.VQLPIPSWithDiscriminator
+      params:
+        disc_conditional: false
+        disc_in_channels: 3
+        disc_start: 0
+        disc_weight: 0.75
+        codebook_weight: 1.0
+
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 8
+    num_workers: 16
+    wrap: true
+    train:
+      target: ldm.data.openimages.FullOpenImagesTrain
+      params:
+        crop_size: 256
+    validation:
+      target: ldm.data.openimages.FullOpenImagesValidation
+      params:
+        crop_size: 256

models/first_stage_models/vq-f8-n256/config.yaml

@@ -0,0 +1,48 @@
+model:
+  base_learning_rate: 4.5e-06
+  target: ldm.models.autoencoder.VQModel
+  params:
+    embed_dim: 4
+    n_embed: 256
+    monitor: val/rec_loss
+    ddconfig:
+      double_z: false
+      z_channels: 4
+      resolution: 256
+      in_channels: 3
+      out_ch: 3
+      ch: 128
+      ch_mult:
+      - 1
+      - 2
+      - 2
+      - 4
+      num_res_blocks: 2
+      attn_resolutions:
+      - 32
+      dropout: 0.0
+    lossconfig:
+      target: taming.modules.losses.vqperceptual.VQLPIPSWithDiscriminator
+      params:
+        disc_conditional: false
+        disc_in_channels: 3
+        disc_start: 250001
+        disc_weight: 0.75
+        codebook_weight: 1.0
+
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 10
+    num_workers: 20
+    wrap: true
+    train:
+      target: ldm.data.openimages.FullOpenImagesTrain
+      params:
+        size: 384
+        crop_size: 256
+    validation:
+      target: ldm.data.openimages.FullOpenImagesValidation
+      params:
+        size: 384
+        crop_size: 256

models/first_stage_models/vq-f8/config.yaml

@@ -0,0 +1,48 @@
+model:
+  base_learning_rate: 4.5e-06
+  target: ldm.models.autoencoder.VQModel
+  params:
+    embed_dim: 4
+    n_embed: 16384
+    monitor: val/rec_loss
+    ddconfig:
+      double_z: false
+      z_channels: 4
+      resolution: 256
+      in_channels: 3
+      out_ch: 3
+      ch: 128
+      ch_mult:
+      - 1
+      - 2
+      - 2
+      - 4
+      num_res_blocks: 2
+      attn_resolutions:
+      - 32
+      dropout: 0.0
+    lossconfig:
+      target: taming.modules.losses.vqperceptual.VQLPIPSWithDiscriminator
+      params:
+        disc_conditional: false
+        disc_in_channels: 3
+        disc_num_layers: 2
+        disc_start: 1
+        disc_weight: 0.6
+        codebook_weight: 1.0
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 10
+    num_workers: 20
+    wrap: true
+    train:
+      target: ldm.data.openimages.FullOpenImagesTrain
+      params:
+        size: 384
+        crop_size: 256
+    validation:
+      target: ldm.data.openimages.FullOpenImagesValidation
+      params:
+        size: 384
+        crop_size: 256

models/ldm/bsr_sr/config.yaml

@@ -0,0 +1,80 @@
+model:
+  base_learning_rate: 1.0e-06
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0155
+    log_every_t: 100
+    timesteps: 1000
+    loss_type: l2
+    first_stage_key: image
+    cond_stage_key: LR_image
+    image_size: 64
+    channels: 3
+    concat_mode: true
+    cond_stage_trainable: false
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 64
+        in_channels: 6
+        out_channels: 3
+        model_channels: 160
+        attention_resolutions:
+        - 16
+        - 8
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 2
+        - 4
+        num_head_channels: 32
+    first_stage_config:
+      target: ldm.models.autoencoder.VQModelInterface
+      params:
+        embed_dim: 3
+        n_embed: 8192
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: false
+          z_channels: 3
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+    cond_stage_config:
+      target: torch.nn.Identity
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 64
+    wrap: false
+    num_workers: 12
+    train:
+      target: ldm.data.openimages.SuperresOpenImagesAdvancedTrain
+      params:
+        size: 256
+        degradation: bsrgan_light
+        downscale_f: 4
+        min_crop_f: 0.5
+        max_crop_f: 1.0
+        random_crop: true
+    validation:
+      target: ldm.data.openimages.SuperresOpenImagesAdvancedValidation
+      params:
+        size: 256
+        degradation: bsrgan_light
+        downscale_f: 4
+        min_crop_f: 0.5
+        max_crop_f: 1.0
+        random_crop: true

models/ldm/celeba256/config.yaml

@@ -0,0 +1,70 @@
+model:
+  base_learning_rate: 2.0e-06
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0195
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    cond_stage_key: class_label
+    image_size: 64
+    channels: 3
+    cond_stage_trainable: false
+    concat_mode: false
+    monitor: val/loss
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 64
+        in_channels: 3
+        out_channels: 3
+        model_channels: 224
+        attention_resolutions:
+        - 8
+        - 4
+        - 2
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 3
+        - 4
+        num_head_channels: 32
+    first_stage_config:
+      target: ldm.models.autoencoder.VQModelInterface
+      params:
+        embed_dim: 3
+        n_embed: 8192
+        ddconfig:
+          double_z: false
+          z_channels: 3
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+    cond_stage_config: __is_unconditional__
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 48
+    num_workers: 5
+    wrap: false
+    train:
+      target: ldm.data.faceshq.CelebAHQTrain
+      params:
+        size: 256
+    validation:
+      target: ldm.data.faceshq.CelebAHQValidation
+      params:
+        size: 256

models/ldm/cin256/config.yaml

@@ -0,0 +1,80 @@
+model:
+  base_learning_rate: 1.0e-06
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0195
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    cond_stage_key: class_label
+    image_size: 32
+    channels: 4
+    cond_stage_trainable: true
+    conditioning_key: crossattn
+    monitor: val/loss_simple_ema
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32
+        in_channels: 4
+        out_channels: 4
+        model_channels: 256
+        attention_resolutions:
+        - 4
+        - 2
+        - 1
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 4
+        num_head_channels: 32
+        use_spatial_transformer: true
+        transformer_depth: 1
+        context_dim: 512
+    first_stage_config:
+      target: ldm.models.autoencoder.VQModelInterface
+      params:
+        embed_dim: 4
+        n_embed: 16384
+        ddconfig:
+          double_z: false
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 2
+          - 4
+          num_res_blocks: 2
+          attn_resolutions:
+          - 32
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.ClassEmbedder
+      params:
+        embed_dim: 512
+        key: class_label
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 64
+    num_workers: 12
+    wrap: false
+    train:
+      target: ldm.data.imagenet.ImageNetTrain
+      params:
+        config:
+          size: 256
+    validation:
+      target: ldm.data.imagenet.ImageNetValidation
+      params:
+        config:
+          size: 256

models/ldm/ffhq256/config.yaml

@@ -0,0 +1,70 @@
+model:
+  base_learning_rate: 2.0e-06
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0195
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    cond_stage_key: class_label
+    image_size: 64
+    channels: 3
+    cond_stage_trainable: false
+    concat_mode: false
+    monitor: val/loss
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 64
+        in_channels: 3
+        out_channels: 3
+        model_channels: 224
+        attention_resolutions:
+        - 8
+        - 4
+        - 2
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 3
+        - 4
+        num_head_channels: 32
+    first_stage_config:
+      target: ldm.models.autoencoder.VQModelInterface
+      params:
+        embed_dim: 3
+        n_embed: 8192
+        ddconfig:
+          double_z: false
+          z_channels: 3
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+    cond_stage_config: __is_unconditional__
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 42
+    num_workers: 5
+    wrap: false
+    train:
+      target: ldm.data.faceshq.FFHQTrain
+      params:
+        size: 256
+    validation:
+      target: ldm.data.faceshq.FFHQValidation
+      params:
+        size: 256

models/ldm/inpainting_big/config.yaml

@@ -0,0 +1,67 @@
+model:
+  base_learning_rate: 1.0e-06
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0205
+    log_every_t: 100
+    timesteps: 1000
+    loss_type: l1
+    first_stage_key: image
+    cond_stage_key: masked_image
+    image_size: 64
+    channels: 3
+    concat_mode: true
+    monitor: val/loss
+    scheduler_config:
+      target: ldm.lr_scheduler.LambdaWarmUpCosineScheduler
+      params:
+        verbosity_interval: 0
+        warm_up_steps: 1000
+        max_decay_steps: 50000
+        lr_start: 0.001
+        lr_max: 0.1
+        lr_min: 0.0001
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 64
+        in_channels: 7
+        out_channels: 3
+        model_channels: 256
+        attention_resolutions:
+        - 8
+        - 4
+        - 2
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 3
+        - 4
+        num_heads: 8
+        resblock_updown: true
+    first_stage_config:
+      target: ldm.models.autoencoder.VQModelInterface
+      params:
+        embed_dim: 3
+        n_embed: 8192
+        monitor: val/rec_loss
+        ddconfig:
+          attn_type: none
+          double_z: false
+          z_channels: 3
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: ldm.modules.losses.contperceptual.DummyLoss
+    cond_stage_config: __is_first_stage__

models/ldm/layout2img-openimages256/config.yaml

@@ -0,0 +1,81 @@
+model:
+  base_learning_rate: 2.0e-06
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0205
+    log_every_t: 100
+    timesteps: 1000
+    loss_type: l1
+    first_stage_key: image
+    cond_stage_key: coordinates_bbox
+    image_size: 64
+    channels: 3
+    conditioning_key: crossattn
+    cond_stage_trainable: true
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 64
+        in_channels: 3
+        out_channels: 3
+        model_channels: 128
+        attention_resolutions:
+        - 8
+        - 4
+        - 2
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 3
+        - 4
+        num_head_channels: 32
+        use_spatial_transformer: true
+        transformer_depth: 3
+        context_dim: 512
+    first_stage_config:
+      target: ldm.models.autoencoder.VQModelInterface
+      params:
+        embed_dim: 3
+        n_embed: 8192
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: false
+          z_channels: 3
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.BERTEmbedder
+      params:
+        n_embed: 512
+        n_layer: 16
+        vocab_size: 8192
+        max_seq_len: 92
+        use_tokenizer: false
+    monitor: val/loss_simple_ema
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 24
+    wrap: false
+    num_workers: 10
+    train:
+      target: ldm.data.openimages.OpenImagesBBoxTrain
+      params:
+        size: 256
+    validation:
+      target: ldm.data.openimages.OpenImagesBBoxValidation
+      params:
+        size: 256

models/ldm/lsun_beds256/config.yaml

@@ -0,0 +1,70 @@
+model:
+  base_learning_rate: 2.0e-06
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0195
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    cond_stage_key: class_label
+    image_size: 64
+    channels: 3
+    cond_stage_trainable: false
+    concat_mode: false
+    monitor: val/loss
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 64
+        in_channels: 3
+        out_channels: 3
+        model_channels: 224
+        attention_resolutions:
+        - 8
+        - 4
+        - 2
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 3
+        - 4
+        num_head_channels: 32
+    first_stage_config:
+      target: ldm.models.autoencoder.VQModelInterface
+      params:
+        embed_dim: 3
+        n_embed: 8192
+        ddconfig:
+          double_z: false
+          z_channels: 3
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+    cond_stage_config: __is_unconditional__
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 48
+    num_workers: 5
+    wrap: false
+    train:
+      target: ldm.data.lsun.LSUNBedroomsTrain
+      params:
+        size: 256
+    validation:
+      target: ldm.data.lsun.LSUNBedroomsValidation
+      params:
+        size: 256

models/ldm/lsun_churches256/config.yaml

@@ -0,0 +1,92 @@
+model:
+  base_learning_rate: 5.0e-05
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0155
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    loss_type: l1
+    first_stage_key: image
+    cond_stage_key: image
+    image_size: 32
+    channels: 4
+    cond_stage_trainable: false
+    concat_mode: false
+    scale_by_std: true
+    monitor: val/loss_simple_ema
+    scheduler_config:
+      target: ldm.lr_scheduler.LambdaLinearScheduler
+      params:
+        warm_up_steps:
+        - 10000
+        cycle_lengths:
+        - 10000000000000
+        f_start:
+        - 1.0e-06
+        f_max:
+        - 1.0
+        f_min:
+        - 1.0
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32
+        in_channels: 4
+        out_channels: 4
+        model_channels: 192
+        attention_resolutions:
+        - 1
+        - 2
+        - 4
+        - 8
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 2
+        - 4
+        - 4
+        num_heads: 8
+        use_scale_shift_norm: true
+        resblock_updown: true
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config: '__is_unconditional__'
+
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 96
+    num_workers: 5
+    wrap: false
+    train:
+      target: ldm.data.lsun.LSUNChurchesTrain
+      params:
+        size: 256
+    validation:
+      target: ldm.data.lsun.LSUNChurchesValidation
+      params:
+        size: 256

models/ldm/semantic_synthesis256/config.yaml

@@ -0,0 +1,59 @@
+model:
+  base_learning_rate: 1.0e-06
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0205
+    log_every_t: 100
+    timesteps: 1000
+    loss_type: l1
+    first_stage_key: image
+    cond_stage_key: segmentation
+    image_size: 64
+    channels: 3
+    concat_mode: true
+    cond_stage_trainable: true
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 64
+        in_channels: 6
+        out_channels: 3
+        model_channels: 128
+        attention_resolutions:
+        - 32
+        - 16
+        - 8
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 4
+        - 8
+        num_heads: 8
+    first_stage_config:
+      target: ldm.models.autoencoder.VQModelInterface
+      params:
+        embed_dim: 3
+        n_embed: 8192
+        ddconfig:
+          double_z: false
+          z_channels: 3
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.SpatialRescaler
+      params:
+        n_stages: 2
+        in_channels: 182
+        out_channels: 3

models/ldm/semantic_synthesis512/config.yaml

@@ -0,0 +1,78 @@
+model:
+  base_learning_rate: 1.0e-06
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0205
+    log_every_t: 100
+    timesteps: 1000
+    loss_type: l1
+    first_stage_key: image
+    cond_stage_key: segmentation
+    image_size: 128
+    channels: 3
+    concat_mode: true
+    cond_stage_trainable: true
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 128
+        in_channels: 6
+        out_channels: 3
+        model_channels: 128
+        attention_resolutions:
+        - 32
+        - 16
+        - 8
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 4
+        - 8
+        num_heads: 8
+    first_stage_config:
+      target: ldm.models.autoencoder.VQModelInterface
+      params:
+        embed_dim: 3
+        n_embed: 8192
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: false
+          z_channels: 3
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.SpatialRescaler
+      params:
+        n_stages: 2
+        in_channels: 182
+        out_channels: 3
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 8
+    wrap: false
+    num_workers: 10
+    train:
+      target: ldm.data.landscapes.RFWTrain
+      params:
+        size: 768
+        crop_size: 512
+        segmentation_to_float32: true
+    validation:
+      target: ldm.data.landscapes.RFWValidation
+      params:
+        size: 768
+        crop_size: 512
+        segmentation_to_float32: true

models/ldm/text2img256/config.yaml

@@ -0,0 +1,77 @@
+model:
+  base_learning_rate: 2.0e-06
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.0015
+    linear_end: 0.0195
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: image
+    cond_stage_key: caption
+    image_size: 64
+    channels: 3
+    cond_stage_trainable: true
+    conditioning_key: crossattn
+    monitor: val/loss_simple_ema
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 64
+        in_channels: 3
+        out_channels: 3
+        model_channels: 192
+        attention_resolutions:
+        - 8
+        - 4
+        - 2
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 3
+        - 5
+        num_head_channels: 32
+        use_spatial_transformer: true
+        transformer_depth: 1
+        context_dim: 640
+    first_stage_config:
+      target: ldm.models.autoencoder.VQModelInterface
+      params:
+        embed_dim: 3
+        n_embed: 8192
+        ddconfig:
+          double_z: false
+          z_channels: 3
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.BERTEmbedder
+      params:
+        n_embed: 640
+        n_layer: 32
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 28
+    num_workers: 10
+    wrap: false
+    train:
+      target: ldm.data.previews.pytorch_dataset.PreviewsTrain
+      params:
+        size: 256
+    validation:
+      target: ldm.data.previews.pytorch_dataset.PreviewsValidation
+      params:
+        size: 256

scripts/download_first_stages.sh

@@ -0,0 +1,41 @@
+#!/bin/bash
+wget -O models/first_stage_models/kl-f4/model.zip https://ommer-lab.com/files/latent-diffusion/kl-f4.zip
+wget -O models/first_stage_models/kl-f8/model.zip https://ommer-lab.com/files/latent-diffusion/kl-f8.zip
+wget -O models/first_stage_models/kl-f16/model.zip https://ommer-lab.com/files/latent-diffusion/kl-f16.zip
+wget -O models/first_stage_models/kl-f32/model.zip https://ommer-lab.com/files/latent-diffusion/kl-f32.zip
+wget -O models/first_stage_models/vq-f4/model.zip https://ommer-lab.com/files/latent-diffusion/vq-f4.zip
+wget -O models/first_stage_models/vq-f4-noattn/model.zip https://ommer-lab.com/files/latent-diffusion/vq-f4-noattn.zip
+wget -O models/first_stage_models/vq-f8/model.zip https://ommer-lab.com/files/latent-diffusion/vq-f8.zip
+wget -O models/first_stage_models/vq-f8-n256/model.zip https://ommer-lab.com/files/latent-diffusion/vq-f8-n256.zip
+wget -O models/first_stage_models/vq-f16/model.zip https://ommer-lab.com/files/latent-diffusion/vq-f16.zip
+
+
+
+cd models/first_stage_models/kl-f4
+unzip -o model.zip
+
+cd ../kl-f8
+unzip -o model.zip
+
+cd ../kl-f16
+unzip -o model.zip
+
+cd ../kl-f32
+unzip -o model.zip
+
+cd ../vq-f4
+unzip -o model.zip
+
+cd ../vq-f4-noattn
+unzip -o model.zip
+
+cd ../vq-f8
+unzip -o model.zip
+
+cd ../vq-f8-n256
+unzip -o model.zip
+
+cd ../vq-f16
+unzip -o model.zip
+
+cd ../..

scripts/download_models.sh

@@ -0,0 +1,49 @@
+#!/bin/bash
+wget -O models/ldm/celeba256/celeba-256.zip https://ommer-lab.com/files/latent-diffusion/celeba.zip
+wget -O models/ldm/ffhq256/ffhq-256.zip https://ommer-lab.com/files/latent-diffusion/ffhq.zip
+wget -O models/ldm/lsun_churches256/lsun_churches-256.zip https://ommer-lab.com/files/latent-diffusion/lsun_churches.zip
+wget -O models/ldm/lsun_beds256/lsun_beds-256.zip https://ommer-lab.com/files/latent-diffusion/lsun_bedrooms.zip
+wget -O models/ldm/text2img256/model.zip https://ommer-lab.com/files/latent-diffusion/text2img.zip
+wget -O models/ldm/cin256/model.zip https://ommer-lab.com/files/latent-diffusion/cin.zip
+wget -O models/ldm/semantic_synthesis512/model.zip https://ommer-lab.com/files/latent-diffusion/semantic_synthesis.zip
+wget -O models/ldm/semantic_synthesis256/model.zip https://ommer-lab.com/files/latent-diffusion/semantic_synthesis256.zip
+wget -O models/ldm/bsr_sr/model.zip https://ommer-lab.com/files/latent-diffusion/sr_bsr.zip
+wget -O models/ldm/layout2img-openimages256/model.zip https://ommer-lab.com/files/latent-diffusion/layout2img_model.zip
+wget -O models/ldm/inpainting_big/model.zip https://ommer-lab.com/files/latent-diffusion/inpainting_big.zip
+
+
+
+cd models/ldm/celeba256
+unzip -o celeba-256.zip
+
+cd ../ffhq256
+unzip -o ffhq-256.zip
+
+cd ../lsun_churches256
+unzip -o lsun_churches-256.zip
+
+cd ../lsun_beds256
+unzip -o lsun_beds-256.zip
+
+cd ../text2img256
+unzip -o model.zip
+
+cd ../cin256
+unzip -o model.zip
+
+cd ../semantic_synthesis512
+unzip -o model.zip
+
+cd ../semantic_synthesis256
+unzip -o model.zip
+
+cd ../bsr_sr
+unzip -o model.zip
+
+cd ../layout2img-openimages256
+unzip -o model.zip
+
+cd ../inpainting_big
+unzip -o model.zip
+
+cd ../..

scripts/evaluate_model.py

@@ -0,0 +1,89 @@
+import argparse, os, sys, glob
+
+sys.path.append(os.path.join(sys.path[0], '..'))
+
+import torch
+import numpy as np
+from omegaconf import OmegaConf
+from PIL import Image
+from tqdm import tqdm, trange
+from einops import rearrange
+from torchvision.utils import make_grid
+
+from ldm.util import instantiate_from_config
+from ldm.models.diffusion.ddim import DDIMSampler
+from ldm.models.diffusion.plms import PLMSSampler
+from ldm.data.personalized import PersonalizedBase
+from evaluation.clip_eval import LDMCLIPEvaluator
+
+def load_model_from_config(config, ckpt, verbose=False):
+    print(f"Loading model from {ckpt}")
+    pl_sd = torch.load(ckpt, map_location="cpu")
+    sd = pl_sd["state_dict"]
+    model = instantiate_from_config(config.model)
+    m, u = model.load_state_dict(sd, strict=False)
+    if len(m) > 0 and verbose:
+        print("missing keys:")
+        print(m)
+    if len(u) > 0 and verbose:
+        print("unexpected keys:")
+        print(u)
+
+    model.cuda()
+    model.eval()
+    return model
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--prompt",
+        type=str,
+        nargs="?",
+        default="a painting of a virus monster playing guitar",
+        help="the prompt to render"
+    )
+
+    parser.add_argument(
+        "--ckpt_path", 
+        type=str, 
+        default="/data/pretrained_models/ldm/text2img-large/model.ckpt", 
+        help="Path to pretrained ldm text2img model")
+
+    parser.add_argument(
+        "--embedding_path", 
+        type=str, 
+        help="Path to a pre-trained embedding manager checkpoint")
+
+    parser.add_argument(
+        "--data_dir",
+        type=str,
+        help="Path to directory with images used to train the embedding vectors"
+    )
+
+    opt = parser.parse_args()
+
+
+    config = OmegaConf.load("configs/latent-diffusion/txt2img-1p4B-eval_with_tokens.yaml")  # TODO: Optionally download from same location as ckpt and chnage this logic
+    model = load_model_from_config(config, opt.ckpt_path)  # TODO: check path
+    model.embedding_manager.load(opt.embedding_path)
+
+    device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
+    model = model.to(device)
+
+    evaluator = LDMCLIPEvaluator(device)
+
+    prompt = opt.prompt
+
+    data_loader = PersonalizedBase(opt.data_dir, size=256, flip_p=0.0)
+
+    images = [torch.from_numpy(data_loader[i]["image"]).permute(2, 0, 1) for i in range(data_loader.num_images)]
+    images = torch.stack(images, axis=0)
+
+    sim_img, sim_text = evaluator.evaluate(model, images, opt.prompt)
+
+    output_dir = os.path.join(opt.out_dir, prompt.replace(" ", "-"))
+
+    print("Image similarity: ", sim_img)
+    print("Text similarity: ", sim_text)

scripts/inpaint.py

@@ -0,0 +1,98 @@
+import argparse, os, sys, glob
+from omegaconf import OmegaConf
+from PIL import Image
+from tqdm import tqdm
+import numpy as np
+import torch
+from main import instantiate_from_config
+from ldm.models.diffusion.ddim import DDIMSampler
+
+
+def make_batch(image, mask, device):
+    image = np.array(Image.open(image).convert("RGB"))
+    image = image.astype(np.float32)/255.0
+    image = image[None].transpose(0,3,1,2)
+    image = torch.from_numpy(image)
+
+    mask = np.array(Image.open(mask).convert("L"))
+    mask = mask.astype(np.float32)/255.0
+    mask = mask[None,None]
+    mask[mask < 0.5] = 0
+    mask[mask >= 0.5] = 1
+    mask = torch.from_numpy(mask)
+
+    masked_image = (1-mask)*image
+
+    batch = {"image": image, "mask": mask, "masked_image": masked_image}
+    for k in batch:
+        batch[k] = batch[k].to(device=device)
+        batch[k] = batch[k]*2.0-1.0
+    return batch
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--indir",
+        type=str,
+        nargs="?",
+        help="dir containing image-mask pairs (`example.png` and `example_mask.png`)",
+    )
+    parser.add_argument(
+        "--outdir",
+        type=str,
+        nargs="?",
+        help="dir to write results to",
+    )
+    parser.add_argument(
+        "--steps",
+        type=int,
+        default=50,
+        help="number of ddim sampling steps",
+    )
+    opt = parser.parse_args()
+
+    masks = sorted(glob.glob(os.path.join(opt.indir, "*_mask.png")))
+    images = [x.replace("_mask.png", ".png") for x in masks]
+    print(f"Found {len(masks)} inputs.")
+
+    config = OmegaConf.load("models/ldm/inpainting_big/config.yaml")
+    model = instantiate_from_config(config.model)
+    model.load_state_dict(torch.load("models/ldm/inpainting_big/last.ckpt")["state_dict"],
+                          strict=False)
+
+    device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
+    model = model.to(device)
+    sampler = DDIMSampler(model)
+
+    os.makedirs(opt.outdir, exist_ok=True)
+    with torch.no_grad():
+        with model.ema_scope():
+            for image, mask in tqdm(zip(images, masks)):
+                outpath = os.path.join(opt.outdir, os.path.split(image)[1])
+                batch = make_batch(image, mask, device=device)
+
+                # encode masked image and concat downsampled mask
+                c = model.cond_stage_model.encode(batch["masked_image"])
+                cc = torch.nn.functional.interpolate(batch["mask"],
+                                                     size=c.shape[-2:])
+                c = torch.cat((c, cc), dim=1)
+
+                shape = (c.shape[1]-1,)+c.shape[2:]
+                samples_ddim, _ = sampler.sample(S=opt.steps,
+                                                 conditioning=c,
+                                                 batch_size=c.shape[0],
+                                                 shape=shape,
+                                                 verbose=False)
+                x_samples_ddim = model.decode_first_stage(samples_ddim)
+
+                image = torch.clamp((batch["image"]+1.0)/2.0,
+                                    min=0.0, max=1.0)
+                mask = torch.clamp((batch["mask"]+1.0)/2.0,
+                                   min=0.0, max=1.0)
+                predicted_image = torch.clamp((x_samples_ddim+1.0)/2.0,
+                                              min=0.0, max=1.0)
+
+                inpainted = (1-mask)*image+mask*predicted_image
+                inpainted = inpainted.cpu().numpy().transpose(0,2,3,1)[0]*255
+                Image.fromarray(inpainted.astype(np.uint8)).save(outpath)

scripts/prune-ckpt.py

@@ -0,0 +1,58 @@
+import os
+import torch
+import argparse
+import glob
+
+
+parser = argparse.ArgumentParser(description='Pruning')
+parser.add_argument('--ckpt', type=str, default=None, help='path to model ckpt')
+args = parser.parse_args()
+ckpt = args.ckpt
+
+def prune_it(p, keep_only_ema=False):
+    print(f"prunin' in path: {p}")
+    size_initial = os.path.getsize(p)
+    nsd = dict()
+    sd = torch.load(p, map_location="cpu")
+    print(sd.keys())
+    for k in sd.keys():
+        if k != "optimizer_states":
+            nsd[k] = sd[k]
+    else:
+        print(f"removing optimizer states for path {p}")
+    if "global_step" in sd:
+        print(f"This is global step {sd['global_step']}.")
+    if keep_only_ema:
+        sd = nsd["state_dict"].copy()
+        # infer ema keys
+        ema_keys = {k: "model_ema." + k[6:].replace(".", ".") for k in sd.keys() if k.startswith("model.")}
+        new_sd = dict()
+
+        for k in sd:
+            if k in ema_keys:
+                new_sd[k] = sd[ema_keys[k]].half()
+            elif not k.startswith("model_ema.") or k in ["model_ema.num_updates", "model_ema.decay"]:
+                new_sd[k] = sd[k].half()
+
+        assert len(new_sd) == len(sd) - len(ema_keys)
+        nsd["state_dict"] = new_sd
+    else:
+        sd = nsd['state_dict'].copy()
+        new_sd = dict()
+        for k in sd:
+            new_sd[k] = sd[k].half()
+        nsd['state_dict'] = new_sd
+
+    fn = f"{os.path.splitext(p)[0]}-pruned.ckpt" if not keep_only_ema else f"{os.path.splitext(p)[0]}-ema-pruned.ckpt"
+    print(f"saving pruned checkpoint at: {fn}")
+    torch.save(nsd, fn)
+    newsize = os.path.getsize(fn)
+    MSG = f"New ckpt size: {newsize*1e-9:.2f} GB. " + \
+          f"Saved {(size_initial - newsize)*1e-9:.2f} GB by removing optimizer states"
+    if keep_only_ema:
+        MSG += " and non-EMA weights"
+    print(MSG)
+
+
+if __name__ == "__main__":
+    prune_it(ckpt)

scripts/sample_diffusion.py

@@ -0,0 +1,313 @@
+import argparse, os, sys, glob, datetime, yaml
+import torch
+import time
+import numpy as np
+from tqdm import trange
+
+from omegaconf import OmegaConf
+from PIL import Image
+
+from ldm.models.diffusion.ddim import DDIMSampler
+from ldm.util import instantiate_from_config
+
+rescale = lambda x: (x + 1.) / 2.
+
+def custom_to_pil(x):
+    x = x.detach().cpu()
+    x = torch.clamp(x, -1., 1.)
+    x = (x + 1.) / 2.
+    x = x.permute(1, 2, 0).numpy()
+    x = (255 * x).astype(np.uint8)
+    x = Image.fromarray(x)
+    if not x.mode == "RGB":
+        x = x.convert("RGB")
+    return x
+
+
+def custom_to_np(x):
+    # saves the batch in adm style as in https://github.com/openai/guided-diffusion/blob/main/scripts/image_sample.py
+    sample = x.detach().cpu()
+    sample = ((sample + 1) * 127.5).clamp(0, 255).to(torch.uint8)
+    sample = sample.permute(0, 2, 3, 1)
+    sample = sample.contiguous()
+    return sample
+
+
+def logs2pil(logs, keys=["sample"]):
+    imgs = dict()
+    for k in logs:
+        try:
+            if len(logs[k].shape) == 4:
+                img = custom_to_pil(logs[k][0, ...])
+            elif len(logs[k].shape) == 3:
+                img = custom_to_pil(logs[k])
+            else:
+                print(f"Unknown format for key {k}. ")
+                img = None
+        except:
+            img = None
+        imgs[k] = img
+    return imgs
+
+
+@torch.no_grad()
+def convsample(model, shape, return_intermediates=True,
+               verbose=True,
+               make_prog_row=False):
+
+
+    if not make_prog_row:
+        return model.p_sample_loop(None, shape,
+                                   return_intermediates=return_intermediates, verbose=verbose)
+    else:
+        return model.progressive_denoising(
+            None, shape, verbose=True
+        )
+
+
+@torch.no_grad()
+def convsample_ddim(model, steps, shape, eta=1.0
+                    ):
+    ddim = DDIMSampler(model)
+    bs = shape[0]
+    shape = shape[1:]
+    samples, intermediates = ddim.sample(steps, batch_size=bs, shape=shape, eta=eta, verbose=False,)
+    return samples, intermediates
+
+
+@torch.no_grad()
+def make_convolutional_sample(model, batch_size, vanilla=False, custom_steps=None, eta=1.0,):
+
+
+    log = dict()
+
+    shape = [batch_size,
+             model.model.diffusion_model.in_channels,
+             model.model.diffusion_model.image_size,
+             model.model.diffusion_model.image_size]
+
+    with model.ema_scope("Plotting"):
+        t0 = time.time()
+        if vanilla:
+            sample, progrow = convsample(model, shape,
+                                         make_prog_row=True)
+        else:
+            sample, intermediates = convsample_ddim(model,  steps=custom_steps, shape=shape,
+                                                    eta=eta)
+
+        t1 = time.time()
+
+    x_sample = model.decode_first_stage(sample)
+
+    log["sample"] = x_sample
+    log["time"] = t1 - t0
+    log['throughput'] = sample.shape[0] / (t1 - t0)
+    print(f'Throughput for this batch: {log["throughput"]}')
+    return log
+
+def run(model, logdir, batch_size=50, vanilla=False, custom_steps=None, eta=None, n_samples=50000, nplog=None):
+    if vanilla:
+        print(f'Using Vanilla DDPM sampling with {model.num_timesteps} sampling steps.')
+    else:
+        print(f'Using DDIM sampling with {custom_steps} sampling steps and eta={eta}')
+
+
+    tstart = time.time()
+    n_saved = len(glob.glob(os.path.join(logdir,'*.png')))-1
+    # path = logdir
+    if model.cond_stage_model is None:
+        all_images = []
+
+        print(f"Running unconditional sampling for {n_samples} samples")
+        for _ in trange(n_samples // batch_size, desc="Sampling Batches (unconditional)"):
+            logs = make_convolutional_sample(model, batch_size=batch_size,
+                                             vanilla=vanilla, custom_steps=custom_steps,
+                                             eta=eta)
+            n_saved = save_logs(logs, logdir, n_saved=n_saved, key="sample")
+            all_images.extend([custom_to_np(logs["sample"])])
+            if n_saved >= n_samples:
+                print(f'Finish after generating {n_saved} samples')
+                break
+        all_img = np.concatenate(all_images, axis=0)
+        all_img = all_img[:n_samples]
+        shape_str = "x".join([str(x) for x in all_img.shape])
+        nppath = os.path.join(nplog, f"{shape_str}-samples.npz")
+        np.savez(nppath, all_img)
+
+    else:
+       raise NotImplementedError('Currently only sampling for unconditional models supported.')
+
+    print(f"sampling of {n_saved} images finished in {(time.time() - tstart) / 60.:.2f} minutes.")
+
+
+def save_logs(logs, path, n_saved=0, key="sample", np_path=None):
+    for k in logs:
+        if k == key:
+            batch = logs[key]
+            if np_path is None:
+                for x in batch:
+                    img = custom_to_pil(x)
+                    imgpath = os.path.join(path, f"{key}_{n_saved:06}.png")
+                    img.save(imgpath)
+                    n_saved += 1
+            else:
+                npbatch = custom_to_np(batch)
+                shape_str = "x".join([str(x) for x in npbatch.shape])
+                nppath = os.path.join(np_path, f"{n_saved}-{shape_str}-samples.npz")
+                np.savez(nppath, npbatch)
+                n_saved += npbatch.shape[0]
+    return n_saved
+
+
+def get_parser():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "-r",
+        "--resume",
+        type=str,
+        nargs="?",
+        help="load from logdir or checkpoint in logdir",
+    )
+    parser.add_argument(
+        "-n",
+        "--n_samples",
+        type=int,
+        nargs="?",
+        help="number of samples to draw",
+        default=50000
+    )
+    parser.add_argument(
+        "-e",
+        "--eta",
+        type=float,
+        nargs="?",
+        help="eta for ddim sampling (0.0 yields deterministic sampling)",
+        default=1.0
+    )
+    parser.add_argument(
+        "-v",
+        "--vanilla_sample",
+        default=False,
+        action='store_true',
+        help="vanilla sampling (default option is DDIM sampling)?",
+    )
+    parser.add_argument(
+        "-l",
+        "--logdir",
+        type=str,
+        nargs="?",
+        help="extra logdir",
+        default="none"
+    )
+    parser.add_argument(
+        "-c",
+        "--custom_steps",
+        type=int,
+        nargs="?",
+        help="number of steps for ddim and fastdpm sampling",
+        default=50
+    )
+    parser.add_argument(
+        "--batch_size",
+        type=int,
+        nargs="?",
+        help="the bs",
+        default=10
+    )
+    return parser
+
+
+def load_model_from_config(config, sd):
+    model = instantiate_from_config(config)
+    model.load_state_dict(sd,strict=False)
+    model.cuda()
+    model.eval()
+    return model
+
+
+def load_model(config, ckpt, gpu, eval_mode):
+    if ckpt:
+        print(f"Loading model from {ckpt}")
+        pl_sd = torch.load(ckpt, map_location="cpu")
+        global_step = pl_sd["global_step"]
+    else:
+        pl_sd = {"state_dict": None}
+        global_step = None
+    model = load_model_from_config(config.model,
+                                   pl_sd["state_dict"])
+
+    return model, global_step
+
+
+if __name__ == "__main__":
+    now = datetime.datetime.now().strftime("%Y-%m-%d-%H-%M-%S")
+    sys.path.append(os.getcwd())
+    command = " ".join(sys.argv)
+
+    parser = get_parser()
+    opt, unknown = parser.parse_known_args()
+    ckpt = None
+
+    if not os.path.exists(opt.resume):
+        raise ValueError("Cannot find {}".format(opt.resume))
+    if os.path.isfile(opt.resume):
+        # paths = opt.resume.split("/")
+        try:
+            logdir = '/'.join(opt.resume.split('/')[:-1])
+            # idx = len(paths)-paths[::-1].index("logs")+1
+            print(f'Logdir is {logdir}')
+        except ValueError:
+            paths = opt.resume.split("/")
+            idx = -2  # take a guess: path/to/logdir/checkpoints/model.ckpt
+            logdir = "/".join(paths[:idx])
+        ckpt = opt.resume
+    else:
+        assert os.path.isdir(opt.resume), f"{opt.resume} is not a directory"
+        logdir = opt.resume.rstrip("/")
+        ckpt = os.path.join(logdir, "model.ckpt")
+
+    base_configs = sorted(glob.glob(os.path.join(logdir, "config.yaml")))
+    opt.base = base_configs
+
+    configs = [OmegaConf.load(cfg) for cfg in opt.base]
+    cli = OmegaConf.from_dotlist(unknown)
+    config = OmegaConf.merge(*configs, cli)
+
+    gpu = True
+    eval_mode = True
+
+    if opt.logdir != "none":
+        locallog = logdir.split(os.sep)[-1]
+        if locallog == "": locallog = logdir.split(os.sep)[-2]
+        print(f"Switching logdir from '{logdir}' to '{os.path.join(opt.logdir, locallog)}'")
+        logdir = os.path.join(opt.logdir, locallog)
+
+    print(config)
+
+    model, global_step = load_model(config, ckpt, gpu, eval_mode)
+    print(f"global step: {global_step}")
+    print(75 * "=")
+    print("logging to:")
+    logdir = os.path.join(logdir, "samples", f"{global_step:08}", now)
+    imglogdir = os.path.join(logdir, "img")
+    numpylogdir = os.path.join(logdir, "numpy")
+
+    os.makedirs(imglogdir)
+    os.makedirs(numpylogdir)
+    print(logdir)
+    print(75 * "=")
+
+    # write config out
+    sampling_file = os.path.join(logdir, "sampling_config.yaml")
+    sampling_conf = vars(opt)
+
+    with open(sampling_file, 'w') as f:
+        yaml.dump(sampling_conf, f, default_flow_style=False)
+    print(sampling_conf)
+
+
+    run(model, imglogdir, eta=opt.eta,
+        vanilla=opt.vanilla_sample,  n_samples=opt.n_samples, custom_steps=opt.custom_steps,
+        batch_size=opt.batch_size, nplog=numpylogdir)
+
+    print("done.")

scripts/stable_txt2img.py

@@ -0,0 +1,292 @@
+import argparse, os, sys, glob
+import torch
+import numpy as np
+from omegaconf import OmegaConf
+from PIL import Image
+from tqdm import tqdm, trange
+from itertools import islice
+from einops import rearrange
+from torchvision.utils import make_grid, save_image
+import time
+from pytorch_lightning import seed_everything
+from torch import autocast
+from contextlib import contextmanager, nullcontext
+
+from ldm.util import instantiate_from_config
+from ldm.models.diffusion.ddim import DDIMSampler
+from ldm.models.diffusion.plms import PLMSSampler
+
+
+def chunk(it, size):
+    it = iter(it)
+    return iter(lambda: tuple(islice(it, size)), ())
+
+
+def load_model_from_config(config, ckpt, verbose=False):
+    print(f"Loading model from {ckpt}")
+    pl_sd = torch.load(ckpt, map_location="cpu")
+    if "global_step" in pl_sd:
+        print(f"Global Step: {pl_sd['global_step']}")
+    sd = pl_sd["state_dict"]
+    model = instantiate_from_config(config.model)
+    m, u = model.load_state_dict(sd, strict=False)
+    if len(m) > 0 and verbose:
+        print("missing keys:")
+        print(m)
+    if len(u) > 0 and verbose:
+        print("unexpected keys:")
+        print(u)
+
+    model.cuda()
+    model.eval()
+    return model
+
+
+def main():
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--prompt",
+        type=str,
+        nargs="?",
+        default="a painting of a virus monster playing guitar",
+        help="the prompt to render"
+    )
+    parser.add_argument(
+        "--outdir",
+        type=str,
+        nargs="?",
+        help="dir to write results to",
+        default="outputs/txt2img-samples"
+    )
+    parser.add_argument(
+        "--skip_grid",
+        action='store_true',
+        help="do not save a grid, only individual samples. Helpful when evaluating lots of samples",
+    )
+    parser.add_argument(
+        "--skip_save",
+        action='store_true',
+        help="do not save individual samples. For speed measurements.",
+    )
+    parser.add_argument(
+        "--ddim_steps",
+        type=int,
+        default=50,
+        help="number of ddim sampling steps",
+    )
+    parser.add_argument(
+        "--plms",
+        action='store_true',
+        help="use plms sampling",
+    )
+    parser.add_argument(
+        "--laion400m",
+        action='store_true',
+        help="uses the LAION400M model",
+    )
+    parser.add_argument(
+        "--fixed_code",
+        action='store_true',
+        help="if enabled, uses the same starting code across samples ",
+    )
+    parser.add_argument(
+        "--ddim_eta",
+        type=float,
+        default=0.0,
+        help="ddim eta (eta=0.0 corresponds to deterministic sampling",
+    )
+    parser.add_argument(
+        "--n_iter",
+        type=int,
+        default=2,
+        help="sample this often",
+    )
+    parser.add_argument(
+        "--H",
+        type=int,
+        default=512,
+        help="image height, in pixel space",
+    )
+    parser.add_argument(
+        "--W",
+        type=int,
+        default=512,
+        help="image width, in pixel space",
+    )
+    parser.add_argument(
+        "--C",
+        type=int,
+        default=4,
+        help="latent channels",
+    )
+    parser.add_argument(
+        "--f",
+        type=int,
+        default=8,
+        help="downsampling factor",
+    )
+    parser.add_argument(
+        "--n_samples",
+        type=int,
+        default=3,
+        help="how many samples to produce for each given prompt. A.k.a. batch size",
+    )
+    parser.add_argument(
+        "--n_rows",
+        type=int,
+        default=0,
+        help="rows in the grid (default: n_samples)",
+    )
+    parser.add_argument(
+        "--scale",
+        type=float,
+        default=7.5,
+        help="unconditional guidance scale: eps = eps(x, empty) + scale * (eps(x, cond) - eps(x, empty))",
+    )
+    parser.add_argument(
+        "--from-file",
+        type=str,
+        help="if specified, load prompts from this file",
+    )
+    parser.add_argument(
+        "--config",
+        type=str,
+        default="configs/stable-diffusion/v1-inference.yaml",
+        help="path to config which constructs model",
+    )
+    parser.add_argument(
+        "--ckpt",
+        type=str,
+        default="models/ldm/stable-diffusion-v1/model.ckpt",
+        help="path to checkpoint of model",
+    )    
+    parser.add_argument(
+        "--seed",
+        type=int,
+        default=42,
+        help="the seed (for reproducible sampling)",
+    )
+    parser.add_argument(
+        "--precision",
+        type=str,
+        help="evaluate at this precision",
+        choices=["full", "autocast"],
+        default="autocast"
+    )
+
+
+    parser.add_argument(
+        "--embedding_path", 
+        type=str, 
+        help="Path to a pre-trained embedding manager checkpoint")
+
+    opt = parser.parse_args()
+
+    if opt.laion400m:
+        print("Falling back to LAION 400M model...")
+        opt.config = "configs/latent-diffusion/txt2img-1p4B-eval.yaml"
+        opt.ckpt = "models/ldm/text2img-large/model.ckpt"
+        opt.outdir = "outputs/txt2img-samples-laion400m"
+
+    seed_everything(opt.seed)
+
+    config = OmegaConf.load(f"{opt.config}")
+    model = load_model_from_config(config, f"{opt.ckpt}")
+    #model.embedding_manager.load(opt.embedding_path)
+
+    device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
+    model = model.to(device)
+
+    if opt.plms:
+        sampler = PLMSSampler(model)
+    else:
+        sampler = DDIMSampler(model)
+
+    os.makedirs(opt.outdir, exist_ok=True)
+    outpath = opt.outdir
+
+    batch_size = opt.n_samples
+    n_rows = opt.n_rows if opt.n_rows > 0 else batch_size
+    if not opt.from_file:
+        prompt = opt.prompt
+        assert prompt is not None
+        data = [batch_size * [prompt]]
+
+    else:
+        print(f"reading prompts from {opt.from_file}")
+        with open(opt.from_file, "r") as f:
+            data = f.read().splitlines()
+            data = list(chunk(data, batch_size))
+
+    sample_path = os.path.join(outpath, "samples")
+    os.makedirs(sample_path, exist_ok=True)
+    base_count = len(os.listdir(sample_path))
+    grid_count = len(os.listdir(outpath)) - 1
+
+    start_code = None
+    if opt.fixed_code:
+        start_code = torch.randn([opt.n_samples, opt.C, opt.H // opt.f, opt.W // opt.f], device=device)
+
+    precision_scope = autocast if opt.precision=="autocast" else nullcontext
+    with torch.no_grad():
+        with precision_scope("cuda"):
+            with model.ema_scope():
+                tic = time.time()
+                all_samples = list()
+                for n in trange(opt.n_iter, desc="Sampling"):
+                    for prompts in tqdm(data, desc="data"):
+                        uc = None
+                        if opt.scale != 1.0:
+                            uc = model.get_learned_conditioning(batch_size * [""])
+                        if isinstance(prompts, tuple):
+                            prompts = list(prompts)
+                        c = model.get_learned_conditioning(prompts)
+                        shape = [opt.C, opt.H // opt.f, opt.W // opt.f]
+                        samples_ddim, _ = sampler.sample(S=opt.ddim_steps,
+                                                         conditioning=c,
+                                                         batch_size=opt.n_samples,
+                                                         shape=shape,
+                                                         verbose=False,
+                                                         unconditional_guidance_scale=opt.scale,
+                                                         unconditional_conditioning=uc,
+                                                         eta=opt.ddim_eta,
+                                                         x_T=start_code)
+
+                        x_samples_ddim = model.decode_first_stage(samples_ddim)
+                        x_samples_ddim = torch.clamp((x_samples_ddim + 1.0) / 2.0, min=0.0, max=1.0)
+
+                        if not opt.skip_save:
+                            for x_sample in x_samples_ddim:
+                                x_sample = 255. * rearrange(x_sample.cpu().numpy(), 'c h w -> h w c')
+                                Image.fromarray(x_sample.astype(np.uint8)).save(
+                                    os.path.join(sample_path, f"{base_count:05}.jpg"))
+                                base_count += 1
+
+                        if not opt.skip_grid:
+                            all_samples.append(x_samples_ddim)
+
+                if not opt.skip_grid:
+                    # additionally, save as grid
+                    grid = torch.stack(all_samples, 0)
+                    grid = rearrange(grid, 'n b c h w -> (n b) c h w')
+                    
+                    for i in range(grid.size(0)):
+                        save_image(grid[i, :, :, :], os.path.join(outpath,opt.prompt+'_{}.png'.format(i)))
+                    grid = make_grid(grid, nrow=n_rows)
+
+                    # to image
+                    grid = 255. * rearrange(grid, 'c h w -> h w c').cpu().numpy()
+                    Image.fromarray(grid.astype(np.uint8)).save(os.path.join(outpath, f'{prompt.replace(" ", "-")}-{grid_count:04}.jpg'))
+                    grid_count += 1
+                    
+                    
+
+                toc = time.time()
+
+    print(f"Your samples are ready and waiting for you here: \n{outpath} \n"
+          f" \nEnjoy.")
+
+
+if __name__ == "__main__":
+    main()

scripts/txt2img.py

@@ -0,0 +1,184 @@
+import argparse, os, sys, glob
+import torch
+import numpy as np
+from omegaconf import OmegaConf
+from PIL import Image
+from tqdm import tqdm, trange
+from einops import rearrange
+from torchvision.utils import make_grid, save_image
+
+from ldm.util import instantiate_from_config
+from ldm.models.diffusion.ddim import DDIMSampler
+from ldm.models.diffusion.plms import PLMSSampler
+
+def load_model_from_config(config, ckpt, verbose=False):
+    print(f"Loading model from {ckpt}")
+    pl_sd = torch.load(ckpt, map_location="cpu")
+    sd = pl_sd["state_dict"]
+    model = instantiate_from_config(config.model)
+    m, u = model.load_state_dict(sd, strict=False)
+    if len(m) > 0 and verbose:
+        print("missing keys:")
+        print(m)
+    if len(u) > 0 and verbose:
+        print("unexpected keys:")
+        print(u)
+
+    model.cuda()
+    model.eval()
+    return model
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--prompt",
+        type=str,
+        nargs="?",
+        default="a painting of a virus monster playing guitar",
+        help="the prompt to render"
+    )
+
+    parser.add_argument(
+        "--outdir",
+        type=str,
+        nargs="?",
+        help="dir to write results to",
+        default="outputs/txt2img-samples"
+    )
+    parser.add_argument(
+        "--ddim_steps",
+        type=int,
+        default=200,
+        help="number of ddim sampling steps",
+    )
+
+    parser.add_argument(
+        "--plms",
+        action='store_true',
+        help="use plms sampling",
+    )
+
+    parser.add_argument(
+        "--ddim_eta",
+        type=float,
+        default=0.0,
+        help="ddim eta (eta=0.0 corresponds to deterministic sampling",
+    )
+    parser.add_argument(
+        "--n_iter",
+        type=int,
+        default=1,
+        help="sample this often",
+    )
+
+    parser.add_argument(
+        "--H",
+        type=int,
+        default=256,
+        help="image height, in pixel space",
+    )
+
+    parser.add_argument(
+        "--W",
+        type=int,
+        default=256,
+        help="image width, in pixel space",
+    )
+
+    parser.add_argument(
+        "--n_samples",
+        type=int,
+        default=4,
+        help="how many samples to produce for the given prompt",
+    )
+
+    parser.add_argument(
+        "--scale",
+        type=float,
+        default=5.0,
+        help="unconditional guidance scale: eps = eps(x, empty) + scale * (eps(x, cond) - eps(x, empty))",
+    )
+
+    parser.add_argument(
+        "--ckpt_path", 
+        type=str, 
+        default="/data/pretrained_models/ldm/text2img-large/model.ckpt", 
+        help="Path to pretrained ldm text2img model")
+
+    parser.add_argument(
+        "--embedding_path", 
+        type=str, 
+        help="Path to a pre-trained embedding manager checkpoint")
+
+    opt = parser.parse_args()
+
+
+    config = OmegaConf.load("configs/latent-diffusion/txt2img-1p4B-eval_with_tokens.yaml")  # TODO: Optionally download from same location as ckpt and chnage this logic
+    model = load_model_from_config(config, opt.ckpt_path)  # TODO: check path
+    #model.embedding_manager.load(opt.embedding_path)
+
+    device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
+    model = model.to(device)
+
+    if opt.plms:
+        sampler = PLMSSampler(model)
+    else:
+        sampler = DDIMSampler(model)
+
+    os.makedirs(opt.outdir, exist_ok=True)
+    outpath = opt.outdir
+
+    prompt = opt.prompt
+
+
+    sample_path = os.path.join(outpath, "samples")
+    os.makedirs(sample_path, exist_ok=True)
+    base_count = len(os.listdir(sample_path))
+
+    all_samples=list()
+    with torch.no_grad():
+        with model.ema_scope():
+            uc = None
+            if opt.scale != 1.0:
+                uc = model.get_learned_conditioning(opt.n_samples * [""])
+            for n in trange(opt.n_iter, desc="Sampling"):
+                c = model.get_learned_conditioning(opt.n_samples * [prompt])
+                shape = [4, opt.H//8, opt.W//8]
+                samples_ddim, _ = sampler.sample(S=opt.ddim_steps,
+                                                 conditioning=c,
+                                                 batch_size=opt.n_samples,
+                                                 shape=shape,
+                                                 verbose=False,
+                                                 unconditional_guidance_scale=opt.scale,
+                                                 unconditional_conditioning=uc,
+                                                 eta=opt.ddim_eta)
+
+                x_samples_ddim = model.decode_first_stage(samples_ddim)
+                x_samples_ddim = torch.clamp((x_samples_ddim+1.0)/2.0, min=0.0, max=1.0)
+
+                for x_sample in x_samples_ddim:
+                    x_sample = 255. * rearrange(x_sample.cpu().numpy(), 'c h w -> h w c')
+                    Image.fromarray(x_sample.astype(np.uint8)).save(os.path.join(sample_path, f"{base_count:04}.jpg"))
+                    base_count += 1
+                all_samples.append(x_samples_ddim)
+
+
+    # additionally, save as grid
+    grid = torch.stack(all_samples, 0)
+    grid = rearrange(grid, 'n b c h w -> (n b) c h w')
+    
+    for i in range(grid.size(0)):
+           save_image(grid[i, :, :, :], os.path.join(outpath,opt.prompt+'_{}.png'.format(i)))
+           
+    grid = make_grid(grid, nrow=opt.n_samples)
+    
+
+    # to image
+    grid = 255. * rearrange(grid, 'c h w -> h w c').cpu().numpy()
+    Image.fromarray(grid.astype(np.uint8)).save(os.path.join(outpath, f'{prompt.replace(" ", "-")}.jpg'))
+    
+    
+
+    print(f"Your samples are ready and waiting four you here: \n{outpath} \nEnjoy.")

setup.py

@@ -0,0 +1,13 @@
+from setuptools import setup, find_packages
+
+setup(
+    name='latent-diffusion',
+    version='0.0.1',
+    description='',
+    packages=find_packages(),
+    install_requires=[
+        'torch',
+        'numpy',
+        'tqdm',
+    ],
+)