Official code for our paper
A Latent Space of Stochastic Diffusion Models for Zero-Shot Image Editing and Guidance
Chen Henry Wu, Fernando De la Torre
Carnegie Mellon University
ICCV 2023
This paper was first on arxiv in Oct, 2022 (link), under the title
Unifying Diffusion Models' Latent Space, with Applications to CycleDiffusion and Guidance
We modified the content and title and submitted it to ICCV 2023. The original version of the code, discussion, and GitHub issues can be found here.
- Create an environment by running
conda env create -f environment.yml
conda activate generative_prompt
pip install git+https://github.com/openai/CLIP.git
- Install
torch
andtorchvision
based on your CUDA version. - Install taming-transformers by running
cd ../
git clone git@github.com:CompVis/taming-transformers.git
cd taming-transformers/
pip install -e .
cd ../
- Set up wandb for logging (registration is required). You should modify the
setup_wandb
function inmain.py
to accomodate your wandb credentials. You may want to run something like
wandb login
- Most data for zero-shot image-to-image translation are already included in data/. Some images are from the AFHQ validation set, detailed below.
- Prepare the AFHQ validation set for unpaired image-to-image translation (also for some images used by zero-shot image-to-image translation) by running
git clone git@github.com:clovaai/stargan-v2.git
cd stargan-v2/
bash download.sh afhq-v2-dataset
- Stable Diffusion
cd ckpts/
mkdir stable_diffusion
cd stable_diffusion/
# Download pre-trained checkpoints for Stable Diffusion here.
# You should download this version: https://huggingface.co/CompVis/stable-diffusion-v-1-4-original
# Due to licence issues, we cannot share the pre-trained checkpoints directly.
- Latent Diffusion Model
cd ckpts/
wget https://www.dropbox.com/s/9lpdgs83l7tjk6c/ldm_models.zip
unzip ldm_models.zip
cd ldm_models/
mkdir text2img-large
cd text2img-large/
wget https://ommer-lab.com/files/latent-diffusion/nitro/txt2img-f8-large/model.ckpt
wget https://www.dropbox.com/s/7pdttimz78ll0km/txt2img-1p4B-eval.yaml
- DDPM (AFHQ-Dog and FFHQ are from ILVR; CelebAHQ is from SDEdit; AFHQ-Cat and -Wild are trained by ourselves)
cd ckpts/
mkdir ddpm
cd ddpm/
# Update Aug 4, 2023: it seems that the link below, originally from SDEdit, is broken. Please find other sources for CelebA-HQ (cf. issue #24)
wget https://image-editing-test-12345.s3-us-west-2.amazonaws.com/checkpoints/celeba_hq.ckpt
wget https://www.dropbox.com/s/g4h8sv07i3hj83d/ffhq_10m.pt
wget https://www.dropbox.com/s/u74w8vaw1f8lc4k/afhq_dog_4m.pt
wget https://www.dropbox.com/s/8i5aznjwdl3b5iq/cat_ema_0.9999_050000.pt
wget https://www.dropbox.com/s/tplximipy8zxaub/wild_ema_0.9999_050000.pt
wget https://www.dropbox.com/s/vqm6bxj0zslrjxv/configs.zip
unzip configs.zip
- Zero-shot image-to-image translation with Stable Diffusion v1-4. We divided the 128 test samples into 8 groups (16 samples in each group), so the averaged metrics are reported.
export CUDA_VISIBLE_DEVICES=0
export RUN_NAME=translate_text2img256_stable_diffusion_stochastic_1
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1405 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 4 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
export CUDA_VISIBLE_DEVICES=1
export RUN_NAME=translate_text2img256_stable_diffusion_stochastic_2
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1424 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 4 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
export CUDA_VISIBLE_DEVICES=2
export RUN_NAME=translate_text2img256_stable_diffusion_stochastic_3
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1423 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 4 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
export CUDA_VISIBLE_DEVICES=3
export RUN_NAME=translate_text2img256_stable_diffusion_stochastic_4
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1422 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 4 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
export CUDA_VISIBLE_DEVICES=4
export RUN_NAME=translate_text2img256_stable_diffusion_stochastic_5
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1429 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 4 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
export CUDA_VISIBLE_DEVICES=5
export RUN_NAME=translate_text2img256_stable_diffusion_stochastic_6
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1428 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 4 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
export CUDA_VISIBLE_DEVICES=6
export RUN_NAME=translate_text2img256_stable_diffusion_stochastic_7
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1427 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 4 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
export CUDA_VISIBLE_DEVICES=7
export RUN_NAME=translate_text2img256_stable_diffusion_stochastic_8
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1426 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 4 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
- Zero-shot image-to-image translation with the LDM text-to-image checkpoint. We divided the 128 test samples into 8 groups, so the averaged metrics are reported.
export CUDA_VISIBLE_DEVICES=0
export RUN_NAME=translate_text2img256_latentdiff_stochastic_1
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1465 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 16 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
export CUDA_VISIBLE_DEVICES=1
export RUN_NAME=translate_text2img256_latentdiff_stochastic_2
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1485 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 16 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
export CUDA_VISIBLE_DEVICES=2
export RUN_NAME=translate_text2img256_latentdiff_stochastic_3
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1486 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 16 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
export CUDA_VISIBLE_DEVICES=3
export RUN_NAME=translate_text2img256_latentdiff_stochastic_4
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1487 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 16 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
export CUDA_VISIBLE_DEVICES=4
export RUN_NAME=translate_text2img256_latentdiff_stochastic_5
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1488 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 16 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
export CUDA_VISIBLE_DEVICES=5
export RUN_NAME=translate_text2img256_latentdiff_stochastic_6
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1489 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 16 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
export CUDA_VISIBLE_DEVICES=6
export RUN_NAME=translate_text2img256_latentdiff_stochastic_7
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1411 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 16 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
export CUDA_VISIBLE_DEVICES=7
export RUN_NAME=translate_text2img256_latentdiff_stochastic_8
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1412 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 16 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
- AFHQ-Cat to AFHQ-Dog with DDIM
$\eta=0.1$
export CUDA_VISIBLE_DEVICES=1
export RUN_NAME=translate_afhqcat256_to_afhqdog256_ddim_eta01
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1446 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 1 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
- AFHQ-Wild to AFHQ-Dog with DDIM
$\eta=0.1$
export CUDA_VISIBLE_DEVICES=5
export RUN_NAME=translate_afhqwild256_to_afhqdog256_ddim_eta01
export SEED=42
nohup python -m torch.distributed.launch --nproc_per_node 1 --master_port 1498 main.py --seed $SEED --cfg experiments/$RUN_NAME.cfg --run_name $RUN_NAME$SEED --logging_strategy steps --logging_first_step true --logging_steps 4 --evaluation_strategy steps --eval_steps 50 --metric_for_best_model CLIPEnergy --greater_is_better false --save_strategy steps --save_steps 50 --save_total_limit 1 --load_best_model_at_end --gradient_accumulation_steps 4 --num_train_epochs 0 --adafactor false --learning_rate 1e-3 --do_eval --output_dir output/$RUN_NAME$SEED --overwrite_output_dir --per_device_train_batch_size 1 --per_device_eval_batch_size 1 --eval_accumulation_steps 4 --ddp_find_unused_parameters true --verbose true > $RUN_NAME$SEED.log 2>&1 &
If you find this repository helpful, please cite it as
@inproceedings{cyclediffusion,
title={Unifying Diffusion Models' Latent Space, with Applications to {CycleDiffusion} and Guidance},
author={Chen Henry Wu and Fernando De la Torre},
booktitle={ArXiv},
year={2022},
}
or
@inproceedings{cyclediffusion,
title={A Latent Space of Stochastic Diffusion Models for Zero-Shot Image Editing and Guidance},
author={Chen Henry Wu and Fernando De la Torre},
booktitle={ICCV},
year={2023},
}
We use the X11 License. This license is identical to the MIT License, but with an extra sentence that prohibits using the copyright holders' names (Carnegie Mellon University in our case) for advertising or promotional purposes without written permission.
Issues are welcome if you have any questions about the code. If you would like to discuss the method, please contact Chen Henry Wu.