This repository reproduces image editing methods using diffusion inversion methods.
Keywords: Diffusion Model, Image Inversion, Image Editing
- p2p_requirements.txt: for models in
run_editing_p2p.py - masactrl_requirements.txt: for models in
run_editing_masactrl.py - pnp_requirements.txt: for models in
run_editing_pnp.py
For example, if you want to use the models in run_editing_p2p.py, you need to install the environment as follows:
conda create -n p2p python=3.9 -y
conda activate p2p
conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.3 -c pytorch
pip install -r environment/p2p_requirements.txtpython run_editing_p2p.py \
--data_path "/home/poong/junseok/PIE_Bench" \
--output_path "/home/poong/junseok/evaluation/PIE_Bench_output" \
--edit_category_list 0 1 2 3 4 5 6 7 8 9 \
--edit_method_list "directinversion+p2p"python evaluation/evaluate.py \
--annotation_mapping_file "/home/poong/junseok/PIE_Bench/mapping_file.json" \
--metrics structure_distance psnr_unedit_part lpips_unedit_part mse_unedit_part ssim_unedit_part clip_similarity_source_image clip_similarity_target_image clip_similarity_target_image_edit_part \
--src_image_folder "/home/poong/junseok/PIE_Bench/annotation_images" \
--tgt_methods 1_ddim+pnp 1_null-text-inversion+pnp 1_directinversion+pnp 4_edit-friendly-inversion+pnp \
--result_path pnp_results.csv \
--evaluate_whole_tableIf you want average the results, run evaluation/evaluate_final.py
Reproduced quantitative results
- NT : Null-Text Inversion
- EF : Edit-Friendly DDPM Inversion
- Direct : Direct Inversion
| Prompt-to-prompt | Plug-and-Play | MasaCtrl | |
|---|---|---|---|
| Source Prompts | an orange cat sitting on top of a fence | a cute little duck with big eyes | blue light, a black and white cat is playing with a flower |
| Target Prompts | an black cat sitting on top of a fence | a cute little marmot with big eyes | blue light, a black and white dog is playing with a flower |
| DDIM |  |
 |
 |
| NT |  |
 |
 |
| EF |  |
 |
 |
| Direct |  |
 |
 |
Our code is modified on the basis of direct-inversion, thanks to contributors!