RikuYokoo/Council-GAN

Implementation for our paper Breaking the Cycle - Colleagues are all you need (CVPR 2020)

Council-GAN

Implementation of our paper Breaking the Cycle - Colleagues are all you need (CVPR 2020)

Paper

Ori Nizan , Ayellet Tal, Breaking the Cycle - Colleagues are all you need [Project]

Temporary Telegram Bot

Send image to this telegram bot and it will send you back its female translation using our implementation

Usage

Install requirements

conda env create -f conda_requirements.yml

Downloading the dataset

Download the selfie to anime dataset:

bash ./scripts/download.sh U_GAT_IT_selfie2anime

Download the celeba glasses removal dataset:

bash ./scripts/download.sh celeba_glasses_removal

Download the celeba male to female dataset:

bash ./scripts/download.sh celeba_male2female

use your on dataset:

├──datasets
    └──DATASET_NAME
        ├──testA
            ├──im1.png
            ├──im2.png
            └── ...
        ├──testB
            ├──im3.png
            ├──im4.png
            └── ...
        ├──trainA
            ├──im5.png
            ├──im6.png
            └── ...
        └──trainB
            ├──im7.png
            ├──im8.png
            └── ...

and change the data_root attribute to ./datasets/DATASET_NAME in the yaml file

Training:

Selfie to anime:

python train.py --config configs/anime2face_council_folder.yaml --output_path ./outputs/council_anime2face_256_256 --resume

Glasses removel:

python train.py --config configs/galsses_council_folder.yaml --output_path ./outputs/council_glasses_128_128 --resume

Male to female:

python train.py --config configs/male2female_council_folder.yaml --output_path ./outputs/male2famle_256_256 --resume

Testing:

for converting all the images in input_folder using all the members in the council:

python test_on_folder.py --config configs/anime2face_council_folder.yaml --output_folder ./outputs/council_anime2face_256_256 --checkpoint ./outputs/council_anime2face_256_256/anime2face_council_folder/checkpoints/01000000 --input_folder ./datasets/selfie2anime/testB --a2b 0

or using spsified memeber:

python test_on_folder.py --config configs/anime2face_council_folder.yaml --output_folder ./outputs/council_anime2face_256_256 --checkpoint ./outputs/council_anime2face_256_256/anime2face_council_folder/checkpoints/b2a_gen_3_01000000.pt --input_folder ./datasets/selfie2anime/testB --a2b 0

Download Pretrain Models

Download pretrain male to female model:

bash ./scripts/download.sh pretrain_male_to_female

Then to convert images in --input_folder run:

python test_on_folder.py --config pretrain/m2f/256/male2female_council_folder.yaml --output_folder ./outputs/male2famle_256_256 --checkpoint pretrain/m2f/256/01000000 --input_folder ./datasets/celeba_male2female/testA --a2b 1

Download pretrain glasses removal model:

bash ./scripts/download.sh pretrain_glasses_removal

Then to convert images in --input_folder run:

python test_on_folder.py --config pretrain/glasses_removal/128/galsses_council_folder.yaml --output_folder ./outputs/council_glasses_128_128 --checkpoint pretrain/glasses_removal/128/01000000 --input_folder ./datasets/glasses/testA --a2b 1

Download pretrain selfie to anime model:

bash ./scripts/download.sh pretrain_selfie_to_anime

Then to convert images in --input_folder run:

python test_on_folder.py --config pretrain/anime/256/anime2face_council_folder.yaml --output_folder ./outputs/council_anime2face_256_256 --checkpoint pretrain/anime/256/01000000 --input_folder ./datasets/selfie2anime/testB --a2b 0

Test GUI:

test GUI on pretrain model:

male2female

python test_gui.py --config pretrain/m2f/128/male2female_council_folder.yaml --checkpoint pretrain/m2f/128/a2b_gen_0_01000000.pt --a2b 1

glasses Removal

python test_gui.py --config pretrain/glasses_removal/128/galsses_council_folder.yaml --checkpoint pretrain/glasses_removal/128/a2b_gen_3_01000000.pt --a2b 1

selfie2anime

python test_gui.py --config pretrain/anime/256/anime2face_council_folder.yaml --checkpoint pretrain/anime/256/b2a_gen_3_01000000.pt --a2b 0

Citation

@inproceedings{nizan2020council,
  title={Breaking the Cycle - Colleagues are all you need},
  author={Ori Nizan and Ayellet Tal},
  booktitle={IEEE conference on computer vision and pattern recognition (CVPR)},
  year={2020}
}

Acknowledgement

In this work we based our code on MUNIT implementation. Please cite the original MUNIT if you use their part of the code.