This is my 3rd year course work in the HSE University.
Weights can be downloaded using download_weights.py or here.
In case using link, make sure that pretrained_ckpts directory looks like below.
pretrained_ckpts/
├── auxiliray (need for training)
│ ├── model_ir_se50.pth
│ └── model.pth
├── e4s
│ └── iteration_100000.pt
├── face_parsing
│ ├── segnext.large.512x512.celebamaskhq.160k.py
│ └── best_mIoU_iter_150000.pth
├── gpen
│ └── weights
│ ├── GPEN-BFR-512.pth
│ ├── ParseNet-latest.pth
│ ├── realesrnet_x4.pth
│ └── RetinaFace-R50.pth
├── stylegan2 (need for training)
│ └── stylegan2-ffhq-config-f.pt
├── shape_predictor_68_face_landmarks.dat
├── vox.pt
├── hopenet_robust_alpha1.pkl (need for metric calculation)
├── WFLW_4HG.pth (need for metric calculation)
└── arcface.pt (need for metric calculation)
python face_swap.py --source=path_to_photos/source --target=path_to_photos/targetFor more information please check options/swap_options.py or face_swap.py -h.
python -m torch.distributed.launch \
--nproc_per_node=4 \
--nnodes=1 \
--node_rank=0 \
--master_addr=localhost \
--master_port=22221 \
train.py --exp_dir='name of your experiment'For more information please check options/train_options.py or train.py -h.
All experiments losses and configs can be founded on wandb.
This repository is not a new method but it is a huge improvements of E4S(CVPR2023). Thanks to the authors for sharing their code.