Identity-Preserving Face Swapping via Dual Surrogate Generative Models

 

This is the repository of the paper Identity-Preserving Face Swapping via Dual Surrogate Generative Models. For now, we upload the inference code and checkpoint.

Getting Started

Environment

pip install -r requirements.txt

Then download ID encoder weight ms1mv3_arcface_r100_fp16_backbone.pth from:

https://onedrive.live.com/?id=4A83B6B633B029CC!5577&resid=4A83B6B633B029CC!5577&authkey=!AFZjr283nwZHqbA&cid=4a83b6b633b029cc

and should be placed in ./model/arcface/

Inference Checkpoints

You can download the checkpoints from [https://1drv.ms/f/c/64d71f39113d98e4/ElBkLV2YQXdHgJbsc2Aboy8BBhhvct14hvW8sGD87F2Nzg?e=U2Yqxj] and place them at ./.

Inference

Before swapping, use facealign.sh to align the face images.

After alignment, inference_adapter.sh is utilized to swapping

bash facealign.sh
bash inference_adapter.sh

Training

Download the training data from [https://1drv.ms/f/c/64d71f39113d98e4/El8ChUj0d5BIk5yMGkiyR8kB450SvhZYY6d4sm5sksZIeA?e=p4Dk8T] and place them at ./train_data. Then run the following scirpt

bash train_adapter.sh

And the results can be found in ./expr/train_smswap_faceshifter_adapter.

License and Citation

CSCS is released only for academic research. Researchers are allowed to use this code and weights freely for non-commercial purposes.

Reference Format:

@article{huang2024cscs,
  title={Identity-Preserving Face Swapping via Dual Surrogate Generative Models},
  author={Huang, Ziyao and Tang, Fan and Zhang, Yong and Cao, Juan and Li, Chengyu and Tang, Sheng and Li, Jintao and Lee, Tong-Yee},
  journal={ACM Transactions on Graphics},
  volume={43},
  number={5},
  pages={1--19},
  year={2024},
  publisher={ACM New York, NY, USA}
}