Awesome
Identity-Preserving Face Swapping via Dual Surrogate Generative Models
<a href='https://bone-11.github.io/cs-cs/'><img src='https://img.shields.io/badge/Project-Page-Green'></a>
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:
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.