DGUA-FAS

The implementation of Domain-Generalized Face Anti-Spoofing with Unknown Attacks

The architecture of the proposed DGUA-FAS method:

Congifuration Environment

• python 3.9.12
• torch 1.10.0
• torchvision 0.11.1
• cuda 12.0

Requirements

Install MobileViT and modify base_cls.py to our version

git clone https://github.com/apple/ml-cvnets
cd ml-cvnets
git checkout 84d992f413e52c0468f86d23196efd9dad885e6f

# replace ./cvnets/models/classification/base_cls.py to our version
pip install -r requirements.txt
pip install --editable .
pip install pandas
pip install tensorboard
cd ..

Our data Pre-processing is like SSDG, so please ref their dataset setting.

# After setting up the dataset path, run below codes.
cd ./data_label
python generate_label.py

Training

cd ./experiment/m/
python train.py

The file config.py contains all the hype-parameters used during training.

Testing

Change the name of testing dataset in config.py and run like this:

python dg_test.py

We also provide our pretrained model [Google drive]

Acknowledgment

This work can not be finished well without the following reference, many thanks for the author's contribution:

SSDG, ml-cvnets, DiVT

Citation

Please cite our works if the code is helpful to your research.

@INPROCEEDINGS{10223078,
  author={Hong, Zong-Wei and Lin, Yu-Chen and Liu, Hsuan-Tung and Yeh, Yi-Ren and Chen, Chu-Song},
  booktitle={2023 IEEE International Conference on Image Processing (ICIP)},
  title={Domain-Generalized Face Anti-Spoofing with Unknown Attacks},
  year={2023},
  volume={},
  number={},
  pages={820-824},
  doi={10.1109/ICIP49359.2023.10223078}}