search

YuchenLiu98 / ECCV2022-SDA-FAS

Source-Free Domain Adaptation with Contrastive Domain Alignment and Self-supervised Exploration for Face Anti-Spoofing, ECCV2022
MIT License
19 stars 2 forks source link

readme

ECCV2022-SDA-FAS

The implementation of "Source-Free Domain Adaptation with Contrastive Domain Alignment and Self-supervised Exploration for Face Anti-Spoofing", ECCV2022.

The motivation of our proposed SDA-FAS:

The framework of our proposed SDA-FAS:

Congifuration Environment

Data Preparation

Dataset

Download the OULU-NPU, CASIA-FASD, Idiap Replay-Attack, MSU-MFSD, and CelebA-Spoof datasets.

Data Pre-processing.

MTCNN is used for face detection and alignment. All the cropped faces are resized as (256,256,3).

Data Organization

└── Data_Dir
   ├── OULU_NPU
   ├── CASIA_MFSD
   ├── REPLAY_ATTACK
   ├── MSU_MFSD
   ├── CelebA-Spoof
   └── ...

Training

Move to the folder $root/SDA-FAS/experiment/testing_scenarios/ and run:

python train_SDAFAS.py

Testing

Move to the folder $root/SDA-FAS/experiment/testing_scenarios/ and run:

python test_SDAFAS.py

Trained Models

Scenarios HTER(%) AUC(%) Trained models
O&C&I to M 5.00 96.60 model
O&M&I to C 2.40 99.42 model
O&C&M to I 2.25 99.64 model
I&C&M to O 5.07 99.00 model

Access code for Baidu is sdaf

Citation

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


@inproceedings{liu2022source,
    author = {Liu, Yuchen and Chen, Yabo and Dai, Wenrui and Gou, Mengran and Huang, Chun-Ting and Xiong, Hongkai},
    title = {Source-Free Domain Adaptation with Contrastive Domain Alignment and Self-supervised Exploration for Face Anti-Spoofing},
    booktitle = {ECCV},
    year = {2022}
}