usdan-pr's Introduction

USDAN

The implementation of Unified unsupervised and semi-supervised domain adaptation network for cross-scenario face anti-spoofing, which is accepted by Pattern Recognition 2021.

An overview of the proposed USDAN method:

Congifuration Environment

python 3.6
pytorch 0.4
torchvision 0.2
cuda 8.0

Pre-training

Dataset.

Download the CASIA-FASD, Idiap Replay-Attack, and MSU-MFSD datasets.

Data Pre-processing.

MTCNN algorithm is utilized for face detection and face alignment. All the detected faces are normalized to 224$\times$224$\times$3, where only RGB channels are utilized for training. The exact codes that we used can be found here.

Put the processed frames in the path $root/processed_data

To be specific, we first utilize the MTCNN algorithm to process every frame of each video. And then, we utilize the get_files function in the utils/utils.py to sample frames during training. Finally, the information of selected frames are saved to the choose_*.json file.

Data Label Generation.

Move to the $root/USDAN_*/msu_casia/data_label/ and generate the data label list:

python generate_label.py

Training

Move to the folder $root/USDAN_*/msu_casia/ and just run like this:

python train_USDAN_*.py

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

Testing

Run like this:

python da_test.py

Citation

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

@InProceedings{Jia_2021_PR_USDAN,
    author = {Yunpei Jia and Jie Zhang and Shiguang Shan and Xilin Chen},
    title = {Unified Unsupervised and Semi-supervised Domain Adaptation Network for Cross-scenario Face Anti-spoofing},
    booktitle = {Pattern Recognition},
    year = {2021}
}

usdan-pr's People