This repo provides PyTorch implementation of the paper One-Class Knowledge Distillation for Face Presentation Attack Detection to appear on IEEE Transactions on Information Forensics & Security (TIFS).
-
Please request and download the datasets. You may use the following address:
๐ NTU ROSE-YOUTU
๐ CASIA FASD
๐ IDIAP REPLAY-ATTACK
๐ MSU MFSD
๐ OULU-NPU
-
Please install dlib(19.24.0) and opencv(4.5.5) in your anaconda environment.
-
Please download the pretrained shape predictor model from here.
-
Please use preprocessing.py to get face images from videos. The prepocessed data for client-specific one-class domain adaptation setting are available here.
-
Please find the data division of the challenging experimental setting here.
๐ Please use the example script train_teacher.py to train the Teacher Network.
๐ Please use the example script train_student.py to train the Student Network.
๐ You may use the example script evaluation_student.py to evalute the pretrained model.
๐ Everything in this repo can NOT be used for commercial purpose.
๐ If you have any questions, feel free to open an issue or contact me via email.
๐ The implementation of sparse learning in our codes is based on library.
๐ If you use this repo in your work, please use the following citation.
@ARTICLE{9782427, author={Li, Zhi and Cai, Rizhao and Li, Haoliang and Lam, Kwok-Yan and Hu, Yongjian and Kot, Alex C.}, journal={IEEE Transactions on Information Forensics and Security}, title={One-Class Knowledge Distillation for Face Presentation Attack Detection}, year={2022}, volume={}, number={}, pages={1-1}, doi={10.1109/TIFS.2022.3178240} }