depth-face-anti-spoofing

This project is generated by pytorch-template.

Dataset

Method

The idea is that using DQNet model to predict the depth of face and use the mean of the predicted detph as output score. By using depth_face.sh, we can generate the ground truth depth of real face. And the depth of a spoof face is an all zero image. After training DQNet for a few epochs, we add an auxiliary branch to train a binary classification to output real or spoof face.

Reference

• The depth of real face is generated by 3DDFA_V2.
• The DQNet structure is borrowed from Face De-Spoofing: Anti-Spoofing via Noise Modeling.