Cross Modal Focal Loss for RGBD Face Anti-Spoofing

This package is part of the Bob toolkit and it allows to reproduce the experimental results published in the following paper:

@inproceedings{GeorgeCVPR2021,
    author = {Anjith George, Sebastien Marcel},
    title = {Cross Modal Focal Loss for RGBD Face Anti-Spoofing},
    year = {2021},
    booktitle = {IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
}

If you use this package and/or its results, please cite the paper [GM21].

The main idea in the paper the multi-head architecture and the use of the proposed Cross-Modal Focal Loss to improve the performance of multi-channel CNN models.

Fig. 1 Diagram of the proposed framework (The architecture is referred to as RGBD-MH).

User guide

• Training CNN using Cross Modal Focal Loss
  • Preprocessing data
  • Training the CNN Model with CMFL
  • Running experiments with the trained model
  • Evaluating results
• Using pretrained models
• References