ngoanpv / ECCV20-STDN-1

Source code for ECCV 2020 paper: On Disentangling Spoof Trace for Generic Face Anti-Spoofing

On Disentangling Spoof Traces for Generic Face Anti-Spoofing

Yaojie Liu, Joel Stehouwer, Xiaoming Liu

[Paper] [Project]

Setup

Tested on Python 3.6 & Tensorflow 1.13.0. As it uses the contrib package, the code should work with Tensorflow >1.8.0 and <1.13.0. The code should be easy to transfer to keras package.

Training

First create a folder "./log/". To run the training code:

python train.py

The face-anti-spoofing databases (e.g. SiW-M, SiW, and Oulu-NPU) have to be applied separately. We provide the train/test split we used for SiW-M protocol I. We provide some data samples to illustrate the data format and structure requirement in "./data/" folder. The video should pre-processed into frames of cropped face, saved in one folder under either "live/" or "spoof/". The landmark (68) should be provided in the "XXX.npy" file.

Testing

To run the testing code:

python test.py

It saves the scores in "./log/XXX/test/score.txt" file.

Acknowledge

If you find this code useful, please cite the paper:

@inproceedings{eccv20yaojie,
    title={On Disentangling Spoof Traces for Generic Face Anti-Spoofing},
    author={Yaojie Liu, Joel Stehouwer, Xiaoming Liu},
    booktitle={In Proceeding of European Conference on Computer Vision (ECCV 2020)},
    address={Virtual},
    year={2020}
}

If you have any question, please contact: Yaojie Liu