Face De-Spoofing: Anti-Spoofing via Noise Modeling

Amin Jourabloo*, Yaojie Liu*, Xiaoming Liu

alt text

Setup

Install the Tensorflow >=1.1, <2.0.

The source code files:

  1. "Architecture.py": Contains the architectures and the definitions of the loss functions.
  2. "data_train.py" : Contains the functions for reading the training data.
  3. "Train.py" : The main training file that read the training data, computes the loss functions and backpropagates error.
  4. "facepad-test.py": It performs the testing on the test videos and generates the score for each frame.

Training

To run the training code: source ~/tensorflow/bin/activate python /data/train_demo/code/Train.py deactivate

Testing

To run the testing code on a test video ("Test_video.avi"):

  1. python facepad-test.py -input Test_video.avi -isVideo 1
  2. It will generate a txt file in the Score folder which contains the score for each frame.

Acknowledge

Please cite the paper:

@inproceedings{eccv18jourabloo,
    title={Face De-Spoofing: Anti-Spoofing via Noise Modeling},
    author={Amin Jourabloo*, Yaojie Liu*, Xiaoming Liu},
    booktitle={In Proceeding of European Conference on Computer Vision (ECCV 2018)},
    address={Munich, Germany},
    year={2018}
}

@inproceedings{eccv18jourabloo,
    title={Learning Deep Models for Face Anti-Spoofing: Binary or Auxiliary Supervision},
    author={Yaojie Liu*, Amin Jourabloo*, Xiaoming Liu},
    booktitle={In Proceeding of IEEE Computer Vision and Pattern Recognition (CVPR 2018)},
    address={Salt Lake City, UT},
    year={2018}
}

If you have any question, please contact: Amin Jourabloo