/FAS_ModelZoo_v4

DepthSupervision

Primary LanguagePython

Learning Deep Models for Face Anti-Spoofing: Binary or Auxiliary Supervision

Environment

  • HUAWEI CLOUD
  • Python 3.7
  • Tensorflow-1.15

Datasets

Installation

  • Clone FAS_ModelZoo_v4 repository. We'll call the directory that you cloned ReId_Eigen as $ROOT_PATH. 
    git clone --recursive https://github.com/liuajian/FAS_ModelZoo_v4.git

Requirements

  • Account: hw80211537
  • Obs: ajian3

Usage

  • data_url: Oulu-Train
  • train_url: Jobs
  • Start training: python train_depth_yun.py
  • Start testing: python test_depth_yun.py

Results

  • The testing results of these methods based on multi-shot setting are as follows(%): 
    ---------------------------------------------
|  Method   | APCER(%) | BPCER(%) | ACER(%) |
|Aux(Depth) |   2.7    |   2.7    |   2.7   |
|   Ours    |   3.5    |   1.4    |   2.4   |
---------------------------------------------
Note that the metric of ACER is the final indicator, and the smaller value means better performance.

Citation

Please cite the following papers in your publications if it helps your research:
@inproceedings{Liu2018Learning,
title={Learning deep models for face anti-spoofing: Binary or auxiliary supervision},
author={Liu, Yaojie and Jourabloo, Amin and Liu, Xiaoming},
booktitle={CVPR},
year={2018}
}

Questions

Please contact 'ajianliu92@gmail.com'