EPCR

A PyTorch implementation for the paper [Consistency Regularization for Deep Face Anti-Spoofing] by Zezheng Wang

Dependencies

If you don't have python 3.8 environment:

conda create -n EPCR python=3.8
conda activate EPCR

Then install the required packages:

pip install -r requirements.txt

Download OCIM datasets, preprocess and prepare the train/test list

Run

Run this command to train DG with TransFAS architecture and with normal protocols

bash configs/run_ocim_epcr_ssdg_transfas_cim-o.sh

Run this command to train DG with semi-supervised protocols of leave-one-out protocols.

bash configs/run_ocim_leave_one_om-i_c.sh

Run this command to train DG with semi-supervised protocols of intra-labeled protocols.

bash configs/run_ocim_semi_Intra_live0.2_spoof0.2.sh

Semi-Supervised protocols

https://pan.baidu.com/s/1prFW1NMRwtutpugaqlbr1Q code: pfix

Pretrained odels

https://pan.baidu.com/s/1qL9xUGOuAJbyP4ppXdy44A code: jdfk