Roll-Face / anti-spoofing

In verification services related to face recognition, such as eKYC and face access control, it is vital to know whether the input face is real or fake

Step 1: Extract frame from video

python src/generate.py

Step 2: Crop image to scale

python src/generate_patches.py

├───scale_1.0
│   └───liveness_face
│       └───datasets
│           └───datasets_train
|   └───label_list.txt
├───scale_2.7
│   └───liveness_face
│       └───datasets
│           └───datasets_train
|   └───label_list.txt
└───scale_4
    └───liveness_face
        └───datasets
            └───datasets_train 
    └───label_list.txt

Config params: ./src/default_config.py

Step 3: training model

python train.py

Step 4: Demo

python src/demo.py --model_dir ./resources/ckpt_onnx/ --input_data ./data/0.mp4

refers:

• https://github.com/minivision-ai/Silent-Face-Anti-Spoofing/blob/master/README_EN.md
• https://github.com/Elroborn/Face-anti-spoofing-based-on-color-texture-analysis
• https://github.com/mnikitin/Learn-Convolutional-Neural-Network-for-Face-Anti-Spoofing