DeepPixFace-PyTorch

A PyTorch implementation of Deep Pixel-wise Binary Supervision for Face Presentation Attack Detection (ICB'19).

Table of Contents:

• DeepPixFace-PyTorch
  • Data
  • Model
  • Experiment configs
  • Requirements
  • Usage
  • Results
  • Future Work
  • References
  • License

Data:

• CASIA Face-Antispoofing dataset was used for training and inference.
• Can be downloaded from here.
  • Requires signing and mailing agreement to competition organizers.
• Research paper for the CASIA-SURF Dataset

Model:

• DenseNet based implementation with dual supervision.

Experiment configs:

- Input size: 224 x 224 x 3
- Batch size: 8
- Learning rate: 1e-4
- Optimizer: Adam
- Model was trained locally on a GTX 1050Ti. 
- Small batch-size of 8 is used to ensure model meets local compute requirements.

Usage:

• Clone the repository here
• python main.py
• To run on a GPU, you need to enable cuda in the config file.

Future-Work:

• Update repo with demo visualizations corresponding to implementation

Requirements:

Check requirements.txt.

References:

• DeepPixBis Official Implementation: https://bit.ly/31kfLY7

License:

This project is licensed under Apache-2.0 License - see the LICENSE file for details.