Generalizable Method for Face Anti-Spoofing with Semi-Supervised Learning

Nikolay Sergievskiy, Roman Vlasov, Roman Trusov

This is the official implementation of the paper Generalizable Method for Face Anti-Spoofing with Semi-Supervised Learning by the ML research team from Entry

arXiv

Dataset and models status

This work is done using a proprietary dataset, which is why we cannot share the data or pretrained models.

Requirements

• VISSL
• OpenCV, PIL
• timm
• tqdm

WandB

We use WandB for model comparison and monitoring, and the training/validation script relies on it heavily, so by default you will need an account to export results there.

Structure

• code/train_config.yaml - main configuration for training/eval
• code/run.py - training/validation script with CLI
• code/dataset - package with utilities for loading and augmenting data
• code/entry_antispoof - package with utils, loss functions, and network definitions

Contacts

Roman Trusov