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