Supervised Contrastive Learning for Snapshot Spectral Imaging Face Anti-Spoofing [CVPR2024 Workshop]

This repository is the official implementation of Supervised Contrastive Learning for Snapshot Spectral Imaging Face Anti-Spoofing(https://arxiv.org/abs/2405.18853). We rank first at the Chalearn Snapshot Spectral Imaging Face Anti-spoofing Challenge on CVPR 2024; the paper is accepted by CVPR 2024 workshop.

Reproduction Codes for Snapshot Spectral Imaging Face Anti-spoofing Challenge

Here we provide the training and testing codes for the reproduction of our result. To facilitate the reproduction of the result, we also provide the training log file log.csv and the result file result.txt.

1. Environment Setting

In order to reproduce the same result, the environment setting is important. Please make sure the basic environment setting is the environment py310 + torch2.2.1 + cuda12.1 with 1 NVIDIA Tesla A100 80G GPU, and our algorithm can be deterministic under this environment. You can run the following script to configure the necessary python package:

pip install albumentations==1.3.0
pip install kornia==0.7.2
pip install pytorch_metric_learning==2.4.1
pip install torch_ema==0.3
pip install numpy==1.26.4
pip install pandas==2.2.1
pip install timm==0.9.16

2. Training

Before runing the training script python train.py, you need to make some revisions for the preparation.

• training image path: please download and extrat the HySpeFAS training data HySpeFAS_trainval.zip, and revise training image path /path/to/HySpeFAS_trainval/images/ at the line 118 and 121 of train.py.

• model saving path: please revise model saving path at the line 283 of train.py.

• ImageNet pretrained model path: please download the ImageNet pretrained model via the url https://github.com/lukemelas/EfficientNet-PyTorch/releases/download/1.0/adv-efficientnet-b3-cdd7c0f4.pth, and revise the ImageNet pretrained model path at the line 327 of efficientnet/utils.py

After that, you can run the the training script python train.py to train the target model. The model and log files will be saved in the model saving path above.

3. Testing

Before runing the testing script python test.py, you need to make some revisions for the preparation.

• testing image path: please download and extrat the HySpeFAS testing data HySpeFAS_test.zip, and revise testing image path /path/to/HySpeFAS_test/images/ at the line 49 and 52 of test.py.

• inference model path: Based on the path of the model file at the last training epoch, please revise inference model path /path/to/mat_b3_256_sfas_model_0/model_epoch_30.pth at the line 91 of test.py.

• inference threshold (optional) : you also need to revise probability threshold for testing data at the line 119 of test.py, based on the validation threshold in the training log, but we have done it for you.

After that, you can run the the testing script python test.py to get the result file result.txt.