This repository contains the implementation of the paper titled "Deep Pixel-wise Binary Supervision for Face Anti-Spoofing" in the TensorFlow framework. The paper introduces a novel technique for face anti-spoofing using deep pixel-wise binary supervision.
Face anti-spoofing is a crucial task in computer vision, especially in security-sensitive applications such as face recognition systems. Spoofing attacks involve presenting a fake face or image to a face recognition system to gain unauthorized access. The proposed technique in this paper leverages deep pixel-wise binary supervision to enhance the robustness of face anti-spoofing systems against such attacks.
The paper detailing the approach implemented in this repository can be found on arXiv.
The page of this task on PapersWithCode.
The implementation is provided in TensorFlow, a popular deep learning framework. The codebase includes the necessary scripts to train, evaluate, and test the face anti-spoofing model using the deep pixel-wise binary supervision technique.
- Python 3.x
- TensorFlow
- Other dependencies specified in
requirements.txt
Install the required dependencies using:
pip install -r requirements.txt
- Training: Train the face anti-spoofing model using the provided training script. Customize the training parameters as needed.
python Train.py
- Testing: Test the trained model on unseen data using the testing script. This script provides predictions and performance metrics on a real-time frames from your camera.
python Test.py
The results obtained from the experiments conducted with the implemented technique are summarized in the paper. Additional details and analysis can be found in the paper and supplementary materials.
This project is licensed under the MIT License - see the LICENSE file for details.
- This project inspired from Face-Anti-Spoofing-using-DeePixBiS
- Utilizing code from the DeepFace repository for implementing a robust face detection system leveraging MediaPipe technology.