Multi-Modal Face Anti-Spoofing Ensemble

This repository contains an ensemble of different techniques for face anti-spoofing, combining approaches such as binary pixel-wise segmentation, patch analysis, depth-based analysis, and deepfake detection. It aims to provide a comprehensive solution to detecting fake or spoofed faces in images.

Folders

• Binary_Pixel: Contains the implementation of a binary pixel-wise segmentation approach for anti-spoofing.

• Patch_Depth: Implements a patch and depth-based approach for anti-spoofing.

• MesoNet: Implements a deepfake detection technique using MesoNet.

• Ensemble(Final): Combines and integrates the above approaches into an ensemble for improved accuracy.

• MTCNN_Face_Extraction: Implements face extraction using the MTCNN model.

• PRNet: Implements depth-based face construction using PRNet.

• SimSwap: Not an added folder but used to generate deepfake data on our dataset. Uses SimSwap repo.

Requirements

Try to run the code in a conda environment

conda create --name fas python=3.8
conda activate fas

In case of a dlib error

pip install cmake
conda install -c conda-forge dlib

To run the code in this repository, you need to install the required dependencies listed in the requirements.txt file.

pip install -r requirements.txt

Requirements differences in Mac

• Upgrade macOS to version 12+
• Remove torch, torchvision from requirements.txt
• Install miniconda
• Run the following before following the other conda instructions

conda config --set auto_activate_base false

• Brew install the rust compiler

brew install cmake protobuf rust

• Install Tensorflow and Pytorch using the follows

conda install -c apple tensorflow-deps
pip install tensorflow-macos
pip install tensorflow-metal
conda install pytorch torchvision torchaudio -c pytorch

• Other installations which might be required for installation: hdf5, h5py, chardet, scipy

Running the code

• Create the environment from the above instructions
• Make Ensemble(Final) as current directory.
• Create a folder test_img_folder and add images to it
• Run the command

python ensemble.py