Transfer-Learning-Face-Anti-Spoofing-Attack-Model

Authors: Renato-Castro, Jonathan-Javier and Rodolfo-Galvez.

This repository contains the jupyter and mathematica notebooks of the Transfer Learning Model on VGG16, ResNet 50 architecture with ImageNet weights and FaceNet, and also our code proposed of the technique anti spoofing in the paper "Face Liveness Detection Based on Perceptual Image Quality Assessment Features with Multi-scale Analysis". Our project is about a Transfer Learning CNN on VGG16 architecture because it has the best results with an accuracy of 93,016 % in CASIA dataset, 97,321 % MDP dataset and 91,142 % NUAA dataset, we use a "real-time" landmark detection however the transfer learning model and landmark detection don't work together yet (our future work is that these two technologies can work together to reduce errors and improve the accuracy). This also contains the full paper of the "Anti Spoofing Face Detection Technique based on Transfer Learning Convolutional Neural Networks and Real-Time Facial Landmark Detection" to the Latinx in AI - ICML Workshop call.

Link to download the weights of models: https://drive.google.com/drive/folders/1d6Fx-6NVEyJE5DfEPXA4u03KtGVDhfQ2?usp=sharing