BaseFaceRecognition-keras

Introduction

This is a simple implementation of face recognition using keras.

Dataset

The dataset is from CASIA-WebFace. It contains 494414 images of 10575 identities. The images are cropped and aligned with 5 landmarks.

Model

The model is based on FaceNet. Its backbone is MobileNet. The model is trained with triplet loss.

Usage

Train

python train.py

Predict

python predict.py

Reference

[1] FaceNet: A Unified Embedding for Face Recognition and Clustering

[2] MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications