SmooFaceEngine

An open source face recognition engine named SmooFaceEngine.

Detailed technical principles: 《Python人脸识别:从入门到工程实践》

勘误表(errate)

Chinese Wiki

Introduction

Let us see something about this project now.

What is this project?

This project is an open source project about face recognition. In the project, we implemented a face recognition engine that was one-shot training.

Why did this?

JUST FOR FUN! I think I may be not a face recognition expert, but a geeker.

Principle of SmooFace

In this project, we implemented some CNN, such as VGGNet, VIPL face net, ResNet and XCEPTION etc.

We used AM-Softmax loss as the cost function here, rather than triple loss or other metric learning loss functions. This is because AM-Softmax has less training time, but the accuracy is still good.

How to use?

This project is a demo. In order to see the experimental results, I trained the model with a small data set. I used data augmentation in this project, so that I can get a robust model. If you want to use this project in your production environment, you should do more.

Train

python train.py

Predict

python predict.py

Web API

http://127.0.0.1:8080/test

Dependencies

Python 3.5
flask 1.1.1
h5py 2.10.0
Keras 2.3.1
numpy 1.17.1
scipy 1.3.2
tensorflow 1.14.0

Reference

Papers

You can search the following papers in Google Scholar

AM-Softmax
Sphere face
FaceNet
ResNet
Xception
MobileNet v1,v2,v3
VIPL Face net

Open source projects

  1. https://github.com/xiangrufan/keras-mtcnn
  2. https://github.com/happynear/AMSoftmax
  3. https://github.com/Joker316701882/Additive-Margin-Softmax
  4. https://github.com/hao-qiang/AM-Softmax
  5. https://github.com/ageitgey/face_recognition
  6. https://github.com/oarriaga/face_classification
  7. https://github.com/seetaface/SeetaFaceEngine
  8. https://github.com/jiankangdeng/handbook

LICENSE

Apache license version 2.0

How to contribute

There are many bugs here, so you could send some pull requests or give some issues for this project. Thank you very much :)

TODO

  1. give train.py arguments: for different training set
  2. refactor: to optimize code
  3. etc.