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

Let us see something about this project now.

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

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

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.

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.

python train.py

python predict.py

http://127.0.0.1:8080/test

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

You can search the following papers in Google Scholar

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

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

Apache license version 2.0

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

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