akfheaven / Peppa_Pig_Face_Engine

A simple face detect and alignment method, which is easy and stable.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Peppa_Pig_Face_Engine

introduction

It is a simple demo including face detection and face aligment, and some optimizations were made to make the result smooth.

CAUTION: this is the tensorflow2.0 branch, if you want to work on tf1, please switch to tf1 branch, it still works.

Purpose: I want to make a face analyzer including face detect and face alignment. Most of the face keypoints codes that opensourced are neither stable nor smooth, including some research papers. And the commercial sdk is pretty expensive. So, there is Peppa_Pig_Face_Engine.

I think it is pretty cool, see the demo:

click the gif to see the video: demo

and with face mask: face mask

requirment

  • tensorflow2.0 (tensorflow1 need to switch to tf1 branch )

  • opencv

  • python 3.6

  • easydict

useage

  1. download pretrained model, put them into ./model
  • detector

     Lightnet_0.5  including a tflite model, 
      (time cost: mac i5-8279U@2.4GHz, tf2.0 15ms+, tflite 8ms+-,input shape 320x320, model size 560K)
    
  • keypoints

    shufflenetv2_0.75   including a tflite model, 
     (time cost: mac i5-8279U@2.4GHz, tf2.0 5ms+, tflite 3.7ms+- model size 2.5M)
    

    the dir structure as :

       ./model
       ├── detector
       │   ├── saved_model.pb
       │   └── variables
       │       ├── variables.data-00000-of-00002
       │       ├── variables.data-00001-of-00002
       │       └── variables.index
       ├── keypoints
       │   ├── saved_model.pb
       │   └── variables
       │       ├── variables.data-00000-of-00002
       │       ├── variables.data-00001-of-00002
       │       └── variables.index
    
  1. run python demo.py --cam_id 0 use a camera
    or python demo.py --video test.mp4 detect for a video
    or python demo.py --img_dir ./test detect for images dir no track
    or python demo.py --video test.mp4 --mask True if u want a face mask

Train

The project is based on two of my other repos, and both tensorflow1 and tensorflow2 are supported. If you want to train with your own data, or you want to know the details about the models, click them.

ps

The project is supported by myself, and i need your advice to improve it. Hope the codes can help you, contact me if u have any question. At last, i need your star,also your contribution.

TODO

  • Transfer to tensorflow 2.0
  • small model including tflite
  • Add some GAN model to make it fun ing....
  • 3-d face algorithm
  • maybe a mobile device version, so tired
  • switch to pytorch, HAHA...

About

A simple face detect and alignment method, which is easy and stable.

License:Apache License 2.0


Languages

Language:Python 100.0%