ZHANG-SHI-CHANG / Peppa_Pig_Face_Engine

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

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:

and with face mask:

requirment

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

• opencv

• python 3.6

• easydict

• flask

update

2020.4.28 A new detector based on centernet is trained, here is the link, but i don't know how to integrate it into this project, because it is a tf1 project. I am thinking about to rewrite the project to make it more usefull.

2020.2.4: add http server

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)
  

  • baidu disk ( password yqst )
  • google drive

• 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)
  

  • baidu disk (code fcdc)
  • google drive

  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
  

2. 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

start a http server

1. run python demo.py --web 1

2. test by run python web_demo_test.py,

the result is a json, formate: [{ "bbox": [x1, y1, x2, y2], "landmark": [[x1, y1], [x2, y2],[x3,y3],[...]]}]

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.

• dsfd_light_model tf2 branch
• centernet
• face_landmark

TODO

• Transfer to tensorflow 2.0
• small model including tflite
• add http server demo
• Add some GAN model to make it fun ing....
• 3-d face algorithm
• a mobile device version, it is on the way, i learn a lot about mobile device. **
• switch to pytorch, HAHA... dont worry it wont happen.