中文请参考我的CSDN博客

Introduction

This is a Face-recognition demo that I combined with MTCNN^[1] and FaceNet^[2] under the PyCharm IDE. It can train any dataset and implement face verification, face recognition and face search functions. Finally, I used Tkinter to make the interface, and the user could easily operate it. Anyway if you want to learn face recognition or Tkinter UI, I believe my project will help you.

Results

• Face Verification (Verification ChenZhenyu)

Successful Verification! This image was identified as ChenZhenyu!

• Face Recognition

Detected LiuXin,YangShuntao,LiJialong,ChenZhenyu,LiuKe,ZhangYuchan,ZhangJinrui in this image!

• Face Search (search ChenZhenyu, LiuKe)

Found ChenZhenyu,LiuKe in this image!

Environment

• Windows 10

Python 3.6.4 + TensorFlow 1.6.0 + tk 8.5.19 + dlib 19.4 + opencv 3.4.1 + scikit-image 0.13.0 + scikit-learn 0.19.1

• Ubuntu 16.04

Python 3.6.2 + TensorFlow 1.3.0 + tk 8.5.18 + dlib 19.4 + opencv 3.1.0 + scikit-image 0.13.0 + scikit-learn 0.19.0

Under Ubuntu, it is recommended that you choose English (Chinese display sometimes has problems).

UI Display

Windows (The program can also run under Ubuntu system)

Login

Train

Retrieval

Search the images and display them on the right side of the UI

Predict

Visualize the result of the image on the right side of the UI

Notebook Display

Please see face_recognition_process.ipynb. It will help you to understand my code.

How to use it

• Download my resources and place them in the folder resources

  Google Driver: https://drive.google.com/file/d/19WKS0aDJy9ncmWCR8ScvcmEU8tmDZoEM/view?usp=sharing
  BaiDu Cloud: https://pan.baidu.com/s/1SOBR0eUQvUyYutuhHxLfsQ

• Download the MTCNN weights and place them in the folder models/mtcnn

  Google Driver: https://drive.google.com/file/d/12r98MQsxy4_cPekGpQD7BknPm5Tdyd7b/view?usp=sharing
  BaiDu Cloud: https://pan.baidu.com/s/178wilMLklTdjavE5k_SRDA

• Download the FaceNet weights and place it in the folder models/20170512-110547

  Google Driver: https://drive.google.com/file/d/1aQjhcidR3z2YbjB0qYIFDpJChiP3n7Zy/view?usp=sharing
  BaiDu Cloud: https://pan.baidu.com/s/15UsfsxfLu3iSZSfuqVmx-g

• If you need to perform face alignment, you must download alignment model and put it under the models folder.

  Google Driver: https://drive.google.com/file/d/1C09m6bb2MAkM4G3cBhtsuRuW3S-qmgD5/view?usp=sharing
  BaiDu Cloud: https://pan.baidu.com/s/1uUCs9HaKtkv6Q5ST55nLbw

• After completing the above steps, you can implement face recognition with jupyter notebook.
• If you need to use the UI, open it in PyCharm IDE and run ui_with_tkinter/choose_language.py.

Folder Structure

models/
----------20170512-110547/
--------------------20170512-110547.pd
--------------------model-20170512-110547.ckpt-250000.data-00000-of-00001
--------------------model-20170512-110547.ckpt-250000.index
--------------------model-20170512-110547.meta
----------mtcnn/
--------------------det1.npy
--------------------det2.npy
--------------------det3.npy
----------my_models/
----------shape_predictor_68_face_landmarks.dat
dataset/
----------train_english/
--------------------XiaoMing/
------------------------------XiaoMing1.jpg
------------------------------...
--------------------XiaoHu/
------------------------------XiaoHu1.jpg
------------------------------...
--------------------...
----------train_chinese/
--------------------小明/
------------------------------XiaoMing1.jpg
------------------------------...
--------------------小虎/
------------------------------XiaoHu1.jpg
------------------------------...
--------------------...

Reference

[1] MTCNN
[2] FaceNet