爱美丽 Beauty

简体中文

Features

Beauty is an AI drive app that can help user become beautiful.

it contain those functions:

1. face score cheek

2. face beauty report

3. face imporve proposals

4. face comparison ( pk )

right now, it can only support asian women

and other function is under construction

The latest Android Version download:

https://gitee.com/knifecms/beauty/releases

(there is no web connection data transfer, every function works in mobile locally )

| | | | |---|---|---|

Project Introduce

1.face contour detection

use Dlib

2.face skin detection

byol + lda

3.Overall characteristics

resnet

Sub projects

1. android beauty app

2. deep learning face beauty research

3. asian face leaderboard

   and leaderboard website: http://1mei.fit

Environment

• Python 3.8

Usage in python

1.clone:

git clone https://gitee.com/knifecms/beauty.git

2.Install depend;

2.1 new install:

conda install cmake
conda install nodejs
conda install dlib

2.2 Import conda env：

conda env create -f face.yaml

3.Modify predict.py image path

# change the detect image path
test = "data/2.jpg"

4.Execute:

python predict.py

you can get beauty score in [0-5], the higher the better

5.Interpretation of results:

execute dir landmarks/ 

    1_gen_feature.py 
    
    2_prepare_data.py 
    
gen features in: data/face/features.csv

then run:

python predict_interpret.py

6.run in cam:

python predict_cam.py

7.run web service:

python predict_server.py

or run:

./restart_server.sh

preview：

http://locahost:5000/pred

we use two tech to explain result: lime and shap(recommend)

Todo

1.redesign the face report, do not use AI explain framework but combine small face part scores.

2.颜值解释（已添加点位和身体部位对应名称）； (使用传统切割手段 和 胶囊图网络Capsule GNN 对比使用 https://github.com/benedekrozemberczki/CapsGNN https://github.com/brjathu/deepcaps )

3.use lbph in android to detect skin type

4.使用带语义结构的特征（识别特定皮肤纹理等）

5.端上应用：

由于cordova摄像头插件无法通过录像的方式捕捉人脸轮廓，暂时弃用
Android Native C++配置过于复杂，windows下与python兼容性不好

DEV:

train data:

https://github.com/HCIILAB/SCUT-FBP5500-Database-Release

Directory description:

App     	移动端项目
dl          深度神经网络训练过程
doc         文档
feature     特征处理
landmarks   人脸关键点提取过程
leaderboard 人脸排行榜
logs        日志目录
model       模型二进制文件
static      flask服务静态文件
template    flask服务模版文件
test        测试目录

reference

《女性美容美体小百科》

https://wenku.baidu.com/view/b10e711ba58da0116c1749e6.html

https://wenku.baidu.com/view/29392bbb9fc3d5bbfd0a79563c1ec5da50e2d6eb.html

https://max.book118.com/html/2017/1115/140076049.shtm

Other research progress

https://github.com/bknyaz/beauty_vision

https://github.com/ustcqidi/BeautyPredict

http://antitza.com/assessment_female_beauty.pdf

The Beauty of Capturing Faces: Rating the Quality of Digital Portraits https://arxiv.org/abs/1501.07304v1

SCUT-FBP5500: A Diverse Benchmark Dataset for Multi-Paradigm Facial Beauty Prediction https://arxiv.org/abs/1801.06345v1

Understanding Beauty via Deep Facial Features: https://arxiv.org/pdf/1902.05380.pdf

Welcome contributions

QQ group: 740807335

wechat: