# beauty
**Repository Path**: xiaochonzi/beauty
## Basic Information
- **Project Name**: beauty
- **Description**: 爱美丽是一款美颜智能应用,目标是提高用户颜值,包括颜值评测,缺陷报告,颜值PK等
- **Primary Language**: Python
- **License**: MulanPSL-2.0
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No
## Statistics
- **Stars**: 0
- **Forks**: 48
- **Created**: 2025-03-22
- **Last Updated**: 2025-03-22
## Categories & Tags
**Categories**: Uncategorized
**Tags**: None
## README
# 爱美丽 Beauty
Help you become beautiful [china software copyright 10293875]
------
[简体中文](./readme_cn.md)
## 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.
------
There are two levels of human beauty:
1. Beauty based on bioengineering;
2. Beauty based on genetic appeal;
Bioengineering represents the universal value of biological evolution, while genetic appeals represent the goals pursued by individuals;
Using big data samples to model, and finally giving the best score through model prediction, is the embodiment of universal value and beauty;
Individual pursuit of goals requires the expression and participation of users’ own demands, so the model must run on the terminal and have the ability to interact with users.
------
Everyone's aesthetic standards are different, which determines that the standard of appearance is not the only one.
The beauty of each person is not unique, which also shows that relying on a single standard cannot discover the whole of a person.
But nowadays medical and aesthetic institutions are very popular, trying to obtain customer results through closed tests that guide them to favorable standards.
Such evaluation is not the pursuit of beauty, but the pursuit of money!
Although such behavior by the company is understandable without harming the interests of customers, it has led to many people being misled!
Everyone has their own unique beauty, and what they need is to discover and show themselves.
Therefore, I firmly believe that by adhering to open and transparent evaluation standards, and insisting on close interaction with users,
Only then can we truly excavate and show the beauty of users in a fair way!
Only then can we truly propose meaningful solutions to help users pursue beauty!
Guide and help users achieve themselves!
Therefore, Beauty will always be open source, and the standards of its evaluation model are also open and reproducible.
------
The latest Android Version download:
Tencent app treasure:
https://sj.qq.com/appdetail/com.ml.projects.beautydetection
huawei app market:
https://appgallery.huawei.com/app/C105812963
google play market:
https://play.google.com/store/apps/details?id=com.ml.projects.beautydetection
(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](./App/readme.md)
2. [deep learning face beauty research](./dl/readme.md)
3. [asian face leaderboard](./leaderboard/readme.md)
and leaderboard website: http://1mei.fit
## Environment
- Python 3.8
## Usage in python
#### 1.clone:
git clone https://gitee.com/knifecms/beauty.git
or
git clone https://github.com/showkeyjar/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.face score explain(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.use semantic structural features(识别特定皮肤纹理等)
### DEV:
train data:
https://github.com/HCIILAB/SCUT-FBP5500-Database-Release
Directory description:
App android project
dl deep learning models
doc documents
feature face features
landmarks face landmarks process
leaderboard asian face leaderboard
logs log dirs
model trained models
static flask web assets
template flask templates
test unit test

## 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
Automatic Facial Spots and Acnes Detection System
https://www.researchgate.net/publication/276040820_Automatic_Facial_Spots_and_Acnes_Detection_System
## Welcome contributions
QQ group: 740807335
wechat:
