Human beautifying and Image matting

We build an end-to-end human beautifying and matting system, which can handle any human photos supplied by user. Firstly we beautify the photo depending on the users’ preferences(smooth, lip brighten, whiten, thin, etc.), and then we separate the man from the background(matting). Using this system users can generate ID Photos easily.

example

network

Deployment

Please visit http://104.199.231.92:443/ for live demo.

Back end: Flask

Front end: HTML+JavaScript

Architecture

.
├── Readme.md
├── app.py
├── beautify.py
├── infer.py
├── make_dataset.py
├── segmentation
├── simple_request.py
├── src
│   ├── __init__.py
│   ├── dataset.py
│   ├── resnetUnet.py
│   └── unet_model.py
├── static
│   ├── index.js
│   └── style.css
├── templates
│   ├── index.html
│   └── try.html
├── test.py
├── test_beauty.py
├── tmp_test_beauty.py
├── train.py

Reference

https://blog.csdn.net/grafx/article/details/70232797?locationNum=11&fps=1

https://zhuanlan.zhihu.com/p/29718304

https://www.kaggle.com/laurentmih/aisegmentcom-matting-human-datasets

https://www.kaggle.com/vbookshelf/selfie-segmenter-keras-and-u-net

https://github.com/milesial/Pytorch-UNet

https://www.kaggle.com/rishabhiitbhu/unet-starter-kernel-pytorch-lb-0-88

https://arxiv.org/pdf/1912.03630.pdf

https://arxiv.org/pdf/1902.06838.pdf