Skin Detective AI Engine

This is the AI core engine behind the project Skin-Detective AI which is published on Diagnostics (Scopus Impact Factor 3.9)

Article links

• MDPI Journal
• National Library of Medicine
• Diagnostics 2022, 12(8), 1879

News

• vnexpress
• vnexpress
• tuoitre.vn
• cafef
• video

Reference

• COCO bject detection format

How to use

1. Container

Reproduce the exact environment

docker pull hthquan28/skin-detective

Run Jupyterlab in container, with expose port 8080 and using NVIDIA Driver

docker run -it --gpus all -p 8080:8080 --name skin_container hthquan28/skin-detective

Run on detached mode

docker run -d --gpus all -p 8080:8080 --name skin_container hthquan28/skin-detective

Using terminal in container

docker exec -it skin_container bash

Build container

sh build_local.sh

Note: Due to different architectures of your Graphic Card, you might not able to run it, for more information please refer this article

2. Setup

2.1 Install requirement packages

pip install -r requirements.txt

2.2 Prepare dataset

Object detection dataset should be organized in COCO format

root
| - bbox   <contain COCO format files info>
    | - img1  <coco info of img1>
    | - img2  <coco info of img2>
| - image <contain images>
    | - img1.jpg <img1>
    | - img2.jpg  <img2>
| - models   <store model files>
| - mAP    <eval package>

2.3 Train object detection

python acne_detection.py

Look at help for more detail parameters Model will be trained and stored in ./models folder

2.4 Train Grade Classifier

Run acne_circle_final.ipynb This should output lightgbm model for grade classifer

3. Demo

Run file present2.ipynb to present the result and evaluation