DAME:Automatic detection of melanins and sebums from skin images using generative adversarial network

Lun Hu

Peng Zhou

---

Folders

• data contains data processing related scripts.
• datasets contains pre-training script.
• models contains Pix2pix related deep learning model.
• options contains scripts that configure the relevant parameters.
• util contains frame application related scripts.

Files

• prepare.py contains code about image gray and image enchancement.
• test.py contains related code to test.
• train.py contains related code to train.

Usage

1. git clone this project;

note:ImageDataset folder is test dataset;SourceCodes folder is model source code

2. cd DAME/SourceCodes(root directory),new folder named image, the directory structure is as follows: place the original image in the jpg folder and place the marked image in the mark folder.

image
│   ├── finish
│   ├── origin
│   │   ├── black
│   │   │   ├── jpg
│   │   │   └── mark
│   │   └── oil
│   │       ├── jpg
│   │       └── mark 

3. run python prepare.py and it will generate image after process(DAME or guassian or CLAHE). the directory structure is as follows:

image
│   ├── finish
│   │   ├── black
│   │   │   ├── jpg
│   │   │   └── mark
│   │   └── oil
│   │       ├── jpg
│   │       └── mark
│   ├── origin

note: different processes correspond to different functions DAME:prepare.py/gray,guassian:prepare.py/gaussian,CLAHE:prepare.py/clahe

4. in the root directory, run python datasets/combine_A_and_B.py --fold_A finish/black/jpg --fold_B finish/black/mark --fold_AB finish/black/ --no_multiprocessing, It will generate the dataset required by the pix2pix model.
5. train model:python train.py --dataroot finish/black --model pix2pix --name black
6. test model:python test.py --dataroot ./datasets/black/ --name black --model pix2pix