Semantic-segmentation-framework

This is a semantic segmentation framework, it can be easy to read and learn. It also can add your network and dataset to train.

0. Quick start

1. Git clone from GitHub.

git clone https://github.com/ShanZard/Semantic-segmentation-framework.git

2. cd to Semantic-segmentation-framework

cd Semantic-segmentation-framework

3. install requirements

pip install -r requirements

1. Dataset

1. Should organize dataset like this:

├── data
│   ├── your_dataset
|   |   |——train
|   |   |   ├── images
│   │   │   │   ├── xxx{img_suffix}
│   │   │   │   ├── yyy{img_suffix}
│   │   │   │   ├── zzz{img_suffix}
│   │   │   ├── mask
│   │   │   │   ├── xxx{seg_map_suffix}
│   │   │   │   ├── yyy{seg_map_suffix}
│   │   │   │   ├── zzz{seg_map_suffix}
│   │   ├── val/test
│   │   │   ├── images
│   │   │   │   ├── xxx{img_suffix}
│   │   │   │   ├── yyy{img_suffix}
│   │   │   │   ├── zzz{img_suffix}
│   │   │   ├── mask
│   │   │   │   ├── xxx{seg_map_suffix}
│   │   │   │   ├── yyy{seg_map_suffix}
│   │   │   │   ├── zzz{seg_map_suffix}

Note:

1. images names should be same as mask
2. mask should be [h,w], if your mask is RGB[3,h,w], you can use tools/pre_processdataset.py to convert it .

2. Before train

1. should change utils/palette.py
2. change your args in train.py

3. Test