songcv / de-unet

Config files for my GitHub profile.

DE-unet

Requirements

conda virtual environment

python=3.8 
pytorch==1.13.0
cuda==11.7 
pip install mmcv-full==1.7.0 -f https://download.openmmlab.com/mmcv/dist/cu117/torch1.13.0/index.html

Datasets

Create folder named 'root_path', its structure is

    aeril_root_path/
    ├── images
       ├── train
          ├── xxx.tif
          ├── ...
       ├── val
       ├── test
    ├── masks
      ├── train
          ├── mask.png(0-num_class)
          ├── ...
      ├── val
      ├── test

Training

python tools/train.py config/path --work-dir save/path

Evaluation

Accuracy: python tools/test.py config_file checkpoints_file --eval mIoU

Speed

FPS: python tools/fps.py config_file --height / --width /