Code for paper: DECOR-NET: A COVID-19 LUNG INFECTION SEGMENTATION NETWORK IMPROVED BY EMPHASIZING LOW-LEVEL FEATURES AND DECORRELATING FEATURES. Specifically, we design a channel re-weighting strategy and a decorrelation loss to improve COVID-19 infection segmentation. The model is built based on the MONAI framework.
- Requirements
- Linux
- Python 3.6+
- Dependencies.
- numpy>=1.21.5
- pandas>=1.1.5
- torch>=1.7.0
- torchvision>=0.11.1
- monai>=0.8.1
- pillow
- yaml
- json
- torchmetrics
The preprocessed COVID-19 Challenge dataset can be found in BaiduDisk (key = tu1h).
The COVID-19 Challenge dataset can be found here.
The path of dataset need to be set in ./DecorNet/Unet.yaml before training.
python train.py
Set runs_file in evaluation.sh before running it.
You can find your folder's name in ./DecorNet/runs/
bash evaluation.sh
All the metrics generated by evaluation.sh will be save in the folder of runs_file.
You can find the code for Decor loss in ./Decor_loss.py.