you can train your own dataset
creat a new folder 'data' and 'checkpoint'，then the directory structure is as follows:
-checkpoint
-data
-train (train dataset)
-image (CT images)
-mask (GT images)
-mask_ (EdgeEGT images)
-val (validation dataset)
-image (CT images)
-mask (GT images)
-mask_ (EdgeEGT images)
You can modify the parameter settings in /resources/train_config.yaml
-batch_size
-learning_rate
-weight_decay
-checkpoint_save_dir
-loss_function ...
finally run train.py, the model will saved in checkpoint folder.
Please cite our paper if you find the work useful:
@article{hu2022deep,
author = {Haigen Hu and Leizhao Shen and Qiu Guan and Xiaoxin Li and Qianwei Zhou and Su Ruan},
journal = {Pattern Recognition},
title = {Deep co-supervision and attention fusion strategy for automatic COVID-19 lung infection segmentation on CT images},
year = {2022},
volume = {124},
pages = {108452},
doi = {https://doi.org/10.1016/j.patcog.2021.108452},
}