A novel non-pretrained deep supervision network for polyp segmentation

Zhenni Yu, Li Zhao, Tangfei Liao, Xiaoqin Zhang, Geng Chen, Guobao Xiao Pattern Recognition (PR), 2024, 154, 110554

Usage

Installation

git clone https://github.com/guobaoxiao/NPD-Net
cd NPD-Net

Pre-trained models

• Traindata: download the traindata set from here, put into 'data/TrainDataset'
• Testdata: download the traindata set from here, put into 'data/TestDataset'
• pre-weigth: download the weight of res2net50 from here, put into 'weigth/' download the weight of pvtv2 from here, put into 'weigth/'

Weights

• pre-weigth: gt_encoder_model: loss-10.5362-GTENCODER-NPD_Net-1.pth, put into 'model_path'

  链接：https://pan.baidu.com/s/1p5UBnN5lJFKABD-mmp0xnw 提取码：cevt

Train

python Mytrain.py

Test

python Mytest.py

• For miou and mDice:

python eval.py

• For more detailed metrics :

python MSCAF_COD_evaluation/evaluation.py

Citation

If you find this project useful, please consider citing:

@article{yu2024novel,
  title={A novel non-pretrained deep supervision network for polyp segmentation},
  author={Yu, Zhenni and Zhao, Li and Liao, Tangfei and Zhang, Xiaoqin and Chen, Geng and Xiao, Guobao},
  journal={Pattern Recognition},
  pages={110554},
  year={2024},
  publisher={Elsevier}
}