RSAFormer is a novel network architecture designed for the polyp segmentation task. It also works well in various kinds of medical segmentation tasks. In this model, we use dual decoders to obtain respective features and coarse segmentation maps. The maps are utilized to provide pixel classification information, which can be used in the region self-attention module for the subsequent feature enhancement. The core highlight is the flexible combination of decoders. We highly recommend introducing different decoders in this network to adapt to your task and achieve the best results!
conda create -n RSAFormer python==3.8.16
conda activate RSAFormer
pip install -r requirements.txt
Download pretrained checkpoints from Google Drive and move it to the pretrained_pth directory.
Download dataset from Google Drive and move it to the data directory.
python expr.py
Please cite our paper if you find the work useful:
@article{yin2024rsaformer,
title={RSAFormer: A method of polyp segmentation with region self-attention transformer},
author={Yin, Xuehui and Zeng, Jun and Hou, Tianxiao and Tang, Chao and Gan, Chenquan and Jain, Deepak Kumar and García, Salvador},
journal={Computers in Biology and Medicine},
volume={172},
pages={108268},
year={2024},
publisher={Elsevier}
}
Please contact [email protected] for any further questions.