This repo is the official implementation for the paper: "LGANet: Local-Global Augmentation Network for Skin Lesion Segmentation" at ISBI 2023.
This paper proposes a novel framework, LGANet, for skin lesion segmentation. Particularly, two module, LFM and GAM are constructed. LFM aims at learning local inter-pixel correlations to augment local detailed information around boundary regions, while GAM aims at learning global context at a finer level to augment global information.
Fig.2. The structure of the proposed LGANet. LFM and GAM are integrated into the Transformer encoder based framework to learn local detailed information around boundary and augment global context respectively, where dense concatenations are used for final pixel-level prediction.
You could download the pretrained model from here. Please put it in the " ./pretrained" folder for initialization.
python train.py
python test.py
python evaluate.py
Some of the codes in this repo are borrowed from:
Please cite the following paper if you think this project is useful for your work. Thanks.
@inproceedings{
GuoFWZL2023LGANet,
author = { Guo, Qingqing and Fang, Xianyong and Wang, Linbo and Zhang, Enming and Liu, Zhengyi},
booktitle = {Proceedings of the 20th IEEE International Symposium on Biomedical Imaging - ISBI 2023},
title = {{LGANet: Local-global augmentation network for skin lesion segmentation}},
address = {Cartagena, Colombia},
year = {2023}
}