TCSVT 2025

:pushpin: This is an official PyTorch implementation of De-LightSAM: Modality-Decoupled Lightweight SAM for Generalizable Medical Segmentation

[arXiv] [BibTeX]

📰News

[2025.10.11] The article has been accepted by: IEEE Transactions on Circuits and Systems for Video Technology.

[2024.08.08] The pre-print paper has been uploaded!

[2024.08.07] Paper will be updated soon!

[2024.08.07] Code and model checkpoints are released!

🛠Setup

git clone https://github.com/xq141839/De-LightSAM.git
cd De-LightSAM
conda create -n ESP python=3.10
conda activate ESP
conda install pytorch==1.13.0 torchvision==0.14.0 torchaudio==0.13.0 pytorch-cuda=11.6 -c pytorch -c nvidia
pip install albumentations==0.5.2
pip install pytorch_lightning==1.1.1
pip install monai

Note: Please refer to requirements.txt

📚Data Preparation

The structure is as follows.

De-LightSAM
├── datasets
│   ├── image_1024
│     ├── ISIC_0000000.png
|     ├── ...
|   ├── mask_1024
│     ├── ISIC_0000000.png
|     ├── ...

🎪Segmentation Model Zoo

We provide all pre-trained models here.

MA-Backbone	MC	Checkpoints
TinyViT	Dermoscopy	Link
TinyViT	X-ray	Link
TinyViT	Fundus	Link
TinyViT	Colonoscopy	Link
TinyViT	Ultrasound	Link
TinyViT	Microscopy	Link

📜Citation

If you find this work helpful for your project, please consider citing the following paper:

@article{xu2024delight,
  title={De-LightSAM: Modality-Decoupled Lightweight SAM for Generalizable Medical Segmentation}, 
  author={Qing Xu and Jiaxuan Li and Xiangjian He and Chenxin Li and Fiseha B. Tesem and Wenting Duan and Zhen Chen and Rong Qu and Jonathan M. Garibaldi and Chang Wen Chen},
  journal={IEEE Transactions on Circuits and Systems for Video Technology},
  year={2025}
}