This is an official repository for SegMIC.

Quick Start

Part1. Dataset

We are organizing our benchmark UniMedDB, and we will release the official version. Once you have downloaded UniMedDB.tar.gz, unzip it and place all datasets in Meddata. Specifically, for OOD datasets, you should create a new sub-folder named "not_train" in Meddata, and put all OOD datasets into "Meddata/not_train".

Part2. Train

You can train SegMIC by:

bash train_train_segm_512_base.sh

Part3. Inference

You can infer SegMIC by:

bash inference_per_label.sh

😄 If you find this repository help you well, please cite our paper:

@article{zhao2025segmic,
  title={SegMIC: A Universal Model for Medical Image Segmentation Through In-Context Learning},
  author={Zhao, Jianwei and Yang, Fan and Li, Xin and Jiao, Zicheng and Zhai, Qiang and Li, Xiaomeng and Wu, De and Fu, Huazhu and Cheng, Hong},
  journal={Pattern Recognition},
  pages={112179},
  year={2025},
  publisher={Elsevier}
}