CDSS-Det

Official implementation of Cross-Domain Semi-Supervised Organ Detection - MIDL 2026.

Installation

conda create -n cdss-det python=3.8.18
conda activate cdss-det
conda install pytorch==2.3.0 torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia
pip install -r requirements.txt

Configs

Config files are under config/midl, adjust the dataset path according to your storage and other parameters as you needed.

Datasets

Our preprocessed datasets can be downloaded from here. We don't upload Abdomen-Atlas since it is too large, you can download it from the official source instead.

If you use your own dataset, make sure it is preprocessed in the same way as organdetr/data/preprocessor.py.

Training

Pre-training

First, pre-train the model using source data.

export ORGANDETR_DATA=${your_data_path}
python scripts/sd_train.py --config ${your_config_name}

Cross-Domain training

Second, use the pre-trained model above to train the source, labeled target and unlabeled target data together.

export ORGANDETR_DATA=${your_data_path}
python scripts/midl_ts_cd_train.py --config ${your_config_name}$

Inference

python scripts/ts_test.py --run ${your_config_name} --model ${your_model_checkpoint_name}

Acknowledgements

The codebase is developed based on OrganDETR

If you find this repository useful, please consider giving a star 🌟 and citing the paper:

@inproceedings{
li2026crossdomain,
title={Cross-Domain Semi-Supervised Organ Detection},
author={Nian Li and MORTEZA GHAHREMANI and Bailiang Jian and Pascual Tejero Cervera and Benedikt Wiestler and Marcus Makowski and Christian Wachinger},
booktitle={Medical Imaging with Deep Learning},
year={2026},
url={https://openreview.net/forum?id=NSjBDpsZqV}
}