unimiss-code's Introduction

UniMiSS & UniMiSS+

This is the official pytorch implementation of our ECCV 2022 paper "UniMiSS: Universal Medical Self-Supervised Learning via Breaking Dimensionality Barrier" and extended IEEE-TPAMI paper "UniMiSS+: Universal Medical Self-Supervised Learning From Cross-Dimensional Unpaired Data".

News

5 Aug 2024: 🎉🎉🎉Our extended version UniMiSS+ has been accepted by IEEE-TPAMI (Impact Factor=20.8). Code will be released soon.

To do

UniMiSS+ fine-tuning code
UniMiSS+ pre-training code and weights
UniMiSS fine-tuning code
UniMiSS pre-training code and weights

Citation

If this code is helpful for your study, please cite:

@article{UniMiSS,
  title={UniMiSS: Universal Medical Self-Supervised Learning via Breaking Dimensionality Barrier},
  author={Xie, Yutong and Zhang, Jianpeng and Xia, Yong and Wu, Qi},
  booktitle={ECCV},
  year={2022}
}
@article{UniMiSS+,
  title={UniMiSS+: Universal Medical Self-Supervised Learning From Cross-Dimensional Unpaired Data},
  author={Xie, Yutong and Zhang, Jianpeng and Xia, Yong and Wu, Qi},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2024},
  publisher={IEEE}
}

Acknowledgements

Part of codes is reused from the DINO. Thanks to Caron et al. for the codes of DINO.

Contact

Yutong Xie ([email protected])

unimiss-code's People

Contributors

Stargazers

Watchers

unimiss-code's Issues

About finetune for the 3D downstream tasks

Undoubtly, you done an amzing work and really appreciated for your effort.
If there is fintune code will be more convenient for us to use it and make your work more sense in this filed, would you consider it？

Finetune for 2D classification task

Hi, This is an interesting work. Can you also provide the code for fine-tune on 2d image classification tasks? The hyper-params are missing for these downstream 2d tasks. Thanks a lot.

3d image classification on RICORD

Would you please provide the code for fine-tuning on 3d image classification tasks?That will be really helpful.Thanks a lost.

Resample CT volumes to a unified voxel size of 1.0×1.0×3.0 mm3.

Thank you for your contribution and I can't wait to try running this code, but I don't know how to Resample CT volumes to a unified voxel size of 1.0×1.0× 3.0mm3. Could you tell me what I should do?

Recommend Projects