ConSlide

[ICCV 2023] ConSlide: Asynchronous Hierarchical Interaction Transformer with Breakup-Reorganize Rehearsal for Continual Whole Slide Image Analysis.

Training Data Preparation

We mainly follow the pipeline of CLAM. The modified version of the CLAM code for data preparation will be released later.

Training Example

python utils/main.py --state train --model conslide --dataset seq-wsi --exp_desc conslide --buffer_size 1100 --alpha 0.2 --beta 0.2

Updates / TODOs

Please follow this GitHub for more updates.

• Refine the code.
• Provide code for data preparation.
• Remove dead code.
• Better documentation on interpretability code example.

Reference

If you find our work useful in your research please consider citing our paper:

Huang, Y., Zhao, W., Wang, S., Fu, Y., Jiang, Y., & Yu, L. (2023). ConSlide: Asynchronous Hierarchical Interaction Transformer with Breakup-Reorganize Rehearsal for Continual Whole Slide Image Analysis. In Proceedings of the IEEE/CVF International Conference on Computer Vision (pp. 21349-21360).

@inproceedings{huang2023conslide,
  title={ConSlide: Asynchronous Hierarchical Interaction Transformer with Breakup-Reorganize Rehearsal for Continual Whole Slide Image Analysis},
  author={Huang, Yanyan and Zhao, Weiqin and Wang, Shujun and Fu, Yu and Jiang, Yuming and Yu, Lequan},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={21349--21360},
  year={2023}
}

Acknowledgements

Framework code for Continual Learning was largely adapted via making modifications to Mammoth