A new version of implementation has been released, integrating Swin-MIL and SA-MIL together, enhancing the code's usability. You can find it at https://github.com/Nexuslkl/SA-MIL.
The Swin Transformer backbone is initialized by ImageNet-1K pretrain model of Swin-T. The pretrain model mamed "swin_tiny_patch4_window7_224" can be download from https://github.com/microsoft/Swin-Transformer.
@article{qian2022transformer, title={Transformer based multiple instance learning for weakly supervised histopathology image segmentation}, author={Qian, Ziniu and Li, Kailu and Lai, Maode and Chang, Eric I and Wei, Bingzheng and Fan, Yubo and Xu, Yan and others}, journal={arXiv preprint arXiv:2205.08878}, year={2022} }