PyTorch implementation of the paper

• Noriaki H. and Daisuke F. et al., Multi-scale Domain-adversarial Multiple-instance CNN for Cancer Subtype Classification with Unannotated Histopathological Images, CVPR2020 Proceeding [link]

I confirmed that the source code was running with the following environment.

• Python3.6
• pytorch 1.4.0
• CUDA 10.0
• NVIDIA Quadro RTX 5000
• and python library in requirements.txt

There is no image data here. Therefore, you need to edit the model/dataset.py to fit your data.

Here, I'll explain the case of using two magnifications. ('scale1' and 'scale2')

First, run single scale learning(DA-MIL) for each magnification.

$ python single_scale_learning.py scale=scale1

$ python single_scale_learning.py scale=scale2

After run, parameter-files DAMIL_params_scale1.pth and DAMIL_params_scale2.pth are generated in tmp_storage/.

After each single scale learning, run multi scale learning (MS-DA-MIL).

$ python multi_scale_learning.py