Pretrained Weights for Reproduction

[casperliuliuliu]

Thanks you for your excellent work!

The methods and results described in your paper are extremely valuable for my research. I am currently attempting to reproduce the results of your experiments.
While the provided code is a great resource, the lack of pretrained weights for these components has made the reproduction quite challenging.

Would it be possible for you to release the pretrained weights for the Region-level Encoder, WSI-level Encoder, Transformer Decoder and other modules?

[dddavid4real]

Thank you for your interest.

Since we have several near deadlines on hand, the checkpoint and other codes (e.g. for downstream tasks) should be uploaded in mid/late December.

Actually, the train_wsi_report.sh script is set with the optimal hyperparameters. So if you have the resources to train the model, you could try training it and see if there is inconsistency with the results shown in the paper.