Validation Datasets and Splits

[timat33]

Hi there, thanks a lot for all your work!

I would like to perform some benchmarking on the model and would like to reproduce some of the results in the quantitative evaluation section (4.2) of the paper. I'd be very interested to know the composition of the training set (I see that the 9 held-out test sets are explicitly listed), in particular, since I can't find them listed on the SA-Med2D-20M paper, the training/validation splits. Could you help me with this?

Many thanks,
Tim

[Cjl-MedSeg]

All nine datasets used for zero-shot testing were released after 2023, and we are not authorized to share them. To evaluate these datasets, we first convert them into the same data format as used during training (including preprocessing steps) and then conduct the corresponding tests.