Clarification on preprocessing and reproduction of BTCV results

[tal-grossman]

Hi, and thank you for sharing your excellent work on FATE-SAM. Both the paper and the code are impressive and very valuable for the community.

I have been testing the provided BTCV abdomen example and noticed that the segmentation results are significantly poorer than those reported in the paper. It seems that the example data attached in the repository might not have undergone any preprocessing, and this could be the reason for the performance gap.

When I run the same pipeline on my own preprocessed BTCV data, the results improve somewhat compared to the provided example data, but they are still quite far from the paper’s reported performance.

Could you please clarify:

1. Was any specific preprocessing applied to the BTCV dataset before training or inference (for example, normalization, resampling, cropping, intensity scaling, etc.)?

2. Are there any post-processing steps required for evaluation?

Thanks again for the great work. I would really appreciate any guidance on reproducing the BTCV results more closely.

Best regards,
Tal