Question about the dataset splits

[Mzczz]

Hi, thanks for sharing the SCSegamba code and datasets.

I have three questions about the dataset splits:

1. How were the train/validation/test sets created? Were they randomly split, stratified, or grouped by image source or scene? What random seed was used?

2. For DeepCrack, why does the released split contain 373/58/106 images instead of the commonly used 300/237 split? Are the results in the paper based on the released split?

3. Could you share the split-generation script or filename lists so that the experiments can be reproduced exactly?

Thanks for your help!