bincount in class_percentages fails on float labels

[perctrix]

segmentation.py:142 - torch.bincount requires integer tensors but label can be float after resampling/padding.

[ATATC]

How did you trigger this error? I used align_spacing=True without seeing any error.

[perctrix]

Some transforms such as MONAI's MapLabelValue convert label dtype from torch.long to torch.float32. Then class_percentages calls bincount which rejects float tensors.

[ATATC]

Then I would say the way out is to add ToLong at the end of a Compose. Trying to avoid redundent ops to save work for the upcoming optimization.