Poor results when using previous binary segmentation

Hello!

When I try to input a previous binary segmentation into SAM I observe poor results compared to using logits. Is there any way to perform corrections on a pre-existing binary segmentation?

Start with the repository's example notebooks and reproduce the reported difference between a previous binary segmentation and logits when used as SAM input. Determine whether correction from a pre-existing binary segmentation is supported; done would require a clearly defined correction path and evidence that it improves the reported results.

Written by the indexing model from the issue text.