The binary mask keeps track of which parts of the volume have been proofread by the user, which is useful to know which parts of the ground truth are trustworthy enough to do deep learning on. In other words, we should have a way to distinguish from the "starting", automated segmentation labels, and labels which have been manually painted/separated/whatever.
The binary mask keeps track of which parts of the volume have been proofread by the user, which is useful to know which parts of the ground truth are trustworthy enough to do deep learning on. In other words, we should have a way to distinguish from the "starting", automated segmentation labels, and labels which have been manually painted/separated/whatever.