Not sure if this is expected behavior, but I was using the pixel classifier to create detection objects, and in one case, a detection wrapped around, but did not overlap with, a hole in the parent annotation.
This is inside of a simple tissue detection, the red holes shown are holes in the simple tissue detection where there is whitespace/lipids. I am guessing this is due to the centroid being used when selecting “Insert into hierarchy,” but I had been thinking the position in the hierarchy would be based on overlap. Which is almost certainly much more computationally intensive…