Could you go a bit into their segmentation algorithm that this paper depends on? Do you know how its results might compare to the other segmentation papers we read?
Also, they mention that segments are more invariant in lighting, orientation, and scale than features. Is this intuition necessarily true?