I looked at the Lproduct loss equation, and noticed a strange property - this function seem non convex. E.g. for a GT=0.05 in 1D it would look this (when combined with L1 loss):
I know this actually comes from another paper, but maybe during the work on your analysis you encountered and analysed this part.
Hello,
I looked at the Lproduct loss equation, and noticed a strange property - this function seem non convex. E.g. for a GT=0.05 in 1D it would look this (when combined with L1 loss):
I know this actually comes from another paper, but maybe during the work on your analysis you encountered and analysed this part.
I wonder if this might be related to issues with divergence e.g. https://github.com/JunMa11/SegWithDistMap/issues/19 because this might be a problem, even if it works most of the time.