I have pairs of images which are not the same class, but are from the same meta-class/group. I have a standard CNN which produces a representation for each sample.

If I have several pairs of images belonging to the same meta-class/group, how do I highlight and find common representation within the output representation.

In a way, I am trying to find invariant features for the two images as they belong to the same meta-class/group ? So if my representation is a flattened vector, of say vector size 2048, how can I highlight values which are common for both pairs ?

My ultimate goal: Find common invariant areas within representation for each meta-class/group.

Any advice on what metric or method I can use to find common related areas within the representation would be extremely useful



You must log in to answer this question.