>>12817258I have some binary masks (matrixes of 1's and 0's) that are used to block some unwanted data on images, which will be used in a classification task.
The thing is, it may be possible that these masks are actually carrying information that lets classifiers discriminate within classes, which is something I don't want.
So, I wanna standardize these masks in order to remove that information. For that, and since I can't remove the masks (information below them is lost), I need to cluster masks by shape/size. I'm using Intersection over Union as my measurement of overlapping between masks. I calculated the IoU between every pair of masks, and that's the matrix I'm using.
