I noticed that the outcomes of scipy.ndimage.laplace and CV2.Laplacian of a given RGB array are quite different (normalized or not, if normalized then the array is in float64 format). It seems that these two implementations use the same default kernel, so how to explain the difference?
Is it possible for scipy (in my case it's cupyx.scipy.ndimage) to get the same outcome as CV2.Laplacian?