@apunjani if you have a spare minute could you comment on how the window is normalized?
Is it as simple as
window *= np.sum(window) / np.prod(img.shape) or something like that?
Has anyone else had any experience with switching this setting to false?
Can’t comment re the definition, but in one case (a crowded, small thing with one view in particular that had quite low contrast), setting this to false allowed said view to pop up in the 2D classification where it was previously absent.
@olibclarke I too had a case where turning it off was clearly beneficial. I asked about it because I wanted to understand the rationale for doing it at all.