IMO, your edits begin to uncover a statistical question, but maybe didn't go far enough to convince the reviewers. If all you really/eventually/now want to ask is "Can a filter size of 8 suffice for SSIM?" (or similar, like "what's the minimal reasonable", or "what effect does a small size", etc.) then ask that without any reference to tensorflow (or maybe some minimal such reference, for the default there).
3
(I don't actually know much about SSIM, so grain of salt. And I do worry whether there's an objectively correct answer to some of the phrasings I listed.)
BTW @kjetilbhalvorsen, may I ask you whether "sufficient dimension reduction" is associated with sufficiency or sufficient statistic? I am not familiar with the term and I am not seeing how it's related to the latter.
