I "3d'd" some old war video, it learns "normalization" factors off pushing out a blurred sobel filter.
It works okish sometimes.
https://www.youtube.com/watch?v=0TmkVkKS5MQ&t=159s
Its only prototype, and its not very deep just yet, but i havent had the chance to code the better version, i think this tech should go all the way.
It works through scale invarience, between the pixels.
Start small
You might be on to something. Suggest starting with video which is a lot less complex. As you’re aware, comparing frames (still images) of a video may certainly be used to create a 3D image, but can certainly imaging starting with a complex image might be really difficult. You might start with a cube moving against a solid background.
Looking forward to seeing your progress.
