3 ms·

I think you might be able to use Stable Diffusion with the Depth ControlNet to take your exported image and convert it into a depth map that could then be fed i

• PaulHoule 3y ago
  To do that there is the problem that there are some pixels the left eye sees that the right eye sees and vice versa. Probably you can get an A.I. to fill those pixels but you absolutely have to do it. I think some of the appeal of a good stereogram isn't just that it has the illusion of depth but that it literally has more information than a flat photo because of this, I think it affects emotional scenes in movies since you get to see another 5-10% of the muscles around the edge of the face. Personally I'm more interested in trying to pick out light rays that arrived on the left vs the right side of the camera (synthesize two apertures as opposed to synthesize a narrower or wider aperture)
  • nickthegreek 3y ago
    Ive never seen a stereogram where I can see the emotion in someone's face, that sounds amazing. I used to use a xbox kinect to produce depth data for my stereograms in my younger days, but I could never get fine detail. Do you happen to have a url handy of some top level stereograms?
• orbital-decay 3y ago
  That's not Stable Diffusion though, it's a different third-party model. (MiDaS, for example, but there are many of those) Generally, good depth maps is a holy grail of imaging, but the issue is that it's very sensitive to resolution and correctness, so it must be perfect and have better resolution than the picture to achieve good results. ML-powered depth estimation isn't there yet.