>>12845219Look at here.
First, you have differential equation on neighbouring pixels, to define edges.
You then flip equation, that's projecting shapes to 2D, therefore rotation of cube is inversed...
Then you straight up those edges to lines, count which cube could be rotated like that, what's it's rotation according to shift of line...
Then you map "textures of shapes" to this newly generated 3D cube.
You can use shadows and highlits to deduce source of light, and then reduce it to HDR or whatever you want...
With this HDR you have some shape, that's probably origin of projection....
If it was cube, however encraved shaper there...
However, it's that it doesn't work well on very abstract shapes painted with vanta black, but otherwise it gives you 3D description of objects on picture.
Size's relative, it should be that cameras have some form of laser distance or something, or infra flashlight that is then capcured and distances are deduced, so you know precise picture.
