You can’t reconstruct a 3D scene from a 2D projection unless you take educated guesses.
E.g. When you look at a picture with something that looks like a chair, you assume that it’s indeed a chair, and then you can estimate its size/pose/etc. But there are infinitely many non-chair shapes that would produce the exact same projection. It’s just that you won’t encounter them in real life, except maybe in trickshots like this: https://youtu.be/SKpOKXAVjGo
E.g. When you look at a picture with something that looks like a chair, you assume that it’s indeed a chair, and then you can estimate its size/pose/etc. But there are infinitely many non-chair shapes that would produce the exact same projection. It’s just that you won’t encounter them in real life, except maybe in trickshots like this: https://youtu.be/SKpOKXAVjGo