Called depth estimation from single image. There's ongoing research, here's one example: [0]

Using only the segmented 2D image is what the selfie cam does by the way, at least on Pixel 2 [1].

[0] https://papers.nips.cc/paper/5539-depth-map-prediction-from-...

[1] https://ai.googleblog.com/2017/10/portrait-mode-on-pixel-2-a...