I imagine this is something AI could do if designed and trained to do it. You show it 3D models and also pictures of said 3D models at different angles and it would eventually learn. It would need to be designed for this specific purpose.
Start with something simple like an image of a cube and then a 3D model of a cube. It would eventually be able to tell that the lines represent edges then make a 3D version of it after seeing many 3D versions.