Imagine playing a video game inside your favourite movie, with scenes from the movie exactly recreated in all their detail. Or playing a game at a historical site, building or city that has since been destroyed, with photorealistic appearance as it would have appeared. The goal of this project is to combine state-of-the-art 3D computer vision and procedural content generation to create game-ready scene models and assets from movies, contemporary photos, plans or works of art. 3D reconstruction techniques such as structure-from-motion or deep monocular depth estimation can be used to reconstruct raw models of the observed part of the scene. Deep learning based methods will then be used to extrapolate and clean the models to produce complete scene layouts with photoreal textures.

Sample References:
https://github.com/skanti/scenecad
https://github.com/nianticlabs/monodepth2