A machine learning based image processing and generation framework is disclosed. In some embodiments, depth values of an object or asset in a received input image are at least in part determined using a machine learning based framework that is constrained to a known prescribed environment. Determined depth values facilitate generation of other views of the object or asset.