Scene extrapolation is a challenging variant of the scene completion problem, which aims to predict the missing part(s) of a scene. While the 3D scene completion algorithms in the literature try to fill the occluded part of a scene such as a chair behind a table, this study focuses on extrapolating the available half-scene information to a full one. Our approaches are based on convolutional neural networks (CNN) and generative adversarial networks (GAN). As input, the half of 3D voxelized scenes is taken, then the models complete the other half of scenes as output. The tried models include CNN, GAN, U-Net and hybrid of 3D and 2D inputs. Some of these models are listed below. Read the full paper here.
Read the full details of models, approaches, results, discussion, limitations in my MSc thesis.