Mixing Deep Networks and Entangled Forests for the Semantic Segmentation of 3D Indoor Scenes

This work focuses on semantic segmentation over indoor 3D data, that is, to assign labels to every point in the point clouds representing working spaces: after researching the current state of the art, traditional approaches like random forests and deep neural networks based on PointNet are evaluated. The Superpoint Graph architecture and the 3D Entangled Forests algorithm are selected for mixing their features to try to enhance their performance.