Deep Networks and Random Forests for Semantic Segmentation on 3D Data

This work focus on semantic segmentation over 3D data,firstly by research and study of the state of the art, then by developing methods to transfer features between FuseNet and 3D Entangled Forests.We were able to obtain meaningful insights with regards to the dataset and used learning models, pointing out a possible saturation of the performance due to their combination.We successfully demonstrated usefulness of transferring features and their usability obtaining a performance improvement.