3D-BEVIS: Bird's-Eye-View Instance Segmentation

Recent deep learning models achieve impressive results on 3D scene analysistasks by operating directly on unstructured point clouds. A lot of progress wasmade in the field of object classification and semantic segmentation. However,the task of instance segmentation is less explored. In this work, we present3D-BEVIS, a deep learning framework for 3D semantic instance segmentation onpoint clouds. Following the idea of previous proposal-free instancesegmentation approaches, our model learns a feature embedding and groups theobtained feature space into semantic instances. Current point-based methodsscale linearly with the number of points by processing local sub-parts of ascene individually. However, to perform instance segmentation by clustering,globally consistent features are required. Therefore, we propose to combinelocal point geometry with global context information from an intermediatebird's-eye view representation.