Annotation Tool and Urban Dataset for 3D Point Cloud Semantic Segmentation

Deep learning requires a significant amount of annotated training data in order to accurately separate unstructured 3D point clouds using semantics. Nevertheless, there isn't any free specialist software on the market right now that can effectively annotate big 3D point clouds. By providing PC-Annotate, a free annotation tool for 3D point cloud research, we close this gap. The suggested solution provides essential functions of point cloud registration and the creation of volumetric samples that may be readily understood by modern deep learning point cloud models, in addition to enabling systematic annotation with a variety of essential volumetric shapes. A sizable outdoor urban dataset for 3D semantic segmentation is also introduced by us. The Ouster LiDAR was used to capture the required dataset, PC-Urban, which was then PC-Annotate labeled. It has 25 annotated classes, 66K frames, and more than 4.3 billion points. Finally, we present PC-Urban baseline semantic segmentation outcomes for well-liked contemporary techniques