Explicit modeling of the local 3D structure in point clouds can significantly improve the performance of deep learning models for tasks such as segmentation, classification, and detection.
