ScanObjectNN 3D Point Cloud Dataset

ScanObjectNN is a real 3D point cloud classification dataset that is extremely challenging due to background, missing parts, and deformation characteristics. It consists of occluded objects extracted from real-world indoor scans and contains 2,902 3D objects from 15 categories.

The original object is represented by a list of points with global and local coordinates, normals, color attributes and semantic labels. The dataset also provides partial annotations.