ScanNet

ScanNet is a richly-annotated RGB-D video dataset of indoor scenes, introduced in 2017 by researchers from Stanford, Princeton, and Technical University of Munich. It contains 2.5 million views across 1,513 scenes captured in 707 distinct indoor spaces (offices, apartments, libraries, classrooms, etc.). Each scene provides 3D camera poses, surface reconstructions, and voxel-level semantic segmentations covering 20 object categories. The annotation pipeline combines automated reconstruction with crowdsourced semantic labeling. ScanNet has become one of the most widely-used datasets in 3D computer vision, serving as the benchmark for 3D semantic segmentation, 3D object detection, scene classification, and RGB-D SLAM evaluation. It has accumulated over 4,000 citations and underpins a large portion of 3D scene understanding and embodied AI research.