ScanNet
ScanNet is a large-scale RGB-D video dataset designed for 3D scene understanding containing 2.5 million views across more than 1500 scans. The dataset provides high-quality annotations including 3D camera poses, surface reconstructions, and instance-level semantic segmentations for diverse indoor environments. Data is organized by RGB-D sequences, with each scene stored in a directory containing compressed binary sensor streams, camera trajectories, and reconstructed mesh files. Key assets include cleaned and decimated meshes for semantic annotations, over-segmentation files, and aggregated instance-level metadata in JSON format. Users can access per-vertex segment indices, object groupings with NYU-40 labels, and 2D projections of both raw and filtered annotations as PNG images. The dataset supports various data formats such as binary PLY for meshes and custom compressed binary for sensor streams. It is widely used for developing and evaluating algorithms in semantic segmentation, 3D reconstruction, pose est