ScanNet is a large-scale RGB-D dataset containing 2.5 million views in 1513 scenes, annotated with 3D camera poses, surface reconstructions, and semantic segmentations. The dataset was collected using an easy-to-use and scalable RGB-D capture system, which includes automated surface reconstruction and crowdsourced semantic annotation. The paper introduces a design for efficient 3D data capture and annotation suitable for novice users, and demonstrates the effectiveness of the dataset on several 3D scene understanding tasks, including 3D object classification, semantic voxel labeling, and CAD model retrieval. The contributions of the paper include a large 3D dataset, a design for efficient 3D data capture and annotation, and improved performance on 3D scene understanding tasks using deep learning methods. The dataset and framework are made available to the community to advance research in 3D scene understanding.ScanNet is a large-scale RGB-D dataset containing 2.5 million views in 1513 scenes, annotated with 3D camera poses, surface reconstructions, and semantic segmentations. The dataset was collected using an easy-to-use and scalable RGB-D capture system, which includes automated surface reconstruction and crowdsourced semantic annotation. The paper introduces a design for efficient 3D data capture and annotation suitable for novice users, and demonstrates the effectiveness of the dataset on several 3D scene understanding tasks, including 3D object classification, semantic voxel labeling, and CAD model retrieval. The contributions of the paper include a large 3D dataset, a design for efficient 3D data capture and annotation, and improved performance on 3D scene understanding tasks using deep learning methods. The dataset and framework are made available to the community to advance research in 3D scene understanding.