Indoor Scene 3D Reconstructions with Rich Annotations

The ScanNet dataset provides richly-annotated 3D reconstructions of indoor scenes for computer vision tasks. It is designed for 3D object detection and 3D segmentation applications. The dataset was created by researchers including Angela Dai and Angel X. Chang, with foundational work published at CVPR in 2017.

Use Cases

• Train 3D object detection models using annotated 3D bounding boxes within indoor scene reconstructions.
• Develop 3D semantic segmentation models on point cloud data from indoor environments.
• Benchmark 3D reconstruction algorithms on a standard dataset of indoor scenes.
• Perform scene understanding research using the provided 3D mesh and annotation data.

Strengths

• Dataset is a standard benchmark for 3D indoor scene understanding tasks.
• Provides 3D reconstructions with rich annotations for multiple computer vision tasks.
• Published and cited foundational research paper from CVPR 2017.

Limitations

• Dataset size, specific row/column counts, and file formats are not specified in the provided input.
• The provided description lacks concrete details on the number of scenes or annotations.
• The input does not specify the license, which is critical for determining permissible use.

Provenance

Source

Original research by Dai, Chang, Savva, Halber, Funkhouser, and Nießner, published at CVPR 2017.

Collection Method

3D reconstructions and annotations of indoor scenes, methodology detailed in the cited publication.

Freshness

The dataset entry was last updated on the platform in October 2025.