Other
Matterport3D
Building-scale panoramic RGB-D scene understanding
Stationary multi-camera scans of buildings with aligned views, reconstructed surfaces and object/region annotations.
Access links
Primary places to inspect, download, read about, or reproduce the dataset.
Quick facts
- Vehicle type
- Other
- Environment
- Indoor
- Data origin
- Real-world
- Sensor count
- 6
- Ground truth
- Available
- Calibration
- Parameters Only
- Annotations
- Available
Sensor overview
High-level sensor availability before the detailed sensor records below.
Ground truth
Reference scope, method, rate, coverage, and provenance.
| Scope | Method | Reference system | Count | Rate | Coverage |
Reference provenance
|
|---|---|---|---|---|---|---|
| 6DoF | Structure from motion | Matterport global bundle adjustment | Not reported | Not reported | Full |
Reference provenance
Derived
|
|
||||||
Calibration and synchronization
Reported calibration level, reproducibility signals, and supporting notes.
- Level
- Parameters Only
- Processed parameters
- Available
- Raw calibration data
- Not reported
- Calibration targets
- Not reported
Parameters only
Processed calibration parameters are reported, but raw calibration data is not reported.
Notes
Official file manifest provides original camera intrinsics/poses and undistorted camera parameters. Complete raw calibration sessions and independently measured calibration error are not established.
Known or intentional limitations
Reported constraints and characteristics to check before using the dataset.
Ninety building-scale scenes contain 10,800 panoramas and 194,400 paired RGB-D views. Three RGB-D units rotate through six stops; eighteen views are not eighteen physical cameras or a fixed-rate walking sequence.
Camera poses and building meshes are reconstructed from the same observations. The authors lack independent pose truth; subjective surface registration is not a certified trajectory-accuracy bound.
Depth files encode axial z distance in 0.25 mm increments, not Euclidean ray distance or a demonstrated 0.25 mm sensor accuracy. Missing values and reconstruction errors remain relevant.
Forty canonical object classes differ from freeform object labels and room/region taxonomies. Capture cadence, raw calibration sessions and independent pose/depth errors remain unknown.
Dataset access requires a signed terms request to the authors. Code availability does not grant unrestricted dataset use; no access request was submitted during this review.
Cameras
Camera 1
- Model
- Matterport Pro / PrimeSense Carmine RGB-D
- Setup
- Rgbd
- Count
- 3
- Effective cameras
- 6
- Modality
- Rgbd
- Resolution
- 1280 x 1024
- Rate
- Not reported
- Shutter
- Not Reported
- Lens type
- Not Reported
- HFOV
- Not reported
- VFOV
- Not reported
- Notes
- Three physical RGB-D units provide three RGB and three depth channels. Six stationary yaw stops yield eighteen paired views per panorama; views are not additional physical cameras.
IMUs
Not reported
GNSS
Not reported
LiDAR
Not reported
Radar
Not reported
Additional sensors
Not reported
Annotations
Reported annotation availability and task support.
- Availability
- Available
- Format
- Manual mesh object-instance/category labels; forty canonical object classes distinct from freeform labels and region taxonomy.
- Class count
- 40
Semantic segmentation
Available
Instance segmentation
Available
Object detection
Not reported
Optical flow
Not reported
Depth ground truth
Not reported
Citation
Citation key and BibTeX kept at the end of the page for reference.
@misc{chang2017matterport3d,
author = {{Angel Chang} and {Angela Dai} and {Thomas Funkhouser} and {Maciej Halber} and {Matthias Nießner} and {Manolis Savva} and {Shuran Song} and {Andy Zeng} and {Yinda Zhang}},
title = {Matterport3D: Learning from RGB-D Data in Indoor Environments},
year = {2017},
eprint = {1709.06158},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/1709.06158v1}
}Dataset corrections
Notice a specific issue after reviewing the page?
Report an inaccuracy