List of Datasets

Handheld

ARKitScenes

Mobile indoor reconstruction, oriented furniture detection and depth upsampling

Indoor mobile RGB-D recordings with documented image/depth variants, furniture boxes and subset survey-based references; standalone native IMU delivery is unverified.

Metadata: Complete Metadata source documented Released 2021

Primary places to inspect, download, read about, or reproduce the dataset.

Quick facts

Vehicle type
Handheld
Environment
Indoor
Data origin
Real-world
Sensor count
4
Ground truth
Available
Calibration
Parameters Only
Annotations
Available

Sensor overview

High-level sensor availability before the detailed sensor records below.

Cameras

3 reported

IMUs

Not reported

GNSS

Not reported

LiDAR

Not reported

Radar

Not reported

Magnetometers

Not reported

Barometers

Not reported

UWB

Not reported

Ground truth

Reference scope, method, rate, coverage, and provenance.

Scope Method Reference system Count Rate Coverage Reference provenance
External
Measured using equipment independent of the platform’s normal onboard sensor pipeline.
Derived
Computed primarily from the dataset’s own recorded sensors.
Hybrid
Combines independent reference equipment with onboard measurements.
6DoF Photogrammetry Image registration to FARO geometry · FARO Focus S70 Not reported Not reported Partial · Registered Keyframes Reference provenance
External
Measured using equipment independent of the platform’s normal onboard sensor pipeline.
Derived
Computed primarily from the dataset’s own recorded sensors.
Hybrid
Combines independent reference equipment with onboard measurements.
Hybrid
Reference notes
Keyframe pose estimation uses released images against independent scans; not an independent continuous tracker.
Other Other Rendered registered survey depth · FARO Focus S70 Not reported Not reported Partial · Depth Upsampling Subset Reference provenance
External
Measured using equipment independent of the platform’s normal onboard sensor pipeline.
Derived
Computed primarily from the dataset’s own recorded sensors.
Hybrid
Combines independent reference equipment with onboard measurements.
Hybrid
Reference notes
1920×1440 reference depth for 2257 captures/841 venues; visibility and registration limit coverage.

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

Per-image .pincam intrinsics are documented for the RGB variants. Full raw calibration sessions and independent extrinsic/timing accuracy remain unestablished.

Known or intentional limitations

Reported constraints and characteristics to check before using the dataset.

This entry is limited to documented RGB-D, reconstruction, trajectory and annotation products. The paper mentions IMU, but the official raw asset list/downloader does not establish standalone native inertial delivery; no IMU stream is claimed.

The 3D object-detection product lists 5,047 captures, while the paper repeatedly gives 5,048 overall. Task products have different membership; aggregate recording count remains unknown.

Low-resolution 60 fps, VGA wide 30 fps and filtered high-resolution/ultra-wide 10 fps products share physical cameras. Not every asset exists at each timestamp; filtering and raw/3DOD editions can alter depth/confidence.

Survey-registered keyposes and rendered high-resolution depth cover subsets and reuse image registration. FARO remains reference equipment; registration error, visibility gaps, reflections and withheld test data limit reference coverage.

Cameras

Camera 1

Model
2020 iPad Pro
Setup
Rgbd
Count
1
Effective cameras
2
Modality
Rgbd
Resolution
256 x 192
Rate
60 Hz
Shutter
Not Reported
Lens type
Not Reported
HFOV
Not reported
VFOV
Not reported
Notes
Documented lowres_wide RGB and lowres_depth are 256x192 at 60 fps, with missing assets possible per timestamp. The same physical wide camera also supplies 640x480 at 30 fps and filtered 1920x1440 images at 10 fps; these are not extra cameras.

Camera 2

Model
2020 iPad Pro ultra-wide
Setup
Mono
Count
1
Effective cameras
1
Modality
Rgb
Resolution
640 x 480
Rate
10 Hz
Shutter
Not Reported
Lens type
Not Reported
HFOV
Not reported
VFOV
Not reported
Notes
Documented ultra-wide RGB at 640x480 and 10 fps; timestamps differ from the filtered high-resolution wide-camera products.

IMUs

Not reported

GNSS

Not reported

LiDAR

Not reported

Radar

Not reported

Additional sensors

Odometry

Name
ARKit estimated camera trajectory
Count
1
Notes
The raw release lists lowres_wide.traj with timestamped axis-angle rotation and metric translation. This device estimate is distinct from survey-registered keyframe references.

Annotations

Reported annotation availability and task support.

Availability
Available
Format
Manual oriented 3D furniture boxes and masked, registered high-resolution depth
Class count
17
Semantic segmentation Not reported
Instance segmentation Not reported
Object detection Available
Optical flow Not reported
Depth ground truth Available

Citation

Citation key and BibTeX kept at the end of the page for reference.

Citation key
baruch2021arkitscenes

BibTeX

@inproceedings{
baruch2021arkitscenes,
title={{ARK}itScenes - A Diverse Real-World Dataset for 3D Indoor Scene Understanding Using Mobile {RGB}-D Data},
author={Gilad Baruch and Zhuoyuan Chen and Afshin Dehghan and Tal Dimry and Yuri Feigin and Peter Fu and Thomas Gebauer and Brandon Joffe and Daniel Kurz and Arik Schwartz and Elad Shulman},
booktitle={Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 1)},
year={2021},
url={https://openreview.net/forum?id=tjZjv_qh_CE}
}

Dataset corrections

Notice a specific issue after reviewing the page?

Report an inaccuracy