Marine underwater
uScenes
Underwater RGB and 3D sonar perception
Temporally paired RGB images and metric 3D multibeam sonar observations from a BlueROV2 at a freshwater site: 110 scenes, 95,834 pairs and 277.6 minutes; modalities are not spatially registered.
Access links
Primary places to inspect, download, read about, or reproduce the dataset.
Quick facts
- Vehicle type
- Marine underwater
- Environment
- Not reported
- Data origin
- Real-world
- Sensor count
- 1
- Ground truth
- Not reported
- Calibration
- Partial Raw Data
- Annotations
- Available
Sensor overview
High-level sensor availability before the detailed sensor records below.
Ground truth
Reference scope, method, rate, coverage, and provenance.
Not reported
Calibration and synchronization
Reported calibration level, reproducibility signals, and supporting notes.
- Level
- Partial Raw Data
- Processed parameters
- Available
- Raw calibration data
- Available
- Calibration targets
- 8 x 8 white pin circle grid; four dimension-constrained concrete blocks; round-headed pin board; two flat-headed screw boards at different depths.
Reproducible calibration data
Raw calibration data is reported, so users can reproduce or inspect the calibration process.
Notes
Underwater pinhole intrinsics and five distortion coefficients are reported in Table II. Extrinsic methodology uses EPnP then nonlinear refinement, but the current release is explicitly not spatially registered. Projection tasks require an independently established and validated optic-acoustic transform. Flat-port physics are not fully validated.
Known or intentional limitations
Reported constraints and characteristics to check before using the dataset.
Temporally paired observations are not spatially registered.
Single robot and freshwater site; cross-platform, location and saltwater generalization is not established.
Scene descriptions are coarse metadata; dense object annotations are absent.
Acoustic strength is normalized within each frame and should not be directly compared across frames.
Sequential sonar acquisition can distort point clouds under vehicle or object motion.
The physical effect of the flat-port camera housing is not fully validated.
Cameras
rgb
- Model
- Not reported
- Setup
- Mono
- Count
- 1
- Effective cameras
- 1
- Modality
- Rgb
- Resolution
- 1280 x 720
- Rate
- Not reported
- Shutter
- Not Reported
- Lens type
- Not Reported
- HFOV
- Not reported
- VFOV
- Not reported
- Notes
- DWE underwater RGB camera; exact part number and shutter are not reported. Capture is approximately 30 Hz, but the release pairs nearest images with approximately 6 Hz sonar observations; no uniform delivered RGB cadence is asserted. JPEG images. Timestamp pairing does not imply spatial registration.
IMUs
Not reported
GNSS
Not reported
LiDAR
Not reported
Radar
Not reported
Pitot tube
Not reported
Additional sensors
Other sensor
- Model
- Water Linked Sonar 3D-15
- Count
- 1
- Notes
- 3D multibeam sonar, acquired at approximately 6 Hz from 256 x 64 range observations with 90-degree horizontal and 40-degree vertical field of view. Released metric binary floating arrays contain [x,y,z,s] rows; s is range-compensated and independently normalized per observation, not calibrated reflectivity. Frames contain 1 to 12,627 returns (mean 6,805). Vendor axes are forward/right/down; binary files reverse the lateral sign and require diag(1,-1,1) before a vendor-frame rigid transform.
Annotations
Reported annotation availability and task support.
- Availability
- Available
- Format
- Not reported
- Class count
- Not reported
Semantic segmentation
Not reported
Instance segmentation
Not reported
Object detection
Not available
Optical flow
Not reported
Depth ground truth
Not reported
Citation
Citation key and BibTeX kept at the end of the page for reference.
@misc{dong2026uscenes,
title={uScenes: A Multimodal RGB and 3D Sonar Dataset for Underwater Robot Perception},
author={Dong, Trung Tien and Wu, Zhenqi and Penumarti, Aditya and Zhang, Zi-Hao and Bartlett, Micaiah and Shin, Jane and Lin, Xiaomin},
year={2026},
eprint={2608.27795},
archivePrefix={arXiv},
primaryClass={cs.CV},
note={Version 1, 28 August 2026}
}
Dataset corrections
Notice a specific issue after reviewing the page?
Report an inaccuracy