Datasets

Public datasets let you develop and compare perception, navigation, and mapping algorithms without your own sea time. Links and contents were checked in October 2026. Check each dataset's license before use: many are for non-commercial research only, and some state no license at all.

Quick finder

I want to work on...	Start with
Sonar object detection or classification	Marine Debris FLS, UATD, AI4Shipwrecks; the OpenSonarDatasets index
Underwater SLAM with sonar	Girona Underwater Caves, Aurora
Visual or visual-inertial navigation underwater	Aqualoc, NTNU-ARL multi-camera, FLSea, VAROS (synthetic), MIMIR-UW (synthetic)
Underwater image segmentation or enhancement	SUIM, EUVP, UIEB, TrashCan
USV obstacle detection	LaRS, MODS, MaSTr1325, WaterScenes, MassMIND
USV multi-sensor navigation	Pohang Canal, USVInland
Seafloor mapping	Aurora; MGDS archives
Ocean conditions for mission planning	Copernicus Marine, NOAA, Argo

Sonar

Dataset	Contents	Platform	License
OpenSonarDatasets	An index of about 20 public sonar datasets (side-scan, forward-looking, scanning, multibeam, synthetic aperture). Start here	Various	Per dataset
Marine Debris FLS	ARIS Explorer 3000 forward-looking sonar images for classification, segmentation (11 classes), and matching	Water tank and turntable	Not stated
UATD	About 9,200 forward-looking sonar images with object boxes	Multibeam FLS	See record
AI4Shipwrecks	Side-scan sonar images of shipwrecks with segmentation labels	AUV side-scan	See site
SubPipe	Side-scan sonar, camera, and navigation for pipeline inspection	AUV	See record

Underwater navigation and SLAM

Dataset	Contents	Platform	License
Girona Underwater Caves	Two scanning sonars, DVL, two IMUs, depth, and camera, as ROS bags, in an underwater cave	Sparus II AUV	See site
Aqualoc	Monocular camera, MEMS IMU, and pressure sensor, as ROS bags, at harbor and archaeological sites	ROV	See site
Aurora	Multibeam, side-scan, camera, and navigation from the Haig Fras marine protected area	Autosub 6000 AUV	See record
NTNU-ARL underwater datasets	Five cameras, IMU, and pressure, with motion-capture ground truth	Multi-camera sensor rig	BSD-3
FLSea	Forward-looking stereo and visual-inertial sequences with depth ground truth (Mediterranean and Red Sea)	Stereo and visual-inertial rig	See site
Eiffel Tower	Four visits (2015-2020) to a deep-sea hydrothermal vent, for long-term visual localization	Ifremer deep-sea vehicles	See record
Tasmania Coral Point Count	More than 100,000 stereo pairs from 22 dives, with labels	AUV Sirius	Acknowledgment requested

Synthetic data

Dataset	Contents	License
MIMIR-UW	Three cameras, IMU, ground-truth pose, depth, and segmentation in four environments, rendered with AirSim and Unreal Engine 4	Code GPL-3.0; check the data record
VAROS	Images, IMU, depth gauge, ground-truth poses, depth, and normals	See site

Simulators can also generate data with perfect ground truth; see Simulation.

Underwater imagery

Dataset	Contents	License
SUIM	1,635 images with pixel labels in eight classes	See site
EUVP	Paired and unpaired images for underwater image enhancement	See site
UIEB	950 real underwater images for enhancement benchmarks	See site
TrashCan 1.0	ROV images of marine debris with instance segmentation (from the JAMSTEC J-EDI archive)	Academic use; commercial use needs JAMSTEC permission

Surface vehicle perception

Dataset	Contents	License
LaRS	More than 4,000 key frames from lakes, rivers, and sea with panoptic labels, for obstacle segmentation	See site
MODS	Benchmark with about 81,000 stereo frames and IMU, for USV obstacle detection	See site
MaSTr1325	1,325 USV images with pixel labels (sky, water, obstacle)	See site
WaterScenes	4D radar and camera data with detection and segmentation labels	See site
MassMIND	2,916 long-wave infrared images from Boston Harbor with segmentation labels	See repository
SeaDronesSee	Aerial images over open water, for search and rescue detection	See site
Pohang Canal	Three lidars, marine radar, stereo, infrared and omnidirectional cameras, AHRS, and RTK GNSS over a 7.5 km route	CC BY-NC 4.0
USVInland	Lidar, stereo, millimeter-wave radar, GNSS, and IMU over 26 km of inland waterways	See site

Ocean and geoscience data

Source	Contents
Copernicus Marine Service	Ocean currents, temperature, salinity, and waves: analyses and forecasts
NOAA National Centers for Environmental Information	Bathymetry, ocean profiles (World Ocean Database), and more
Argo	Temperature and salinity profiles from a global float array
MGDS (Marine Geoscience Data System)	Archived multibeam bathymetry and other marine geoscience data, with DOIs
GEBCO	Global gridded bathymetry, for planning and simulation terrain

Use these to plan missions, to set realistic currents and sound speed profiles in simulation, and to validate maps.

Share your dataset

A useful dataset release has:

1. Documentation: sensors with models and mounting, calibration, environmental conditions, and how the data was collected.
2. Standard formats: ROS 2 bags in MCAP for multi-sensor data, plus common image and point cloud formats. Metadata in YAML or JSON.
3. Ground truth, with a description of how it was obtained and its accuracy.
4. A clear license (for example CC BY 4.0) and a citation.
5. Permanent hosting with a DOI (for example Zenodo or an institutional repository).

Then add it to this page with a pull request: name, link, contents, platform, and license.