Multi-Camera Underwater Visual-Inertial Dataset

About

This repository hosts the Multi-Camera Underwater Visual-Inertial Dataset collected by the Autonomous Robots Lab, NTNU.
The dataset features data from a variety of onboard sensors, including:

5 cameras
1 IMU (via the Alphasense Core Research Development Kit)
A barometer
Actuator commands from the ROV

Data was collected across both controlled indoor facilities and natural outdoor underwater environments.

Updates / News

April 28, 2025:
We are uploading new datasets collected in the Trondheim Fjord and at the Marine Cybernetics Lab Pool.
Additionally, we are migrating the dataset repository from github.com/ntnu-arl/underwater-datasets to huggingface.co/datasets/ntnu-arl/underwater-datasets.

Quick Start

You can quickly download the dataset using the huggingface_hub Python library.

1. Install `huggingface_hub`

pip install huggingface_hub

2. Download the Dataset

You can clone the entire dataset repository using:

from huggingface_hub import snapshot_download

# Download the dataset
snapshot_download(
    repo_id="ntnu-arl/underwater-datasets",
    repo_type="dataset",
    local_dir="underwater-datasets",
    local_dir_use_symlinks=False  # Set to False to avoid symlinks
)

This will download the dataset to a local folder called underwater-datasets/.

Note: The full dataset is large (many GBs), so ensure you have enough storage and a stable internet connection.

Alternative: Using `git-lfs`

If you prefer, you can also clone it manually with git-lfs:

sudo apt install git-lfs  # Install git lfs
git lfs install           # Initialize git-lfs
git clone https://huggingface.co/datasets/ntnu-arl/underwater-datasets   # Clone the dataset

3. Use data with underwater state estimation methods

You can try out the dataset on the following undewater state estimation methods:

ReAqROVIO: Refractive Aquatic ROVIO code, homepage
DeepVL: Deep Velocity Learning code, homepage

Platform: Ariel — Underwater Robot with Multi-Camera, IMU, and Compute Suite

The data was gathered using a custom-built underwater robot based on the BlueROV2 Heavy Configuration.

Ariel Underwater Robot

Trajectories with Visual-Inertial Data

General Structure

The dataset is organized into multiple subsets, each corresponding to a different environment.
Each subset directory (e.g., subset-fjord) contains several trajectory directories (traj_<X>), each containing:

A .bag file with raw sensor data
A .tum file with the reference trajectory obtained by running ReAqROVIO using four cameras and the IMU.

Directory structure overview:

underwater-datasets/
├── subset-<environment>/
│   ├── traj_1/
│   │   ├── <trajectory>.bag            # Main ROS bag file
│   │   ├── <trajectory>_baseline.tum    # Reference trajectory (ReAqROVIO output)
│   ├── traj_2/
│   │   ├── <trajectory>.bag
│   │   ├── <trajectory>_baseline.tum
│   └── ...

Available Subsets

Subset: Trondheim Fjord (`subset-fjord`)

This subset includes six trajectories collected by manually piloting Ariel in the Trondheim Fjord.

No.	Length (m)	Duration (s)	Size (GB)
1	142	312	12.5
2	206	499	23.4
3	122	272	13.9
4	165	411	23.4
5	234	440	18.6
6	305	638	26.6

Subset: Marine Cybernetics Lab (`subset-mclab`)

This subset includes two trajectories collected by manually piloting Ariel in the Marine Cybernetics Laboratory at NTNU.

No.	Length (m)	Duration (s)	Size (GB)
1	135	390	15.4
2	125	304	15.0

Calibration

Camera IMU

The cameras were calibrated using the Kalibr toolbox and the calibrations are provided in the output format of Kalibr. The calibration data can be found in the calibrations.

Barometer

The underwater barometer used is the robot uses the follow mapping to obtain depth in meters for fresh water:

barometer_pressure_offset = 2660.0
barometer_pressure_scale = 241.0
depth = - (barometer_measurement - baro_pressure_offset_) / baro_pressure_scale

Reference

If you use this dataset in your research, please cite the following publication:

An Online Self-Calibrating Refractive Camera Model with Application to Underwater Odometry
Mohit Singh, Mihir Dharmadhikari, Kostas Alexis
IEEE International Conference on Robotics and Automation (ICRA), 2024

@INPROCEEDINGS{10610643,
  author={Singh, Mohit and Dharmadhikari, Mihir and Alexis, Kostas},
  booktitle={2024 IEEE International Conference on Robotics and Automation (ICRA)}, 
  title={An Online Self-calibrating Refractive Camera Model with Application to Underwater Odometry}, 
  year={2024},
  pages={10005-10011},
  doi={10.1109/ICRA57147.2024.10610643}
}

Contact

For questions or further information, feel free to reach out: