OA-Stereo Opti-Acoustic Underwater Stereo Dataset

Overview

This dataset accompanies the paper:

OA-Stereo: Self-Supervised Opti-Acoustic Stereo for Robust 3D Perception of Underwater Vehicles (Accepted by IEEE Robotics and Automation Letters (RA-L), 2025)

This dataset is collected based on the DAVE underwater robotics platform, and provides calibrated optical stereo images and forward-looking sonar data for studying opti-acoustic stereo perception, dense depth estimation, and robust 3D reconstruction in underwater environments.

🔗 Project Page (Code & Models): https://github.com/c237814486/OA-Stereo

Sensor Modalities

Each scene contains synchronized and calibrated multi-modal data:

Optical Cameras

• Left and right RGB images
• Camera intrinsics
• Stereo extrinsic parameters

Sonar

• Forward-looking imaging sonar data
• Acoustic projection parameters
• Sonar view images used for cross-modal supervision

Depth / Disparity

• Metric depth and disparity maps are provided for each scene.
• Depth and disparity are accurately aligned with the corresponding RGB cameras.

Dataset Structure

├── dave_sonar
│   ├── scene1
│   │   ├── cam_left/              # Left camera RGB images
│   │   ├── cam_right/             # Right camera RGB images
│   │   ├── sonar/                 # Raw sonar images
│   │   ├── sonar_view/            # Rendered sonar views
│   │   ├── sonar_view_origin/     # Original sonar projections
│   │   ├── depthl/                # Left-view depth (optional)
│   │   ├── displ/                 # Left-view disparity (optional)
│   │   ├── cam_left.yaml          # Camera intrinsics
│   │   ├── cam_right.yaml
│   │   ├── cam_depth.yaml
│   │   └── sonar.yaml             # Sonar parameters
│   ├── scene2
│   ├── scene3
│   └── scene4

Citation

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

@article{cao2025oa,
  title={OA-Stereo: Self-Supervised Opti-Acoustic Stereo for Robust 3D Perception of Underwater Vehicles},
  author={Cao, Yaozhong and Bai, Xuejian and Chu, Hongfei and Wang, Shuo and Tan, Min and Wang, Yu},
  journal={IEEE Robotics and Automation Letters},
  year={2025},
  publisher={IEEE}
}

📊 Structured Schema (Zero-Fabrication)

Estimated Rows: 43,724