PI: Dr. Xi Peng, Associate Professor, Department of Computer and Information Sciences, University of Delaware
Co-PI: Dr. Arthur Trembanis, Professor, School of Marine Science and Policy, University of Delaware
Project: "Advancing Trustworthy AI in Seafloor Mapping," DoD DEPSCoR (FA95502310494), PI: Xi Peng, 2023–2026.
Publication: SeafloorAI: A Large-scale Vision-Language Dataset for Seafloor Geological Survey. NeurIPS, 2024.

🔧 How to Use SeafloorAI Dataset?

Project Homepage;
Read our NeurIPS paper for details;
Open-source data processing code: GitHub;
Download the dataset: Huggingface (this page).

📥 How to Download SeafloorAI Dataset?

Step 1: Fill in Your Online Request (Top)
Step 2: Wait for Approval
Step 3: Download the Data

⭐ Please use SeafloorAI dataset in strict accordance with the CC BY-NC-SA 4.0 license. "Any use of the data must provide proper attribution, be limited to non-commercial purposes, and ensure that any derivative works are distributed under the same license."

🌊 Dataset Abstract

A major obstacle to the advancements of machine learning models in marine science, particularly in sonar imagery analysis, is the scarcity of AI-ready datasets. While there have been efforts to make AI-ready sonar image dataset publicly available, they suffer from limitations in terms of environment setting and scale. To bridge this gap, we introduce SeafloorAI, the first extensive AI-ready dataset for seafloor mapping across 5 geological layers that is curated in collaboration with marine scientists. We further extend the dataset to SeafloorGenAI by incorporating the language component in order to facilitate the development of both visionand language-capable machine learning models for sonar imagery. The dataset consists of 62 geo-distributed data surveys spanning 17,300 square kilometers, with 696K sonar images, 827K annotated segmentation masks, 696K detailed language descriptions and approximately 7M question-answer pairs. By making our data processing source code publicly available, we aim to engage the marine science community to enrich the data pool and inspire the machine learning community to develop more robust models. This collaborative approach will enhance the capabilities and applications of our datasets within both fields.

📊 Dataset Overview

The full dataset comprises 696K sonar images and 827K annotated segmentation masks across 9 regions. This release includes 353K sonar images from 7 regions and the complete set of 827K annotated masks, totaling 178 GB. The dataset is organized into 7 region folders and a split folder that defines the train/validation/test partitions.

Input & Annotation Layers

The dataset includes 11 layers:

Raw Input Signals: Backscatter, Bathymetry, Slope, Rugosity, Longitude, Latitude.
Annotations: Sediment, Physiographic Zone, Habitat, Fault, Fold.

📂 Dataset Structure

The dataset is organized by region with corresponding input signals and annotation layers. Each region contains multi-channel input data and task-specific annotations.

Directory Organization

SeafloorAI/
├── region{1,2,5,6,7}/          # Regions with sediment & physiographic zone
│   ├── input/                  # 6-channel input signals
│   │   └── region*_*.npy       # Shape: (6, 224, 224)
│   ├── sed/                    # Sediment annotations
│   │   └── region*_*.npy       # Shape: (224, 224)
│   └── pzone/                  # Physiographic zone annotations
│       └── region*_*.npy       # Shape: (224, 224)
│
├── region{3,4}/                # Regions with habitat, fault & fold
│   ├── input/                  # 6-channel input signals
│   │   └── region*_*.npy       # Shape: (6, 224, 224)
│   ├── habitat/                # Habitat annotations
│   │   └── region*_*.npy       # Shape: (224, 224)
│   ├── fault/                  # Fault annotations
│   │   └── region*_*.npy       # Shape: (224, 224)
│   └── fold/                   # Fold annotations
│       └── region*_*.npy       # Shape: (224, 224)
│
└── split/                      # Train/validation/test splits
    ├── sed/                    # Splits for sediment task
    │   └── region{1,2,5,6,7}/
    │       ├── train.json
    │       ├── val.json
    │       └── test.json
    ├── pzone/                  # Splits for physiographic zone task
    │   └── region{1,2,5,6,7}/
    │       ├── train.json
    │       ├── val.json
    │       └── test.json
    ├── habitat/                  # Splits for habitat task
    │   └── region{3,4}/
    │       ├── train.json
    │       ├── val.json
    │       └── test.json
    ├── fault/                  # Splits for fault task
    │   └── region{3,4}/
    │       ├── train.json
    │       ├── val.json
    │       └── test.json
    └── fold/                  # Splits for fold task
        └── region{3,4}/
            ├── train.json
            ├── val.json
            └── test.json

Data Format Details

Input Files (input/):

6-channel NumPy arrays with shape (6, 224, 224)
Channels: [backscatter, bathymetry, slope, rugosity, longitude, latitude]
Naming: region{N}_{row}_{col}.npy

Annotation Files:

Single-channel NumPy arrays with shape (224, 224)
Integer labels corresponding to class indices
Naming matches corresponding input file

Split Files:

JSON files containing lists of sample identifiers
Organized by annotation type and region

🖼️ Samples

Region 1 - Sediment & Physiographic Zone

Region 3 - Habitat, Fault & Fold

💻 Visualization & Dataloader

Simple Visualization

To visualize samples and segmentation masks in the dataset, please refer to visualization.ipynb.

PyTorch Dataset Integration

To integrate with deep learning workflows, refer to the SeafloorDataset implementation in seafloor_dataset.py.

The following example demonstrates how to import and use it with a PyTorch DataLoader:

from torch.utils.data import DataLoader
from seafloor_dataset import SeafloorDataset

# Initialize Dataset
dataset = SeafloorDataset(
    data_path='./SeafloorAI',
    anno_path='./SeafloorAI/split',
    layer='sed',
    regions=['region1'],
    split='train',
    input_transform=None,  # Optional: transforms for input
    mask_transform=None    # Optional: transforms for mask
)

# Create DataLoader
loader = DataLoader(dataset, batch_size=4, shuffle=True, num_workers=4)

📋 ToDo

Release data and annotations for regions 1–7
Add sample visualizations and PyTorch dataloader support
Release fault and fold labels for region3 and region4
Release unlabeled data for region8 and region9
Release SeafloorGenAI Dataset

📜 Citation

If you find SeafloorAI useful, please cite the following paper:

@inproceedings{nguyen2024seafloorai,
  title={SeafloorAI: A Large-scale Vision-Language Dataset for Seafloor Geological Survey}, 
  author={Kien X. Nguyen and Fengchun Qiao and Arthur Trembanis and Xi Peng},
  booktitle={Proceedings of the Annual Conference on Neural Information Processing Systems (NeurIPS) Datasets and Benchmarks Track},
  year={2024}
}

📧 Contact & Acknowledgments

We sincerely thank the Department of Defense (DoD) for supporting this research.

We also acknowledge the U.S. Geological Survey (USGS) and the National Oceanic and Atmospheric Administration (NOAA) for providing the raw survey data.

Downloads last month: 16

Total file size:

178 GB