2022-SouthernHydrateRidge-Megafauna

Model Card: YOLOv8 Benthic Megafauna Object Detection Model for Southern Hydrate Ridge, 2022

Overview

This YOLOv8 object detection model is designed to automate detection of megabenthic morphotypes at the Southern Hydrate Ridge methane seep. The model was trained on seafloor imagery captured by the Canadian Scientific Submersible Facility’s ROV ROPOS during the VISIONS’22 University of Washington-NSF-OOI cruise. It is designed to capture common benthic megafauna morphotypes.

Dataset Details

• Number of Images: 577
• Missing Annotations: 0
• Null Examples: 0
• Total Number of Annotations: 6640 (Average of 11.5 annotations per image)
• Number of Classes: 16
• Image Resolution: 3500x2600 (wide orientation)

Class Distribution

Note: Detailed annotation counts are provided for the 10 most common classes.

Model Architecture

The model builds on the YOLO framework.

Training and Validation Splits

The dataset has been partitioned into training, validation, and test sets. Detailed split ratios and performance metrics (precision, recall, mAP, etc.) can be incorporated here as they become available.

Data Collection and Annotation Process

• Data Source: Imagery captured by the Canadian Scientific Submersible Facility’s ROV ROPOS during the VISIONS’22 University of Washington-NSF-OOI cruise.
• Annotation: Annotations were generated using an early iteration of the machine learning assisted manual annotation method described in Bigham & Carter (2026).

Intended Use

This model is intended for:

• Research: Accelerating research on deep marine ecology, particularly methane seeps.
• Environmental Assessment: Semi-automating deep marine monitoring and management efforts.

Limitations

• Class Imbalance: Class imbalance may affect performance on underrepresented classes.
• Specificity: This model was traiined on data exclusively from the Southern Hydrate Ridge methane seep. Transferrability is limited and will require additional model tuning.
• Color Correction: This model was trained on data that was NOT color corrected to remove blue tint from underwater imagery, and does not perform well on corrected images.

Ethical Considerations

ADD

Data Accessibility

ADD

Attribution

Data was collected during the VISIONS’22 University of Washington-NSF-OOI cruises with the Canadian Scientific Submersible Facility’s ROV ROPOS. Annotations generated using an early iteration of methods described in Bigham, K. T., & Carter, A. (2026). A machine learning assisted method for rapidly annotating benthic megafauna in large volumes of marine imagery. Limnology and Oceanography: Methods, e70057. https://doi.org/10.1002/lom3.70057