Model Card for Model ID

This model is designed for object detection in marine environments, specifically targeting benthic megafauna in ROV transects.

Model Details

Training Data

The model is trained on the OceanCV/EcotypeTransect dataset, which contains annotated images of marine fauna and geological features from various methane seep and hydrothermal vent sites along the West Coast.

Training Procedure

Preprocessing

Images were resized to 1024x1024 pixels

Data augmentation techniques included random flipping, rotation, and contrast adjustments

Training Hyperparameters

Training regime: Mixed precision (fp16) Epochs: 200 Batch size: Auto (-1) Optimizer: Auto Learning rate: 0.01 Momentum: 0.937 Weight decay: 0.0005

Model size: 450 MB

Evaluation

Testing Data, Factors & Metrics

Testing Data

The testing dataset consists of near methane seep imagery from a singular 38 ROV surveys conducted in the Northeast Pacific.

Metrics

mAP50: 78.0% mAP50-95: 56.7% Precision: 72.7% Recall: 75.1%

Developed by: Atticus Carter
Model type: YOLOV11
License: MIT
Finetuned from model: FathomNet/MBARI-315k-yolov8

OceanCV
/

OOI_Ecotype