megalodon-2024-yolov11

FathomNet Megalodon Detector

Model Details

• Trained by researchers at the Monterey Bay Aquarium Research Institute (MBARI).
• Ultralytics YOLO11x
• Object detection model
• Fine-tuned to detect 1 class, called 'object', using all FathomNet localizations
• Updated from the yolo11x_megalodon_e70_p10 training run

Intended Use

• Post-process video and images collected by marine researchers
• Can be used to build a localized set of training images, when neither training data nor a model exists for the imagery being analyzed

Factors

• Distribution shifts related to sampling platform, camera parameters, illumination, and deployment environment are expected to impact model performance
• Evaluation was performed on an IID subset of available training data as well as out-of-distribution data

Training Details

• Training run: yolo11x_megalodon_e70_p10
• Max epochs: 70, early stopping patience: 10
• Training stopped at epoch 63
• Best checkpoint: epoch 53 (highest mAP50-95)
• Best mAP50-95: 0.578
• Best mAP50: 0.799
• Full hyperparameters: see args.yaml
• Per-epoch metrics: see results.csv

Training and Evaluation Data

• All publicly-available data on FathomNet
• Training/evaluation dataset: FathomNet/2024-megalodon-benchmark

Deployment

1. Clone this repository
2. In an environment with the ultralytics Python package installed, run:

yolo predict model=mbari-megalodon-yolo11x.pt