NMFS-OSI/yolo11m-esa-coral-icra-detector

Yolo11m ESA Coral ICRA Detector

Model Details

This model was trained to identify and locate Isopora crateriformis (ICRA) using the ultralytics YOLO11 architecture in various underwater conditions.

• Model Architecture: YOLO11m
• Task: Object Detection (Detection)
• Classes: 1 - ICRA (Isopora crateriformis)
• Data Type: Underwater Optical Imagery

Model Weights

• The model's weights can be found here: yolo11-esa-icra-detector.pt

Try a Live Demo here: NMFS-OSI/ESA-Coral-ICRA-Detector-Demo

Models Intended Use

• Detection of ICRA on underwater optical data

Dataset

The training data was collected, parsed, sorted, augmented, and organized from various NOAA Missions:

• Google Cloud Link: Dataset
• METADATA

Dataset Composition:

• Multi-source Dataset: Trained on datasets that include images from various angles.
• TOAL Images: 470 images
• Training Images: 327
• Validation Images: 71
• Test Images: 72
• Train/Val/Test Split Ratio: 7:2:1

Data Augmentations Used

To improve generalization in underwater environments and to increase dataset size, the following augmentations were applied:

• Mosaic: Enabled (Closed last 40 epochs).
• HSV Color Jitter: Hue (0.015), Saturation (0.4), Value (0.3).
• Scaling: Random scaling (0.5).
• MixUp: Disabled to prevent unrealistic object overlap.

Model Performance

• Model Architecture (YOLO11m): Lightweight and optimized for real-time detection in underwater footage.

Metrics

Below are the key metrics from the model evaluation on the validation set:

Results (100 epochs)

• Best epoch: 84
• mAP50 (B) 0.80353
  • Mean average precision calculated at an intersection over union (IoU) threshold of 0.50.

• Precision (B) 0.8188
  • The accuracy of the detected objects, indicating how many detections were correct.

• Recall (B) 0.74149
  • The ability of the model to identify all instances of objects in the images.

• mAP50-95 (B) 0.64382
  • The average of the mean average precision calculated at varying IoU thresholds, ranging from 0.50 to 0.95.

Training Validation Results

Training and Validation Losses

Precision-Recall Curve

F1 Score Curve (F1 = 2 * (Precision * Recall) / (Precision + Recall))

Training Configuration

• Model Weights File: yolo11-esa-icra-detector.pt
• Number of Epochs: 100
• Learning Rate: Optimizer AdamW (Learning Rate: 0.001, Final LR: 0.00003)
• Batch Size: 16
• Image Size: 1024x1024

Deployment

How to Use the Model

To use the trained model, follow these steps:

1. Load the Model:

   from ultralytics import YOLO
   
   # Load the model
   model = YOLO("yolo11-esa-icra-detector.pt")
   results = model.predict("path/to/image_or_video.jpg", imgsz=1024, conf=0.5)
   

Limitations

It may not generalize well to other environments not found in training dataset(Am.Samoa) or non-marine scenarios. Additionally, environmental variations, occlusions, or poor lighting may affect performance.

Additional Notes:

Ethical Considerations:

• The detection results should be validated before using them for critical applications. The model’s performance in new environments might vary, and it may have biases if certain types of corals were underrepresented in the training datasets.

Disclaimer

This repository is a scientific product and is not official communication of the National Oceanic and Atmospheric Administration, or the United States Department of Commerce. All NOAA project content is provided on an ‘as is’ basis and the user assumes responsibility for its use. Any claims against the Department of Commerce or Department of Commerce bureaus stemming from the use of this project will be governed by all applicable Federal law. Any reference to specific commercial products, processes, or services by service mark, trademark, manufacturer, or otherwise, does not constitute or imply their endorsement, recommendation or favoring by the Department of Commerce. The Department of Commerce seal and logo, or the seal and logo of a DOC bureau, shall not be used in any manner to imply endorsement of any commercial product or activity by DOC or the United States Government.