ki-ideenwerkstatt-23/robbenblick

Model Card: YOLOv8m Object Detector

Model Overview

• Model type: YOLOv8m (Ultralytics)
• Task: Object detection
• Framework: PyTorch
• Weights file: iter_run_model_yolov8m.pt_freeze_10_epochs_150.pt
• Version: v1.0.0
• Trained by: KI-Ideenwerkstatt
• Date: 2025-11

Intended Use

This model is intended for:

• Detecting seals in aerial images

Training Data

• Dataset: Aerial_Robben_23_25
• Number of images: ~120
• Annotation format: CVAT
• Source: proprietary

Training Details

• Base model: YOLOv8m pretrained on COCO
• Image size: 9504x6336
• Epochs: 150
• Optimizer: <SGD/AdamW>

Data Augmentation

Training used the following augmentations optimized for aerial images:

• Random rotation up to ±90°
• Translation up to 10% of image size
• Scaling (zoom) up to ±10%
• Horizontal flip (50% probability)
• Vertical flip (50% probability)
• HSV color jitter:
  • Hue ±0.015
  • Saturation ±0.7
  • Value (brightness) ±0.4

Evaluation

• Validation set: Validation split
• Metrics:
  • mAP@0.5: 0.834
  • mAP@0.5:0.95: 0.3175

Limitations

• Performance limited due to small dataset size

Ethical Considerations

• No personal data used
• Not intended for surveillance or biometric identification

License

• Model weights license: agpl-3
• Dataset license: unknown

How to Use

from ultralytics import YOLO
model = YOLO("best.pt")
results = model("image.jpg")