skywalker-object-detection-model

Model Card: Skywalker YOLOv8m

Model Details

• Architecture: YOLOv8m (Medium)
• Framework: PyTorch
• Task: Object Detection (Bounding Box)
• Parameters: ~25.9M
• FLOPs: ~78.9 GFLOPs

Intended Use

• Primary Use Case: Real-time object detection and localization within the specified custom domain (e.g., lightsabers).
• Out of Scope: Tracking (requires deep-sort/ByteTrack integration), instance segmentation, and pose estimation.

Training Data

• Dataset: Custom Domain Dataset (44,000 instances).
• Format: YOLO format (normalized [x_center, y_center, width, height]).
• Split: Configured via data.yaml (Train/Val sets).

Performance Summary (Epoch 1 Dry-Run Baseline)

• Precision: 0.00333
• Recall: 1.000
• mAP@50: 0.00513
• Latency (ONNX, Batch=1): ~25.2ms

Limitations & Ethical Considerations

• Environmental Constraints: Model performance may degrade under severe low-light conditions, heavy occlusion, or significant domain shift (e.g., thermal imaging).
• Bias: The model is inherently biased toward the spatial distribution and lighting conditions present in the training dataset.

Citation

If utilizing this model or repository for academic purposes, please cite:

@software{skywalker_yolov8,
  author = {Rasul Gurbanli},
  title = {Skywalker Object Detection: Production-Grade YOLOv8},
  year = {2026},
  url = {https://github.com/rasul-gurbanli/skywalker-object-detection}
}