In-house object detector for finding queen bees among worker bees, drones, pollen bees, and frame/background content.
Repository: https://github.com/Gratheon/models-queen-bee-detector
It supports two deployment paths:
- browser inference for Live Queen Finder via ONNX +
onnxruntime-web - HTTP inference service for server-side experiments and integrations
Baseline training setup:
- Model:
yolov8n.pt - Image size:
512 - Epochs:
60 - Dataset: merged queen datasets with queen labels normalized to class
queenand non-queen images kept as negative/background samples
Test metrics (weights/best.pt):
- Precision:
0.9727 - Recall:
0.8590 - mAP50:
0.9187 - mAP50-95:
0.6114
Precision is high, but recall still leaves room for missed queens, so detections should be confirmed visually in field use.
