About the model
A YOLO26n detector exported to ONNX. It was trained as part of a study comparing
synthetic images rendered in Blender with real photos, and evaluated on real photos it never saw during training.
Project report and results Synthetic dataset
Known limitations
Works best with one or a few tools in clear view. Heavy clutter, unusual angles,
very small tools and tools that are partly hidden reduce accuracy.
Licensed AGPL-3.0, because the model was trained with Ultralytics YOLO.