WoodDefect Detection Dataset for Automated Visual Inspection in wood Processing
Authors/Creators
Description
Dataset Description: WoodDefect Detection
Domain: Automated visual inspection in hardwood processing
Task: Multi-class defect detection with bounding-box labels
Origin and acquisition
The images were collected by GoldenY through a high-resolution industrial camera under controlled diffuse LED illumination in a European sawmill.
All data were taken from the planed surface of kiln-dried European beech (Fagus sylvatica) boards with thickness 26–50 mm. No magnification filters or digital zoom were applied, guaranteeing a native pixel resolution of ≈ 0.08 mm px⁻¹ on the wood surface.
Volume and splits
Total annotated images: 3 785.
Current partitioning:
– Training: 3105 images (82 %)
– Validation: 680 images (18 %)
The split is stratified so that the global defect-class distribution is preserved within each subset.