Technical articles on vision algorithms, applied mathematics, optics, and machine vision engineering.
VisionLab's Plugin SDK lets you add sub-pixel geometric measurement or AI defect detection to any C/C++ application — without taking on OpenCV or PyTorch as host dependencies.
A deep dive into how RANSAC, sector-centroid sampling, and Devernay sub-pixel edge detection combine to achieve circle-centre accuracy below 0.1 pixels.
PatchCore flips the defect-detection problem on its head — instead of learning what defects look like, it memorises what normal looks like. Here's the math behind it.
VisionLab fits rectangles by finding four edges sequentially with RANSAC, enforcing parallel and perpendicular constraints at each step — more robust than fitting all edges simultaneously.
How VisionLab fits ellipses robustly using algebraic constrained least-squares (Fitzgibbon) combined with a 120° sector-triplet sampling strategy for outlier rejection.
How VisionLab's gradient-orientation based template matching locates parts at any angle and scale — and why it outperforms classic NCC on industrial images.
A practical guide to building YOLO training datasets with VisionLab's AnnotateTab — including directory layout, label format, 4K tiling, and using a trained model to speed up annotation.
A practical guide to YOLOv8 training parameters in VisionLab — VRAM selection table, small-dataset tips, and a deep-dive into the close_mosaic loss explosion bug and its fix.