About VisionLab
VisionLab is an industrial machine vision toolkit for geometric measurement and AI-based defect detection. It ships as a standalone Qt application and as a Plugin SDK that integrates into any C/C++ host in minutes.
Every algorithm is built around the mathematics — sub-pixel edge localisation, RANSAC robust estimation, and Vision Transformer feature spaces — so accuracy and reliability hold up in real production environments.
System Architecture
Two deployment modes sharing one algorithm engine
Standalone Mode
Qt UI
Inspect · Measure · Train
AlgoLib
C++ · OpenCV · LibTorch
Result
CSV · JSON · Overlay
Plugin Mode
Your Application (host)
shared mem
< 2 ms
VisionLab.exe (plugin process)
The plugin runs as a separate process — your host has no OpenCV, PyTorch, or Qt dependency at all.
Algorithm Modules
Circle Fitting
RANSAC + Devernay sub-pixel · < 0.1 px accuracy
Line Fitting
Rotated ROI + sector centroids · < 0.2 px accuracy
Ellipse Fitting
Eccentricity constraint · partial-occlusion robust
Rectangle Fitting
Hough/RANSAC · right-angle orthogonality constraint
Template Matching
Gradient-orientation · rotation invariant
PatchCore Defect Detection
DINOv2 + coreset · 5 normal images to train