microsoft/YOLaT-VectorGraphicsRecognition

Operates directly on SVG primitives and Bézier curves as graph-structured data, eliminating rasterization overhead while preserving structural information. Implements hierarchical recognition across three levels (primitive, curve, point) with position-aware enhancement in YOLaT++, using graph neural networks to process vector geometry natively. Targets CAD and diagram datasets including floorplans and charts, with PyTorch training pipelines supporting configurable graph representations and data augmentation.

101 stars. No commits in the last 6 months.