Resources

Guides

Substantive articles on vector data for AI.

• 01What is SVG training data?Text, paths, geometry, layout and metadata: what a vector file carries that a bitmap cannot.
• 02SVG vs raster for visual AI: what pixels throw awayA single poster compared as PNG and as SVG, field by field.
• 03Why every public SVG dataset is icons, and what that costs a modelSVG-Stack, MMSVG-2M, SVGX and SAgoge side by side with their licences.
• 04How to train a text-to-SVG modelData, normalisation, tokenisation and the evaluation metrics buyers actually run.
• 05Building a layout dataset from editable design filesFrom live text nodes to bounding boxes, roles and z-order.
• 06How to evaluate an SVG dataset before you license itValidity, render consistency, duplicate rate, text editability and provenance.
• 07Floor-plan datasets for architecture AIRPLAN, LIFULL, CubiCasa5K, ResPlan and what none of them include.
• 08Commercial licensing for AI training dataThe clauses that matter: permitted use, derivative models, reconstruction, indemnity.
• 09How SVGZO validates and deduplicates 3.3M vector filesresvg, structural hashes, perceptual hashes and the thresholds we publish.
• 10Documenting training data for the EU AI ActWhat the training-summary template asks and what a dataset vendor should hand over.