Reproduce “NURBGen: High-Fidelity Text-to-CAD Generation through LLM-Driven NURBS Modeling” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Muhammad Usama, Mohammad Sadil Khan, Didier Stricker, Muhammad Zeshan Afzal

NURBGen introduces a framework that leverages Large Language Models (LLMs) to generate high-fidelity 3D CAD models directly from natural language using Non-Uniform Rational B-Splines (NURBS). This method produces editable, industry-standard CAD models by translating text into structured JSON representations of NURBS parameters, achieving improved geometric fidelity and a low invalidity ratio compared to existing text-to-CAD baselines.

Key insights

  • Fine-tuning an LLM to directly output symbolic NURBS parameters enables the generation of industrially compatible, editable CAD models from natural language, bridging the gap between intuitive input and professional output.
  • A hybrid representation, combining untrimmed NURBS with analytic primitives, is crucial for robustly modeling trimmed surfaces and complex geometries, significantly enhancing geometric fidelity and reducing artifacts.
  • Leveraging a multi-view VLM and detailed metadata to automatically generate captions for the ABC dataset provides a scalable solution to the data scarcity problem for text-to-CAD, enabling training on diverse and complex geometries.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.