NURBGen
LLM-driven NURBS text-to-CAD scored on preference, invalidity, CD/HD/JSD/MMD versus Text2CAD, DeepCAD, and GPT-4o.
What it does and how it connects
NURBGen is the first framework to generate high-fidelity 3D CAD models directly from text using Non-Uniform Rational B-Splines (NURBS). Existing text-to-CAD systems either produce meshes or rely on scarce design-history data; NURBGen instead produces editable NURBS surfaces.
It fine-tunes a large language model to translate free-form texts into JSON representations containing NURBS surface parameters, namely control points, knot vectors, degrees and rational weights, which can be directly converted into B-Rep format using Python. A hybrid representation combines untrimmed NURBS with analytic primitives to handle trimmed surfaces and degenerate regions more robustly while reducing token complexity.
The paper also introduces partABC, a curated subset of the ABC dataset consisting of individual CAD components, annotated with detailed captions using an automated annotation pipeline. NURBGen reports strong performance on diverse prompts, surpassing prior methods in geometric fidelity and dimensional accuracy, as confirmed by expert evaluations.
Alternatives
Compare side by side- Text2CAD
A framework and dataset of around 170K models that generates sketch-and-extrude sequences from text, rather than generating NURBS surfaces from text.
- MUSE
Grades text-to-CAD assemblies in a three-stage funnel ending in a VLM judge, rather than generating NURBS models from text.
- Text2CAD-Bench
Tests text-to-parametric-CAD on 600 human-curated examples across four complexity levels, rather than generating NURBS surfaces from text.
- CADBench (Doris et al.)
A unified 18,000-sample benchmark turning mesh, image or render inputs into STEP or mesh, rather than NURBS text-to-CAD generation.
- DeepCAD (dataset)
A dataset of 178K sketch-and-extrude sequences parsed from Onshape, rather than a NURBS text-to-CAD framework.
Sources
Not verified
- License: no license is stated on the arXiv page read.
- A public repo: the abstract states code and dataset will be released publicly, but no repository URL appears on the abstract page read.
- The scoring metrics named in the one-liner (preference, invalidity, CD/HD/JSD/MMD versus Text2CAD, DeepCAD and GPT-4o): directory record; the arXiv abstract read does not state them.
- Pricing: none stated (research framework).
- Real user voices: no real linked post read for this page.
- Related Error Index entries: no matching entry verified.
Last verified November 6, 2026. Listings are informational. The evidence score measures public, verifiable evidence and is not an endorsement or a compatibility claim.