Reproduce “Intent2CAD: How Semantic-Parametric Supervision Shapes Text-to-CAD Generation” 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

Jinbo Yang, Mingyue Yuan, Shikai Jing

Intent2CAD tests whether training text-to-CAD models on code with recoverable features and simple parameter relations improves generation and editing, without changing architecture or decoding. With Qwen3.5-9B, semantic supervision lowers invalid outputs from 2.39% to 0.22% and reduces CD from 1.44 to 0.46; semantic-parametric supervision improves relation-edit success, but not feature-edit success, under explicitly targeted edits.

Summary by alphaXiv.

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