Research map / Image generation / Text-to-image diffusion
Intent2CAD: How Semantic-Parametric Supervision Shapes Text-to-CAD Generation
Jinbo Yang, Shikai Jing, Mingyue Yuan
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces intent-oriented text-to-CAD generation, where design intent is encoded as recoverable code-level signal so generated CAD programs expose reusable, modifiable structure.
Problem
Text-to-CAD generation typically produces executable geometry, but practical CAD workflows also need generated programs that expose reusable structure supporting later modification.
Approach
Frames text-to-CAD as intent-oriented generation in which design intent is expressed as recoverable code-level signal, using semantic-parametric supervision to shape generated CAD programs.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add semantic lifting to the supervision target (primitive-only to semantic) (ablation) | SFT text-to-CadQuery model · IntentCAD-100K test | Invalid Ratio (%): 2.39 -> 0.22 | Improved |
| add semantic lifting to the supervision target (ablation) | SFT text-to-CadQuery model · IntentCAD-100K test | Mean Chamfer Distance (x10^3): 1.44 -> 0.46 | Improved |
| add parametric lifting on top of semantic lifting (ablation) | SFT text-to-CadQuery model · IntentCAD-100K test | IR, AR, Median CD | No clear effect |
| add parametric lifting on top of semantic lifting (ablation) | SFT text-to-CadQuery model · IntentCAD-100K test | Mean Chamfer Distance (x10^3): 0.46 -> 0.57 | No clear effect |
| add semantic lifting to the supervision target (ablation) | zero-shot editing · IntentCAD-Edit-1K | Edit Success Rate (%): 41.80 -> 88.90 | Improved |
| add parametric lifting on top of semantic lifting (ablation) | zero-shot editing · IntentCAD-Edit-1K | CE@ES (%): 91.00 -> 95.82 | Improved |
| add parametric lifting on top of semantic lifting (ablation) | zero-shot editing · IntentCAD-Edit-1K | FE@ES (%): 95.17 -> 92.51 | Mixed |
| add semantic lifting to the supervision target (ablation) | visually grounded editing · Visually grounded editing (200 cases) | Edit Success (%): 36.50 -> 66.00 | Improved |
| add parametric lifting on top of semantic lifting (ablation) | visually grounded editing · Visually grounded editing (200 cases) | Edit Success (%): 66.00 -> 78.50 | Improved |
| add parametric lifting for feature edits in visually grounded editing (ablation) | visually grounded editing · Visually grounded editing (200 cases) | FE@ES (%): 86.25 -> 78.75 | Hurt |
| fine-tune open backbone on IntentCAD SFT | Qwen3.5-9B · IntentCAD-100K | Invalid Ratio (%): 99.86 -> 0.20 | Improved |
Closest papers by meaning
- CAD-Editor: A Locate-then-Infill Framework with Automated Training Data Synthesis for Text-Based CAD Editing (ICML 2025 · similarity 0.89)
- Clarify Before You Draw: Proactive Agents for Robust Text-to-CAD Generation (ICML 2026 · similarity 0.89)
- Text2CAD: Generating Sequential CAD Designs from Beginner-to-Expert Level Text Prompts (NeurIPS 2024 · similarity 0.88)
- FlexCAD: Unified and Versatile Controllable CAD Generation with Fine-tuned Large Language Models (ICLR 2025 · similarity 0.88)
- CAD-Coder: Text-to-CAD Generation with Chain-of-Thought and Geometric Reward (NeurIPS 2025 · similarity 0.87)
- CAD-Llama: Leveraging Large Language Models for Computer-Aided Design Parametric 3D Model Generation (CVPR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
