Research map / Image generation / Autoregressive and latent generation
CAD-Refiner: A Unified Framework for CAD Generation and Iterative Editing
Meng Yuan, Dawei Lin, Hongxia Xie, Tieru Wu, Rui Ma
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace fixed masking ratio 0.5 with dynamic sine schedule in Sequence Injection Strategy (ablation) | CAD-Refiner · MMCAD generation (multimodal input) | Valid Ratio (%): 18.71 -> 52.91 | Improved |
| replace fixed masking ratio 0.9 with dynamic schedule (ablation) | CAD-Refiner · MMCAD generation (multimodal input) | Acc_c (%): 40.98 -> 84.66 | Improved |
| add chain-of-thought prompting to CAD Insighter (ablation) | CAD-Refiner · MMCAD generation (multimodal input) | Acc_p (%): 53.93 -> 59.48 | Improved |
| add CAD Insighter to text-to-CAD (ablation) | CAD-Refiner · MMCAD generation (text input) | Acc_p (%): 51.66 -> 55.52 | Improved |
| add CoT to CAD Insighter on text-to-CAD (ablation) | CAD-Refiner · MMCAD generation (text input) | Acc_p (%): 55.52 -> 57.06 | Improved |
| add CAD Insighter to text-to-CAD (ablation) | CAD-Refiner · MMCAD generation (text input) | inference time (s): 0.45 -> 1.09 | Hurt |
| add CAD Checker adaptive loss weighting (ablation) | CAD-Refiner · MMCAD generation (multimodal input) | Acc_p (%): 58.18 -> 59.48 | Improved |
| add CAD Checker adaptive loss weighting on JSD (ablation) | CAD-Refiner · MMCAD generation (multimodal input) | JSD (x10^2): 17.53 -> 17.51 | No clear effect |
| combine text and graph conditions in the decoder (ablation) | CAD-Refiner · MMCAD generation (multimodal input) | Acc_p (%): 57.78 -> 59.48 (graph only to both) | Improved |
| use multimodal prompts instead of single-modal prompts | CAD-Refiner · MMCAD generation | Acc_c (%): 83.78 -> 84.66 | Improved |
| use CAD-Refiner instead of LLaMA3.1-8B for editing | MMCAD editing · MMCAD editing | Acc_p (%): 64.45 -> 82.55 | Improved |
| use CAD-Refiner instead of LLM baselines for editing validity | MMCAD editing · MMCAD editing | Valid Ratio (%): 78.00 -> 70.69 (Qwen2.5-7B) | Hurt |
Closest papers by meaning
- CAD-Editor: A Locate-then-Infill Framework with Automated Training Data Synthesis for Text-Based CAD Editing (ICML 2025 · similarity 0.88)
- CADMorph: Geometry‑Driven Parametric CAD Editing via a Plan–Generate–Verify Loop (NeurIPS 2025 · similarity 0.88)
- cadrille: Multi-modal CAD Reconstruction with Reinforcement Learning (ICLR 2026 · similarity 0.87)
- FlexCAD: Unified and Versatile Controllable CAD Generation with Fine-tuned Large Language Models (ICLR 2025 · similarity 0.87)
- StepCAD: Mesh-to-CAD Code Generation via LLM Policy and Geometry-Guided Search (NeurIPS 2026 · similarity 0.86)
- CADCrafter: Generating Computer-Aided Design Models from Unconstrained Images (CVPR 2025 · similarity 0.86)
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.
