跳到正文
HuggingFace Daily Papers(社區熱門論文)·· 5 天前AI 評分53

CADMAGEDDON 以本地推理實現高保真 CAD 重建

CADMAGEDDON: High-Fidelity CAD Reconstruction with Local Inference

AI 導讀

CADMAGEDDON 提出以本地推理將網格重建為可編輯 CAD 的方法:幾何工具識別並擬合網格實體,Qwen3.8-27B VLM 則根據渲染結果決定建模與修正步驟。

正文

CADMAGEDDON: mesh → editable CAD with one open-weight model, no frontier API needed 🛠️

Most mesh-to-CAD systems make a model write the whole program, including every coordinate and dimension. We split each feature in two:
• geometric instruments detect and name the entities on the target mesh (levels, fitted lines/arcs, faces, hole stacks, patterns) and fit their exact values;
• a mid-sized VLM (Qwen3.8-27B) looks at renders and decides only what to build next and what to fix, referring to geometry by name.

No reconstruction model is trained. The agent emits CadQuery or native FreeCAD programs, and the model fits on a single 48 GB GPU.

Results (vs. the strongest baseline, CADFather):
• MCB: IoU 0.862 → 0.933, GMS 0.705 → 0.818
• CADENA-Bench: GMS 0.800 → 0.909
• BenchCAD (64³ voxel IoU): 0.991, above the best self-reported frontier entry (0.963)

Code: https://github.com/kulibinai/cadmageddon

來源:HuggingFace Daily Papers(社區熱門論文) · huggingface.co