CADFit: Precise Mesh-to-CAD Program Generation with Hybrid Optimization

ICML 2026

1Massachusetts Institute of Technology    2Siemens Digital Industries Software

CADFit reconstructs editable CAD construction sequences from watertight meshes — extrusions, revolutions, fillets, chamfers, composed with Boolean union and cut.

Abstract

Despite recent progress, recovering parametric CAD construction sequences from geometric input — meshes or point clouds — remains a key challenge for design and manufacturing. Existing CAD reconstruction and generation methods are largely restricted either to difficult-to-edit formats (meshes, B-reps) or to simple sketch-and-extrude pipelines on low-complexity datasets.

We introduce CADFit, a hybrid optimization-based framework that recovers complex, editable CAD construction sequences from meshes by incrementally fitting and validating parametric operations using geometric feedback. Our approach formulates reconstruction as an IoU-driven optimization over structured, executable CAD programs, and supports a rich operator set — extrusions, revolutions, fillets, and chamfers — composed with Boolean union and cut.

Experiments on multiple CAD benchmarks (DeepCAD, Fusion360 Gallery, ABC) show that CADFit outperforms state-of-the-art mesh-to-CAD methods in volumetric IoU and Chamfer Distance, while substantially reducing the Invalid Ratio of reconstructed CAD programs — particularly for complex designs. We further present a multimodal pipeline that enables end-to-end reconstruction of CAD construction sequences from images by combining image-based geometry reconstruction with CADFit.

Video

Pipeline

CADFit pipeline

Given a watertight input mesh, CADFit extracts candidate sketch profiles from detected planar face clusters and axis-aligned slicing planes. For each profile it generates a small set of geometrically consistent extrusion and revolve candidates via one-sided Chamfer parameter sweeps. Candidates are then assembled into a compact base program through IoU-guided backward pruning, executed and validated by the CAD kernel at every step. A bounded number of residual reconstruction iterations stitch in finer features via Boolean union and cut, and a final pass recovers fillets and chamfers.

Mesh-to-CAD Results

Qualitative mesh-to-CAD results

Qualitative reconstructions on DeepCAD, Fusion360, and ABC (easy / medium / hard). CADFit produces a valid, editable CAD program on every input where baselines either output Invalid or visibly diverge from the target geometry.

Image-to-CAD

Image-to-CAD

CADFit is agnostic to where the mesh comes from. Pairing it with a pretrained image-to-3D model — we use Hunyuan3D — and standard post-processing (watertight enforcement, Taubin smoothing, mesh decimation) gives a fully end-to-end image-to-CAD pipeline, with no retraining of CADFit.

Example CadQuery Output

CadQuery code

Every CADFit run emits an executable CadQuery program that reproduces the solid using sketches, extrusions, revolutions, Boolean composition, and finishing features.

BibTeX

@article{nehme2026cadfit,
  title={CADFit: Precise Mesh-to-CAD Program Generation with Hybrid Optimization},
  author={Nehme, Ghadi and Whalen, Eamon and Ahmed, Faez},
  journal={arXiv preprint arXiv:2605.01171},
  year={2026}
}