Title: CADFit: Achieving Precise Mesh-to-CAD Program Generation via Hybrid Optimization

Abstract:

Recovering parametric CAD construction sequences from geometric inputs, such as meshes or point clouds, remains a significant hurdle in design and manufacturing. Current reconstruction and generation techniques are often limited to hard-to-edit formats like meshes or Breps, or they rely on editable but simplistic sketch-and-extrude pipelines and low-complexity datasets. To address these limitations, we present CADFit, a framework for CAD reconstruction based on hybrid optimization. This method recovers complex, editable CAD construction sequences from meshes by incrementally fitting and validating parametric operations through geometric feedback.

CADFit distinguishes itself by treating reconstruction as an Intersection-over-Union (IoU)-driven optimization problem over structured CAD programs. It supports a diverse range of operations, including extrusions, revolutions, fillets, and chamfers. Evaluations across several CAD benchmarks demonstrate that CADFit surpasses current state-of-the-art mesh-to-CAD methods in terms of volumetric Intersection-over-Union and Chamfer Distance. Furthermore, it significantly lowers the Invalid Ratio of reconstructed CAD programs, with notable improvements for complex designs.

Additionally, we introduce a multimodal pipeline that facilitates the end-to-end reconstruction of CAD construction sequences from images. This is achieved by integrating image-based geometry reconstruction with CADFit. By enabling the accurate reconstruction of higher-complexity CAD models, CADFit establishes a practical foundation for creating richer datasets and advancing future learning-based approaches to CAD reverse engineering. The source code is accessible at: https://github.com/ghadinehme/CADFit.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC