[
    {
        "id": "osp-21294",
        "type": "article-journal",
        "title": "StepCAD: Mesh-to-CAD Code Generation via LLM Policy and Geometry-Guided Search",
        "author": [
            {
                "family": "Nehme",
                "given": "Ghadi"
            },
            {
                "family": "Ahmed",
                "given": "Faez"
            }
        ],
        "URL": "https://omanscience.com/ar/articles/stepcad-mesh-to-cad-code-generation-via-llm-policy-and-geometry-guided-search",
        "language": "en",
        "issued": {
            "date-parts": [
                [
                    2026
                ]
            ]
        },
        "abstract": "Recovering executable CAD programs from 3D meshes is challenging due to the compositional nature of CAD construction and the interaction between discrete modeling choices and continuous parameters. Many learning-based methods predict complete programs in a single pass and rely predominantly on sketch-extrude representations, limiting operation diversity and opportunities to correct geometric errors during reconstruction. We introduce StepCAD, a generative optimization approach that combines a state-conditioned CAD policy with geometry-guided search. Given an input mesh, the policy predicts construction actions conditioned on both target and intermediate geometry, and an IoU-guided tree search refines the resulting program through local edits. We also introduce ARCADE-1.5M, a large-scale dataset of 1.5M executable CAD programs spanning diverse operations, sequences with a maximum length of 150+ counted operations, and 12.5M intermediate state-action transitions. Experiments across multiple CAD reconstruction benchmarks show that StepCAD achieves state-of-the-art geometric reconstruction accuracy with consistently high validity, yielding up to 87.2% relative IoU improvement over the strongest evaluated baseline, with particularly large gains on complex shapes. Project page: https://ghadinehme.com/stepcad.github.io/"
    }
]