AI

StepCAD: Mesh-to-CAD Code Generation via LLM Policy and Geometry-Guided Search

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 08/10/2026, 08:18:24 UTC.

Briefings

StepCAD is a generative optimization approach that combines a state-conditioned CAD policy with geometry-guided search to recover executable CAD programs from 3D meshes. The method predicts construction actions conditioned on target and intermediate geometry, then refines the program via local edits guided by an IoU tree search. Experiments show StepCAD achieves state-of-the-art geometric reconstruction accuracy with up to 87.2% relative IoU improvement over the strongest baseline on complex shapes, while maintaining consistently high validity.

Key points

• StepCAD introduces a generative optimization approach combining a state-conditioned CAD policy with geometry-guided search to recover executable CAD programs from 3D meshes.
• The method achieves state-of-the-art geometric reconstruction accuracy with up to 87.2% relative IoU improvement over the strongest baseline on complex shapes.

Context

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.

Limitations

The evidence is limited to a single article published on 2026-09-30. The study uses a single large-scale dataset of 1.5M executable CAD programs and reports results from multiple benchmarks without specifying which specific datasets were used. The reported IoU improvement is relative to a single strongest baseline, and the study does not provide details on the specific CAD reconstruction benchmarks used or the computational resources required for the experiments.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 06/11/2026, 12:22:36 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeia72dsgqlblcdpnbcwuqiadgrvy5ggbllnz3wdwk77a3ecm24u6ri