ConCAD: Constraint-Aware Image-to-CAD Generation with Dual-Granularity Rewards

2026-09-28

AI Summary

This study addresses the limitations of existing image-to-CAD methods in distinguishing structurally distinct shapes with similar volumes and recovering underlying design intent. To this end, we propose a constraint-aware image-to-CAD framework that generates executable parametric programs consistent in both geometry and design intent. Methodologically, we introduce a dual-granularity reward mechanism integrating code-level constraints with execution-level geometric complementarity, define a B-rep geometric constraint satisfaction rate to quantitatively evaluate intent recovery, and employ Group Relative Policy Optimization (GRPO) combined with boundary representation analysis for multi-granularity reinforcement learning. Experiments on the DeepCAD and Zero2CAD datasets demonstrate that our approach significantly outperforms baselines in IoU and Chamfer distance, validating its effectiveness in achieving high-fidelity geometric reconstruction and accurate design intent recovery.

Abstract

Image-to-CAD generation seeks executable parametric programs that recover both the geometry and design intent of a reference object. Existing systems are commonly evaluated by validity and shape overlap, although two solids with similar volume can encode different CAD relations. We introduce ConCAD, a constraint-aware image-to-CAD framework optimized via Group Relative Policy Optimization (GRPO) with rewards at two complementary granularities: a code-level constraint reward and an execution-level geometric reward. This complementary design disambiguates structurally distinct yet volumetrically similar shapes while ensuring valid 3D geometry. To verify that these rewards recover geometry and design intent, we introduce a B-rep geometric constraint satisfaction rate (G-CSR), which analytically extracts and evaluates geometric constraints from boundary representations. Experiments on the DeepCAD and Zero2CAD demonstrate that ConCAD achieves the best IoU and Chamfer Distance over competitive baselines, while also outperforming them on G-CSR, validating its superior recovery of both geometric fidelity and parametric design intent.

Problem

• Image-to-CAD generation
• geometric constraints
• design intent
• parametric programs
• B-rep

Innovation

• Image-to-CAD
• Group Relative Policy Optimization (GRPO)
• Dual-Granularity Rewards
• Geometric Constraint Satisfaction Rate (G-CSR)
• Constraint-Aware Generation