CAD-Judge: Toward Efficient Morphological Grading and Verification for Text-to-CAD Generation

2025-08-05

AI Summary

To address inefficiencies in morphological scoring, high verification costs of vision-language models (VLMs), and reward-hacking risks in text-to-CAD generation, this paper proposes a Compiler-as-a-Judge/Reviewer dual-module framework. It replaces VLM-based validation with a lightweight CAD compiler for efficient syntactic correctness and geometric plausibility assessment; designs a prospect-theory-inspired, deception-resistant reward mechanism; and incorporates a proxy-based generation strategy to enhance robustness. The method achieves state-of-the-art performance across multiple complex CAD benchmarks, accelerates inference and verification by 3.2×, significantly reduces deployment overhead, and intrinsically mitigates reward manipulation. To our knowledge, it is the first approach to establish a verifiable, efficient, and trustworthy closed-loop evaluation framework for text-to-CAD generation.

Technology Category

• Computer Vision: Language and Vision
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Cognitive Modeling & Cognitive Systems: Computational Creativity

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Computer-Aided Design (CAD) models are widely used across industrial design, simulation, and manufacturing processes. Text-to-CAD systems aim to generate editable, general-purpose CAD models from textual descriptions, significantly reducing the complexity and entry barrier associated with traditional CAD workflows. However, rendering CAD models can be slow, and deploying VLMs to review CAD models can be expensive and may introduce reward hacking that degrades the systems. To address these challenges, we propose CAD-Judge, a novel, verifiable reward system for efficient and effective CAD preference grading and grammatical validation. We adopt the Compiler-as-a-Judge Module (CJM) as a fast, direct reward signal, optimizing model alignment by maximizing generative utility through prospect theory. To further improve the robustness of Text-to-CAD in the testing phase, we introduce a simple yet effective agentic CAD generation approach and adopt the Compiler-as-a-Review Module (CRM), which efficiently verifies the generated CAD models, enabling the system to refine them accordingly. Extensive experiments on challenging CAD datasets demonstrate that our method achieves state-of-the-art performance while maintaining superior efficiency.

Problem

• Slow rendering and expensive verification of CAD models
• Risk of reward hacking in CAD model evaluation
• Need for efficient CAD model grading and validation

Innovation

• Compiler-as-a-Judge Module for fast reward
• Agentic CAD generation for robustness
• Compiler-as-a-Review Module for verification