CADReview: Automatically Reviewing CAD Programs with Error Detection and Correction

TL;DR

The paper addresses automatic CAD program review by defining the CADReview task to detect and correct discrepancies between CAD programs and reference images. It introduces ReCAD, a multimodal LLM-based framework with a Feedback Generator that aligns geometric components to code blocks and a Code Editor that performs spatial geometric operations, further enhanced by RL-based refinement. A new CADReview dataset with over 21k program-image pairs supports robust evaluation and comprises both human-made and machine-made CAD programs with annotated feedback. Experimental results show that ReCAD significantly outperforms existing MLLMs in both feedback quality and the fidelity of reconstructed 3D objects, highlighting its potential to streamline AI-assisted industrial design workflows. The work also details dataset construction, localization, feedback generation, and iterative refinement to enable robust CAD-aware program repair.

Abstract

Computer-aided design (CAD) is crucial in prototyping 3D objects through geometric instructions (i.e., CAD programs). In practical design workflows, designers often engage in time-consuming reviews and refinements of these prototypes by comparing them with reference images. To bridge this gap, we introduce the CAD review task to automatically detect and correct potential errors, ensuring consistency between the constructed 3D objects and reference images. However, recent advanced multimodal large language models (MLLMs) struggle to recognize multiple geometric components and perform spatial geometric operations within the CAD program, leading to inaccurate reviews. In this paper, we propose the CAD program repairer (ReCAD) framework to effectively detect program errors and provide helpful feedback on error correction. Additionally, we create a dataset, CADReview, consisting of over 20K program-image pairs, with diverse errors for the CAD review task. Extensive experiments demonstrate that our ReCAD significantly outperforms existing MLLMs, which shows great potential in design applications.

Figures (10)

• Figure 1: Overview of different ways (i.e., designer, GPT-4o and our ReCAD) to CAD review. The green and red denote correct and incorrect prediction, respectively.
• Figure 2: Construction workflow of our CADReview dataset, including program-drawing pairs collection, error creation and feedback annotation.
• Figure 3: Examples of 8 types of CAD program error from our CADReview dataset. Top row: The reference images from one sampled viewpoint. Second row: Error creation on CAD programs, showing the edited code segments. Bottom row: The rendered 3D objects by erroneous programs.
• Figure 4: Overview of our ReCAD. We design geometric component recognition (GCR) and spatial geometric operation (SGO) mechanisms to initialize feedback generator $\phi_1$ and code editor $\phi_2$, respectively.
• Figure 5: Overview of CAD localization. (a) Semantic matching. (b) Coordinate matching and localization.