Img2CADSeq: Image-to-CAD Generation via Sequence-Based Diffusion

2026-05-13

AI Summary

Reconstructing high-quality, industry-standard B-rep CAD models from single-view images remains challenging due to topological complexity and difficulties in modeling CAD construction sequences. This work proposes a multi-stage generative framework that, for the first time, integrates sequential diffusion models with a hierarchical discrete latent space. It encodes CAD operations via a three-level hierarchical codebook and employs a contour-first importance ranking strategy to compress sequence length. The method leverages a coarse-to-fine point cloud intermediate representation and aligns 2D image features with 3D CAD sequences through contrastive learning, enabling a VQ-Diffusion model to directly output standard STEP files. Evaluated on the newly introduced CAD-220K and PrintCAD datasets, the approach significantly outperforms existing methods, producing models readily importable into commercial CAD software.

Abstract

Boundary Representation (BRep) is the standard format for Computer-Aided Design (CAD), yet reconstructing high-quality BReps from single-view images remains challenging due to the complexity of topological constraints and operation sequences. We present Img2CADSeq, a multi-stage pipeline that overcomes these limitations by encoding CAD sequences into a three-level hierarchical codebook. Guided by an importance prioritization, this strategy values profiles over details, compressing long sequences into a stable discrete latent space. To bridge the modality gap, we leverage a coarse-to-fine point cloud intermediate, aligning 2D visual features with 3D CAD sequences via contrastive learning to condition a VQ-Diffusion model. Supported by newly introduced CAD-220K and PrintCAD datasets, our approach ensures robust industrial domain adaptation. Extensive experiments demonstrate that Img2CADSeq significantly outperforms state-of-the-art methods, producing standard STEP files that can be directly used in commercial CAD software.

Problem

• Image-to-CAD
• Boundary Representation
• BRep reconstruction
• single-view image
• CAD generation

Innovation

• sequence-based diffusion
• hierarchical codebook
• BRep reconstruction
• contrastive learning
• VQ-Diffusion