BrepGPT: Autoregressive B-rep Generation with Voronoi Half-Patch

2025-11-27

AI Summary

To address the problem of error accumulation and low efficiency arising from strong geometry-topology coupling in B-rep modeling, this paper proposes the first single-stage autoregressive framework. The core innovation is the Voronoi Half-Patch representation, which unifies faces, edges, vertices, and adjacency relations into local geometric-topological units based on half-edges. A dual VQ-VAE enables compact vertex-level sequence encoding, while a decoder-only Transformer performs end-to-end sequence generation. The method supports unconditional generation—achieving state-of-the-art performance—as well as multimodal conditional generation from categories, point clouds, text, or images. Furthermore, it extends to B-rep auto-completion and interpolation, significantly enhancing generalization and practical applicability.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Diffusion Models for Vision
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems

Abstract

Boundary representation (B-rep) is the de facto standard for CAD model representation in modern industrial design. The intricate coupling between geometric and topological elements in B-rep structures has forced existing generative methods to rely on cascaded multi-stage networks, resulting in error accumulation and computational inefficiency. We present BrepGPT, a single-stage autoregressive framework for B-rep generation. Our key innovation lies in the Voronoi Half-Patch (VHP) representation, which decomposes B-reps into unified local units by assigning geometry to nearest half-edges and sampling their next pointers. Unlike hierarchical representations that require multiple distinct encodings for different structural levels, our VHP representation facilitates unifying geometric attributes and topological relations in a single, coherent format. We further leverage dual VQ-VAEs to encode both vertex topology and Voronoi Half-Patches into vertex-based tokens, achieving a more compact sequential encoding. A decoder-only Transformer is then trained to autoregressively predict these tokens, which are subsequently mapped to vertex-based features and decoded into complete B-rep models. Experiments demonstrate that BrepGPT achieves state-of-the-art performance in unconditional B-rep generation. The framework also exhibits versatility in various applications, including conditional generation from category labels, point clouds, text descriptions, and images, as well as B-rep autocompletion and interpolation.

Problem

• Generates CAD boundary representations using single-stage autoregressive framework
• Unifies geometric and topological elements in coherent Voronoi Half-Patch format
• Enables conditional generation from multiple inputs like text and images

Innovation

• Voronoi Half-Patch representation for unified B-rep decomposition
• Dual VQ-VAEs encode vertex topology and patches into tokens
• Decoder-only Transformer autoregressively predicts tokens for generation