DTGBrepGen: A Novel B-rep Generative Model through Decoupling Topology and Geometry

TL;DR

DTGBrepGen addresses automatic B-rep CAD generation by decoupling topology and geometry. It first learns valid topologies through two Transformer-based stages (edge-face and edge-vertex adjacencies) and then uses Transformer-based diffusion on B-spline control points to generate geometry, enabling accurate, watertight B-rep models. The approach yields superior topology validity and geometric quality across multiple datasets, and supports unconditional and conditional generation (including class- and point-cloud-conditioned outputs). The work demonstrates that explicit topology constraints and continuous B-spline representations substantially improve realism, diversity, and manufacturability of generated CAD models, with code available for reproducibility.

Abstract

Boundary representation (B-rep) of geometric models is a fundamental format in Computer-Aided Design (CAD). However, automatically generating valid and high-quality B-rep models remains challenging due to the complex interdependence between the topology and geometry of the models. Existing methods tend to prioritize geometric representation while giving insufficient attention to topological constraints, making it difficult to maintain structural validity and geometric accuracy. In this paper, we propose DTGBrepGen, a novel topology-geometry decoupled framework for B-rep generation that explicitly addresses both aspects. Our approach first generates valid topological structures through a two-stage process that independently models edge-face and edge-vertex adjacency relationships. Subsequently, we employ Transformer-based diffusion models for sequential geometry generation, progressively generating vertex coordinates, followed by edge geometries and face geometries which are represented as B-splines. Extensive experiments on diverse CAD datasets show that DTGBrepGen significantly outperforms existing methods in both topological validity and geometric accuracy, achieving higher validity rates and producing more diverse and realistic B-reps. Our code is publicly available at https://github.com/jinli99/DTGBrepGen.

Figures (15)

• Figure 1: Overview of DTGBrepGen. The pipeline first generates a valid topological structure by two sequentially Transformer encoder-decoder networks: one to produce edge-face adjacencies, followed by another to establish edge-vertex connections. Based on this topology, the geometry generation process employs Transformer-based diffusion models to progressively generate face bounding boxes, vertex coordinates, edge geometries, and finally face geometries.
• Figure 2: Illustration of B-spline control point generation via diffusion models. Our method learns the distribution of control points to establish face geometry in B-rep models, enabling precise geometric representation.
• Figure 3: Qualitative comparison of B-rep models generated by our method, DeepCAD wu2021deepcad, and BrepGen xu2024brepgen on the DeepCAD dataset.
• Figure 4: Examples of class-conditioned generation results for different furniture categories. Each column shows two instances from the same category, demonstrating our method's ability to capture category-specific features while ensuring structural diversity.
• Figure 5: B-rep models generated based on input point clouds. Each example consists of a point cloud alongside three corresponding generated B-reps.