HiDiGen: Hierarchical Diffusion for B-Rep Generation with Explicit Topological Constraints

Shurui Liu; Weide Chen; Ancong Wu

cs.CVApr 3, 2026

HiDiGen: Hierarchical Diffusion for B-Rep Generation with Explicit Topological Constraints

Authors: Shurui Liu, Weide Chen, Ancong Wu

Organizations: School of Computer Science and Engineering, Sun Yat-sen University, China · School of Intelligent Systems Engineering, Shenzhen Campus of Sun Yat-sen University, China

Abstract

Boundary representation (B-rep) is the standard 3D modeling format in CAD systems, encoding both geometric primitives and topological connectivity. Despite its prevalence, deep generative modeling of valid B-rep structures remains challenging due to the intricate interplay between discrete topology and continuous geometry. In this paper, we propose HiDiGen, a hierarchical generation framework that decouples geometry modeling into two stages, each guided by explicitly modeled topological constraints. Specifically, our approach first establishes face-edge incidence relations to define a coherent topological scaffold, upon which face proxies and initial edge curves are generated. Subsequently, multiple Transformer-based diffusion modules are employed to refine the geometry by generating precise face surfaces and vertex positions, with edge-vertex adjacencies dynamically established and enforced to preserve structural consistency. This progressive geometry hierarchy enables the generation of more novel and diverse shapes, while two-stage topological modeling ensures high validity. Experimental results show that HiDiGen achieves strong performance, generating novel, diverse, and topologically sound CAD models.

Figures & tables

Figure 1: Comparison of learning paradigms for B-rep generation : (a) implicit topology encoding, (b) decoupled topology-geometry generation, and (c) our hierarchical generation framework.

Figure 2: HiDiGen pipeline overview. Our framework generates B-rep models in two hierarchical levels: Level 1 produces face-edge topology ( $\mathbf{EF}_{i}$ ) and coarse geometry (face bounding boxes $\mathbf{B}_{i}$ , edge curves $\mathbf{E}_{i}$ ); Level 2 refines edge-vertex topology ( $\mathbf{EV}_{i}$ ) and detailed geometry (face surfaces $\mathbf{F}_{i}$ , vertex positions $\mathbf{V}_{i}$ ), with each stage conditioned on prior outputs.

Figure 3: Illustration of the contextual conditioned embedding for face geometry generation. We gather the corresponding face bounding boxes and edge geometries, and further derive topological connectivity relationships to form the conditioning embedding.

Dataset		MMD $\downarrow$	COV $\uparrow$	JSD $\downarrow$	Valid $\uparrow$	Unique $\uparrow$	Novel $\uparrow$	CC $\uparrow$	MC $\uparrow$
DeepCAD	DeepCAD	$1.021$	$72.43$	$1.623$	$60.50$	$91.27$	$93.74$	$10.51$	$1.81$
	BrepGen	$0.874$	$71.13$	$1.584$	$69.27$	$99.08$	$99.54$	$9.30$	$0.86$
	DGTBrepGen	$0.867$	76.83	$1.504$	$73.27$	$99.13$	$99.21$	$9.78$	$0.73$
	HiDiGen	0.856	$71.37$	1.424	75.77	99.52	99.71	11.03	1.99
ABC	BrepGen	$1.543$	$66.37$	$2.377$	$53.47$	$97.64$	$98.57$	$10.76$	$1.37$
	DGTBrepGen	$1.455$	68.20	$2.024$	$63.53$	$98.86$	$99.02$	$11.45$	$1.06$

Table 1: Quantitative comparison of shape generation methods on the DeepCAD and ABC datasets. Lower values are better for MMD-CD and JSD (both scaled by

10^{2}

), while higher values are better for COV, Valid, Unique, Novel, CC, and MC. All metrics except CC and MC are reported as percentages. Best results are bolded .

Figure 4: Qualitative comparison of unconditioned B-rep models generated by our method and prior approaches, including DeepCAD [ 24 ] , BrepGen [ 26 ] , and DTGBrepGen [ 14 ] , on the DeepCAD dataset. Our method generates more versatile and consistently watertight solid B-reps, demonstrating geometric fidelity and topological correctness.

Figure 5: HiDiGen failure cases illustrating error propagation between topology and geometry: (a) erroneous topology yields incorrect geometry; (b) inaccurate geometry leads to spurious topological connections.

Figure 6: Point cloud conditioned B-rep generation.

Dataset		MMD $\downarrow$	COV $\uparrow$	JSD $\downarrow$	Valid $\uparrow$
DeepCAD	w/o $\mathbf{B}_{i}$	$0.869$	$70.23$	$1.437$	$75.03$
	w/o $\mathbf{E}_{i}$	$0.871$	$70.87$	$1.441$	$73.27$
	HiDiGen	0.856	71.37	1.424	75.77
ABC	w/o $\mathbf{B}_{i}$	$1.483$	$63.27$	$1.873$	$64.83$
	w/o $\mathbf{E}_{i}$	$1.579$	$61.40$	$1.948$	$62.80$
	HiDiGen	1.414	65.43	1.845	65.77

Table 2: Quantitative comparison of HiDiGen with ablation variants on DeepCAD and ABC datasets.

Figure 7: Failure cases illustrating topological and geometric inconsistencies in generated B-reps.

Figure 8: Additional qualitative results of HiDiGen on the DeepCAD datasets.

Figure 9: Additional qualitative results of HiDiGen on the ABC datasets.

3D Shape Generation

3D shape generation creates and edits 3D objects from learned geometric priors and text, images, sketches, or structural constraints.

See all 104 papers

HiFi-BRep: High-Fidelity Latent Representation for Robust B-Rep Generation
Aug 17, 2026Junhao Hou, Chenqi Luo, Pufan Wang +53D Shape Generation Constrained Generative Modeling
State Key Lab of CAD&CG, Zhejiang University · Hangzhou Dianzi University · Guangzhou University
UniBRep: Learning Unified Geometry and Topology for Image-conditioned B-Rep Generation
Oct 2, 2026Haiyang Ying, Allen Tu, Jiaye Wu +2Parametric CAD Modeling Single-View 3D Reconstruction
University of Maryland, College Park, Maryland, USA · Amazon, Bellevue, Washington, USA
Towards Valid B-Rep Generation: Training-Free Wireframe Anomaly Detection and Repair
Aug 5, 2026Jingyu Wu, Youcheng Cai, Tengyu Luo +1Anomaly Detection
University of Science and Technology of China, Hefei, China

Aug 17, 2026J/K move · Enter open · S save

HiFi-BRep: High-Fidelity Latent Representation for Robust B-Rep Generation

Junhao Hou, Chenqi Luo, Pufan Wang, Jiaying Lu+4

State Key Lab of CAD&CG, Zhejiang University · Hangzhou Dianzi University · Guangzhou University

3D Shape Generation Constrained Generative Modeling Geometric Representation Learning

Open paper AskPDF arXiv

Boundary representation (B-Rep) generation is a fundamental task in computer-aided design, yet the direct synthesis of high-fidelity and structurally valid B-Reps remains a major challenge. Existing deep generative methods suffer from two forms of brittleness: representation brittleness, caused by padding noise and feature contamination in the latent space, and generation brittleness, stemming from sequential error propagation and a train-inference mismatch due to non-differentiable validity enforcement. We propose HiFi-BRep, a novel framework that addresses these limitations through two synergistic contributions. First, a topology-aware encoder constructs a high-fidelity latent representation by eliminating padding via learnable queries and preventing feature contamination with topology-guided attention. Second, a single-stage decoder jointly predicts geometry and topology in parallel, embedding core manifold constraints as a differentiable learning objective. This design ensures mutual guidance between geometry and topology while avoiding cascaded errors. Extensive experiments show that HiFi-BRep significantly outperforms state-of-the-art methods in both structural validity and geometric fidelity, providing a robust solution for high-quality B-Rep synthesis. Code and models are publicly available at https://github.com/1nnoh/HiFi-BRep.

Abstract

3D Shape Generation

Explore similar work

HiFi-BRep: High-Fidelity Latent Representation for Robust B-Rep Generation

UniBRep: Learning Unified Geometry and Topology for Image-conditioned B-Rep Generation

Towards Valid B-Rep Generation: Training-Free Wireframe Anomaly Detection and Repair