HoLa: B-Rep Generation using a Holistic Latent Representation

TL;DR

HoLa presents a holistic latent space for boundary representations (B-Reps) that unifies continuous geometry and discrete topology into a single latent vector defined over surface primitives. A neural intersection module learns the low-order curve geometry from pairs of surfaces, enabling a VAE to encode geometry and topology jointly and an LDM to generate B-Reps under diverse conditions (images, point clouds, sketches, text) with a post-processing step that yields watertight CAD models. This approach reduces ambiguities and inconsistencies of previous multi-step pipelines and achieves substantially higher validity rates than prior methods. The method demonstrates strong unconditional and conditional generation performance across multiple conditioning modalities, with robustness to imperfect inputs, while maintaining efficiency through a unified architecture. Limitations include residual surface-primitives inconsistencies and padding-induced noise, suggesting directions for future global latent compression and enhanced conditioning schemas.

Abstract

We introduce a novel representation for learning and generating Computer-Aided Design (CAD) models in the form of $\textit{boundary representations}$ (B-Reps). Our representation unifies the continuous geometric properties of B-Rep primitives in different orders (e.g., surfaces and curves) and their discrete topological relations in a $\textit{holistic latent}$ (HoLa) space. This is based on the simple observation that the topological connection between two surfaces is intrinsically tied to the geometry of their intersecting curve. Such a prior allows us to reformulate topology learning in B-Reps as a geometric reconstruction problem in Euclidean space. Specifically, we eliminate the presence of curves, vertices, and all the topological connections in the latent space by learning to distinguish and derive curve geometries from a pair of surface primitives via a neural intersection network. To this end, our holistic latent space is only defined on surfaces but encodes a full B-Rep model, including the geometry of surfaces, curves, vertices, and their topological relations. Our compact and holistic latent space facilitates the design of a first diffusion-based generator to take on a large variety of inputs including point clouds, single/multi-view images, 2D sketches, and text prompts. Our method significantly reduces ambiguities, redundancies, and incoherences among the generated B-Rep primitives, as well as training complexities inherent in prior multi-step B-Rep learning pipelines, while achieving greatly improved validity rate over current state of the art: 82% vs. $\approx$50%.

Figures (27)

• Figure 1: Disjoint vs. holistic latent B-Reps. Previous neural B-Rep models all learn separate latent spaces and their respective decoders/generators for surface, curve, vertex primitives, even their relations. In contrast, our HoLa representation is learned to blend the geometry and topology of all primitives into one unified latent space, for B-Rep generation.
• Figure 2: A B-Rep model that consists of surfaces, curves, and their topological connections is encoded into a holistic latent space via a feature fusion model. Then we emply a novel neural intersection module to recover the geometry surfaces and all the lower-order primitives from the latent space.
• Figure 3: Given a GT B-Rep model, we use convolutional, self-attention and cross-attention layers to fuse the features of surfaces and curves and their connectivity to learn a holistic latent representation for the B-Rep model.
• Figure 4: Given a pair of sampled surface latent $z_s$ from our holistic latent space, we use a neural intersection module to identify and recover the intersected curve feature. Then, the decoder network takes the sampled and recovered features to reconstruct the input B-Rep model $(S,C,T_{SC})$.
• Figure 5: Based on the learned latent representation $z_s$, we train a latent diffusion model (LDM) to generate B-Rep models conditioned on various inputs. Condition data is processed by fixed or learned feature extractors to produce a $256$-dimensional condition vector $c$. Then we use a set of self-attention layers to map the noise to the latent vector $\hat{z}$.