BrepGen: A B-rep Generative Diffusion Model with Structured Latent Geometry

BrepGen: A B-rep Generative Diffusion Model with Structured Latent Geometry

July 2024 | XIANG XU, JOSEPH G. LAMBOURNE, PRADEEP KUMAR JAYARAMAN, ZHENGGQING WANG, KARL D.D. WILLIS, YASUTAKA FURUKAWA
BrepGen is a diffusion-based generative model that directly outputs Boundary Representation (B-rep) Computer-Aided Design (CAD) models. The model represents B-rep geometry and topology in a structured latent geometry format, using a hierarchical tree structure where each node encodes the geometry and topology of a face, edge, or vertex. The root node represents the entire CAD solid, with child nodes representing the global position and local geometry of each primitive. Node duplication encodes topological relationships, such as mating and association, implicitly. The model uses Transformer-based diffusion models to sequentially denoise node features, merging duplicated nodes to recover the correct topology. Extensive experiments demonstrate BrepGen's superior performance on various benchmarks, including a newly collected Furniture B-rep Dataset, showcasing its capability to generate complex, free-form, and doubly-curved surfaces. BrepGen also supports applications like CAD autocomplete and design interpolation. The code, pretrained models, and dataset are available online.BrepGen is a diffusion-based generative model that directly outputs Boundary Representation (B-rep) Computer-Aided Design (CAD) models. The model represents B-rep geometry and topology in a structured latent geometry format, using a hierarchical tree structure where each node encodes the geometry and topology of a face, edge, or vertex. The root node represents the entire CAD solid, with child nodes representing the global position and local geometry of each primitive. Node duplication encodes topological relationships, such as mating and association, implicitly. The model uses Transformer-based diffusion models to sequentially denoise node features, merging duplicated nodes to recover the correct topology. Extensive experiments demonstrate BrepGen's superior performance on various benchmarks, including a newly collected Furniture B-rep Dataset, showcasing its capability to generate complex, free-form, and doubly-curved surfaces. BrepGen also supports applications like CAD autocomplete and design interpolation. The code, pretrained models, and dataset are available online.
Reach us at info@study.space