NeuroNURBS: Learning Efficient Surface Representations for 3D Solids

TL;DR

NeuroNURBS introduces a novel method for directly encoding NURBS surface parameters, significantly improving efficiency and precision in 3D solid representation over traditional UV-grid methods.

Contribution

It proposes NeuroNURBS, a learning-based approach that encodes NURBS surfaces directly, reducing GPU and memory usage while enhancing surface generation quality.

Findings

GPU consumption reduced by 86.7% during training

Memory requirement drops by 79.9% for storing 3D solids

Improves FID score from 30.04 to 27.24 in solid generation

Abstract

Boundary Representation (B-Rep) is the de facto representation of 3D solids in Computer-Aided Design (CAD). B-Rep solids are defined with a set of NURBS (Non-Uniform Rational B-Splines) surfaces forming a closed volume. To represent a surface, current works often employ the UV-grid approximation, i.e., sample points uniformly on the surface. However, the UV-grid method is not efficient in surface representation and sometimes lacks precision and regularity. In this work, we propose NeuroNURBS, a representation learning method to directly encode the parameters of NURBS surfaces. Our evaluation in solid generation and segmentation tasks indicates that the NeuroNURBS performs comparably and, in some cases, superior to UV-grids, but with a significantly improved efficiency: for training the surface autoencoder, GPU consumption is reduced by 86.7%; memory requirement drops by 79.9% for storing 3D solids. Moreover, adapting BrepGen for solid generation with our NeuroNURBS improves the FID from 30.04 to 27.24, and resolves the undulating issue in generated surfaces.