Geometry-Informed Neural Networks

Arturs Berzins,Andreas Radler,Eric Volkmann,Sebastian Sanokowski,Sepp Hochreiter,Johannes Brandstetter

2024-10-14

Abstract:Geometry is a ubiquitous tool in computer graphics, design, and engineering. However, the lack of large shape datasets limits the application of state-of-the-art supervised learning methods and motivates the exploration of alternative learning strategies. To this end, we introduce geometry-informed neural networks (GINNs) -- a framework for training shape-generative neural fields without data by leveraging user-specified design requirements in the form of objectives and constraints. By adding diversity as an explicit constraint, GINNs avoid mode-collapse and can generate multiple diverse solutions, often required in geometry tasks. Experimentally, we apply GINNs to several validation problems and a realistic 3D engineering design problem, showing control over geometrical and topological properties, such as surface smoothness or the number of holes. These results demonstrate the potential of training shape-generative models without data, paving the way for new generative design approaches without large datasets.

Machine Learning,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to train a neural network model capable of generating shapes that meet the design requirements in the absence of a large number of shape datasets. Specifically, the authors propose Geometry - Informed Neural Networks (GINNs), which is a method for training shape - generating neural fields by using user - specified design goals and constraints in the absence of data. This method aims to avoid mode - collapse and can generate multiple diverse solutions, which is crucial for the multiple solutions often required in geometric tasks. The main contributions of the paper include: 1. Introducing the GINN framework for training shape - generating neural fields by using design constraints without relying on data and avoiding mode - collapse by adding diversity constraints. 2. Applying GINN to several verification problems and a practical 3D engineering design problem, demonstrating the ability to control geometric and topological properties. Through these contributions, the paper demonstrates the feasibility of training shape - generating models without large - scale datasets, paving the way for new generative design methods.

Geometry-Informed Neural Networks

SDF-based 3D Generative Adversarial Networks for Image and Geometry Generation

Geometry-Informed Neural Operator for Large-Scale 3D PDEs

Geometry-Aware Generative Autoencoders for Warped Riemannian Metric Learning and Generative Modeling on Data Manifolds

Disentangling Geometry and Appearance with Regularised Geometry-Aware Generative Adversarial Networks

Geometry aware 3D generation from in-the-wild images in ImageNet

Geometry Distributions

Geometric Neural Operators (GNPs) for Data-Driven Deep Learning of Non-Euclidean Operators

GeoGen: Geometry-Aware Generative Modeling via Signed Distance Functions

Deep Learning 3d Shape Surfaces Using Geometry Images

Physics-Informed Geometry-Aware Neural Operator

Inverse Graphics GAN: Learning to Generate 3D Shapes from Unstructured 2D Data

Geometric Deep Learning: Going beyond Euclidean data

Learning Geodesics of Geometric Shape Deformations From Images

Generative Teaching Networks: Accelerating Neural Architecture Search by Learning to Generate Synthetic Training Data

SurfNet: Generating 3D shape surfaces using deep residual networks

Learning Manifold Patch-Based Representations of Man-Made Shapes

Shape related constraints aware generation of Mechanical Designs through Deep Convolutional GAN

Neural networks learn to magnify areas near decision boundaries

Geometric Inductive Biases of Deep Networks: The Role of Data and Architecture

Physics-informed neural networks for transformed geometries and manifolds