VoronoiNet: General Functional Approximators with Local Support

TL;DR

This paper introduces VoronoiNet, a differentiable Voronoi diagram-based architecture that enhances generative models by providing detailed 2D and 3D shape reconstructions with a compact latent space.

Contribution

It presents a novel deep architecture embedding differentiable Voronoi diagrams for improved shape representation in generative networks.

Findings

Achieves more detailed shape reconstructions in 2D and 3D.

Provides a highly compact latent embedding.

Shows promising preliminary results compared to implicit occupancy networks.

Abstract

Voronoi diagrams are highly compact representations that are used in various Graphics applications. In this work, we show how to embed a differentiable version of it -- via a novel deep architecture -- into a generative deep network. By doing so, we achieve a highly compact latent embedding that is able to provide much more detailed reconstructions, both in 2D and 3D, for various shapes. In this tech report, we introduce our representation and present a set of preliminary results comparing it with recently proposed implicit occupancy networks.