Neural Convolutional Surfaces

TL;DR

This paper introduces a neural architecture for representing 3D shapes that separates global structure from local details, enabling efficient compression and flexible manipulation of shape features without supervision.

Contribution

It presents a novel neural pipeline and architecture that achieve unsupervised disentanglement of shape geometry, improving compression and enabling shape detail transfer.

Findings

Outperforms existing methods in neural shape compression.

Allows manipulation of global and local shape features independently.

Enables transfer of shape details between surfaces.

Abstract

This work is concerned with a representation of shapes that disentangles fine, local and possibly repeating geometry, from global, coarse structures. Achieving such disentanglement leads to two unrelated advantages: i) a significant compression in the number of parameters required to represent a given geometry; ii) the ability to manipulate either global geometry, or local details, without harming the other. At the core of our approach lies a novel pipeline and neural architecture, which are optimized to represent one specific atlas, representing one 3D surface. Our pipeline and architecture are designed so that disentanglement of global geometry from local details is accomplished through optimization, in a completely unsupervised manner. We show that this approach achieves better neural shape compression than the state of the art, as well as enabling manipulation and transfer of shape details. Project page at http://geometry.cs.ucl.ac.uk/projects/2022/cnnmaps/ .