Shape-Pose Disentanglement using SE(3)-equivariant Vector Neurons

TL;DR

This paper presents an unsupervised method using SE(3)-equivariant Vector Neurons to encode point clouds into canonical shape representations by disentangling shape and pose, ensuring pose-invariance and semantic alignment.

Contribution

The paper introduces a novel auto-encoder architecture based on Vector Neuron Networks that achieves pose-invariant shape encoding with combined rotation and translation equivariance.

Findings

The method produces stable, pose-invariant shape encodings.

It achieves semantic alignment of different shapes within the same class.

Experiments demonstrate superior stability and consistency over existing methods.

Abstract

We introduce an unsupervised technique for encoding point clouds into a canonical shape representation, by disentangling shape and pose. Our encoder is stable and consistent, meaning that the shape encoding is purely pose-invariant, while the extracted rotation and translation are able to semantically align different input shapes of the same class to a common canonical pose. Specifically, we design an auto-encoder based on Vector Neuron Networks, a rotation-equivariant neural network, whose layers we extend to provide translation-equivariance in addition to rotation-equivariance only. The resulting encoder produces pose-invariant shape encoding by construction, enabling our approach to focus on learning a consistent canonical pose for a class of objects. Quantitative and qualitative experiments validate the superior stability and consistency of our approach.