Quaternion Capsule Networks

TL;DR

This paper introduces Quaternion Capsule Networks that utilize quaternion algebra to represent pose information, leading to improved generalization to novel viewpoints, fewer parameters, and competitive performance on benchmark datasets.

Contribution

The paper proposes Quaternion Capsules (QCN), a novel capsule architecture using quaternions for pose representation, enhancing rotation regularization and parameter efficiency.

Findings

QCN generalizes better to unseen viewpoints.

QCN requires fewer parameters than matrix-based capsules.

QCN achieves comparable or superior performance on benchmarks.

Abstract

Capsules are grouping of neurons that allow to represent sophisticated information of a visual entity such as pose and features. In the view of this property, Capsule Networks outperform CNNs in challenging tasks like object recognition in unseen viewpoints, and this is achieved by learning the transformations between the object and its parts with the help of high dimensional representation of pose information. In this paper, we present Quaternion Capsules (QCN) where pose information of capsules and their transformations are represented by quaternions. Quaternions are immune to the gimbal lock, have straightforward regularization of the rotation representation for capsules, and require less number of parameters than matrices. The experimental results show that QCNs generalize better to novel viewpoints with fewer parameters, and also achieve on-par or better performances with the state-of-the-art Capsule architectures on well-known benchmarking datasets.