Abstract:Capsule networks (CapsNets) were introduced to address convolutional neural networks limitations, learning object-centric representations that are more robust, pose-aware, and interpretable. They organize neurons into groups called capsules, where each capsule encodes the instantiation parameters of an object or one of its parts. Moreover, a routing algorithm connects capsules in different layers, thereby capturing hierarchical part-whole relationships in the data. This thesis investigates the intriguing aspects of CapsNets and focuses on three key questions to unlock their full potential. First, we explore the effectiveness of the routing algorithm, particularly in small-sized networks. We propose a novel method that anneals the number of routing iterations during training, enhancing performance in architectures with fewer parameters. Secondly, we investigate methods to extract more effective first-layer capsules, also known as primary capsules. By exploiting pruned backbones, we aim to improve computational efficiency by reducing the number of capsules while achieving high generalization. This approach reduces CapsNets memory requirements and computational effort. Third, we explore part-relationship learning in CapsNets. Through extensive research, we demonstrate that capsules with low entropy can extract more concise and discriminative part-whole relationships compared to traditional capsule networks, even with reasonable network sizes. Lastly, we showcase how CapsNets can be utilized in real-world applications, including autonomous localization of unmanned aerial vehicles, quaternion-based rotations prediction in synthetic datasets, and lung nodule segmentation in biomedical imaging. The findings presented in this thesis contribute to a deeper understanding of CapsNets and highlight their potential to address complex computer vision challenges.

Capsule Networks With Residual Pose Routing

A novel capsule network based on deep routing and residual learning

Adaptive Routing Between Capsules

Training Deep Capsule Networks with Residual Connections

Capsule networks with non-iterative cluster routing

ProtoCaps: A Fast and Non-Iterative Capsule Network Routing Method

DE-CapsNet: A Diverse Enhanced Capsule Network with Disperse Dynamic Routing

Ddrm-Capsnet: Capsule Network Based On Deep Dynamic Routing Mechanism For Complex Data

DeepCaps: Going Deeper With Capsule Networks

Generalized Capsule Networks with Trainable Routing Procedure

Global routing between capsules

Attention routing between capsules

A Receptor Skeleton for Capsule Neural Networks

Weighted-capsule Routing Via a Fuzzy Gaussian Model

On Learning and Learned Data Representation by Capsule Networks

Spiking CapsNet: A spiking neural network with a biologically plausible routing rule between capsules

A lightweight capsule network via channel-space decoupling and self-attention routing

A Fast Routing Capsule Network With Improved Dense Blocks

Efficient-CapsNet: capsule network with self-attention routing

Hierarchical Object-Centric Learning with Capsule Networks

CARNet: Densely Connected Capsules with Capsule-Wise Attention Routing