BDARS_CapsNet: Bi-Directional Attention Routing Sausage Capsule Network

Xin Ning, Weijuan Tian, Weijun Li, Yueyue Lu, Shuai Nie, Linjun Sun, Ziheng ChenIEEE Access 2020 · OpenReview

Introduction

Convolutional Neural Networks (CNNs) have long been the standard for computer vision. However, their reliance on pooling layers leads to the "Picasso problem," where spatial hierarchies are ignored—for instance, a face with a mouth above the eyes might still be classified as a face. Capsule Networks (CapsNets) were introduced to address this by using vectors to represent entities and their poses. Despite their promise, original CapsNets suffered from computational inefficiency and difficulty in handling complex datasets. The BDARS_CapsNet architecture addresses these challenges through two key innovations: bi-directional attention routing and a sausage measure model.

The BDARS_CapsNet Architecture

The proposed network consists of a four-layer architecture designed to maximize information flow and feature representation.

The BDARS_CapsNet architecture integrating bi-directional attention routing and the sausage measure model.

As shown in the architecture diagram, the model transitions from a standard convolutional layer to a Primary Caps layer, followed by a Convolutional Capsule layer, and finally a classification layer. The critical transitions between these layers are governed by bi-directional attention routing and the sausage measure.

Core Innovations

Bi-Directional Attention Routing

Traditional CapsNets use iterative "routing-by-agreement," which is computationally expensive and often unstable. BDARS_CapsNet replaces this with a dual-attention mechanism:

  • Bottom-up attention: Facilitates information feed-forward, identifying how parts constitute a whole.
  • Top-down attention: Provides a feedback mechanism, using high-level category information to refine low-level feature extraction.

This approach allows the routing weights to be learned during standard backpropagation, making the network much faster and more stable than previous iterations.

Sausage Measure Model

To improve how the network measures the relationship between capsules, the authors introduced the "sausage measure model."

Visualization of the Sausage Model used for measuring capsule similarities.

Inspired by covering learning, this model projects capsules onto a 1D manifold (the "sausage"). Unlike simple dot products, this measure captures both similarities and differences on a complex curved surface. By utilizing a Gaussian activation function, the network maintains local responsiveness, allowing it to approximate non-linear functions with high precision while preserving the distinct attributes of each capsule entity.

Experimental Performance

The model was rigorously tested on benchmarks including MNIST, affNIST, CIFAR-10, and SVHN.

  • Accuracy: BDARS_CapsNet achieved 99.67% on MNIST and 82.30% on CIFAR-10 (without data augmentation), outperforming standard CapsNets and EM-routing models.
  • Robustness: On the affNIST dataset, which specifically tests a network's ability to handle tilted or resized images, the model reached 99.44% accuracy, significantly beating the 93.83% of the original CapsNet.
  • Efficiency: The authors demonstrate that their variant achieves better results with 81% fewer parameters than the baseline DeepCaps model, proving that complexity does not always require massive parameter counts.

Conclusion

BDARS_CapsNet represents a significant step forward in disentangled representation learning. By treating routing as a bi-directional attention problem and employing the sausage manifold for feature geometry, the network achieves superior accuracy and robustness with reduced computational overhead. Future research will likely explore topological data analysis to further refine the structural understanding of capsule distributions in even more complex datasets like ImageNet.

Summary updated

Abstract

In order to improve the accuracy of capsule network in disentangled representation, and further expand its application in computer vision, a novel BDARS_CapsNet (bi-directional attention routing sausage capsule network) architecture is proposed in this paper. Firstly, the bi-directional routing, namely bottom-up and top-down attention is used to achieve information feed-forward and feedback mechanism, which contributes to describing the attributes of object entity more accurately and completely. Secondly, inspired by the concept of covering learning, the sausage measure model is introduced into the network. The sausage model measures both the similarities and differences of the capsules and projects them into a more complex curved surface, which makes it possible to approximate any nonlinear function with arbitrary precision and preserving the local responsiveness of capsule entity to the maximum. Finally, the BDARS_CapsNet combines the CNN (Convolutional Neural Network), bi-directional attention routing, and sausage measure into capsule network modeling, and makes full use of high-level category information and low-level vision information; as a result, the reconstruction and classification accuracy is accordingly improved. Experiments demonstrate the effectiveness of proposed information routing, sausage measure, and new framework. Furthermore, the proposed BDARS_CapsNet provides a foundation for future research on disentangled representation learning.
BibTeX
@inproceedings{ning2020bdars,
  author = {Xin Ning and Weijuan Tian and Weijun Li and Yueyue Lu and Shuai Nie and Linjun Sun and Ziheng Chen},
  title = {{BDARS\_CapsNet: Bi-Directional Attention Routing Sausage Capsule Network}},
  booktitle = {IEEE Access},
  year = {2020},
  url = {https://openreview.net/forum?id=lNFMj33q4V},
}
Figure