ME-CapsNet: A Multi-Enhanced Capsule Networks with Routing Mechanism
This work addresses a domain-specific problem for researchers in computer vision by improving Capsule Networks for complex datasets, though it appears incremental as it builds on existing methods.
The paper tackles the challenge of Capsule Networks performing poorly on complex datasets due to excessive feature information by proposing ME-CapsNet, which integrates deeper convolutional layers with Squeeze-Excitation blocks and capsule layers, achieving higher accuracy with minimal model complexity on complex datasets.
Convolutional Neural Networks need the construction of informative features, which are determined by channel-wise and spatial-wise information at the network's layers. In this research, we focus on bringing in a novel solution that uses sophisticated optimization for enhancing both the spatial and channel components inside each layer's receptive field. Capsule Networks were used to understand the spatial association between features in the feature map. Standalone capsule networks have shown good results on comparatively simple datasets than on complex datasets as a result of the inordinate amount of feature information. Thus, to tackle this issue, we have proposed ME-CapsNet by introducing deeper convolutional layers to extract important features before passing through modules of capsule layers strategically to improve the performance of the network significantly. The deeper convolutional layer includes blocks of Squeeze-Excitation networks which use a stochastic sampling approach for progressively reducing the spatial size thereby dynamically recalibrating the channels by reconstructing their interdependencies without much loss of important feature information. Extensive experimentation was done using commonly used datasets demonstrating the efficiency of the proposed ME-CapsNet, which clearly outperforms various research works by achieving higher accuracy with minimal model complexity in complex datasets.