Convolutional neural networks, despite their profound impact in countless domains, suffer from spatial ambiguities and a lack of robustness to pose variations. Capsule networks can likely alleviate these issues by storing and routing the pose information of features through their architectures, seeking agreement between the lower-level predictions of higher-level poses.
We make several contributions to advance the algorithms of capsule networks, with specific real-world applications in biomedical imaging data segmentation and classification: (1) The first ever capsule-based segmentation network in the literature, SegCaps, provided five major novelties. (2) A capsule-based diagnosis network, D-Caps, introduced a novel capsule-average pooling layer. (3) An explainable capsule network, X-Caps, encodes high-level visual object attributes within its capsules by utilizing a multi-task framework and a novel routing sigmoid function, to provide predictions with human-level explanations and a meaningful confidence score.