LE-CapsNet: A Light and Enhanced Capsule Network
This work addresses efficiency and accuracy problems in Capsule Networks for computer vision researchers, but it is incremental as it builds on existing CapsNet methods.
The authors tackled the issues of CapsNet being slow, resource-intensive, and less accurate than CNNs by proposing LE-CapsNet, a lighter and enhanced variant that achieves 76.73% accuracy on CIFAR-10 with 3.8M weights and runs 4x faster, while also improving robustness to affine transformations with 94.3% accuracy on AffNIST compared to CapsNet's 90.52%.
Capsule Network (CapsNet) classifier has several advantages over CNNs, including better detection of images containing overlapping categories and higher accuracy on transformed images. Despite the advantages, CapsNet is slow due to its different structure. In addition, CapsNet is resource-hungry, includes many parameters and lags in accuracy compared to CNNs. In this work, we propose LE-CapsNet as a light, enhanced and more accurate variant of CapsNet. Using 3.8M weights, LECapsNet obtains 76.73% accuracy on the CIFAR-10 dataset while performing inference 4x faster than CapsNet. In addition, our proposed network is more robust at detecting images with affine transformations compared to CapsNet. We achieve 94.3% accuracy on the AffNIST dataset (compared to CapsNet 90.52%).