Adaptive Capsule Network

Jianwei Tao, Xiankun Zhang, Xuexiong Luo, Yuan Wang, Chen Song, Yue Sun

Abstract: Highlights•We propose a novel adaptive capsule layer to generate space-related and high-representation capsule vectors.•Our model structure is more efficient and convenient than existing state-of-the-art CapsNets.•The proposed approach can improve the performance of CapsNets in terms of classification accuracy and robustness