PointCaps: Raw point cloud processing using capsule networks with Euclidean distance routing

Raw point cloud processing using capsule networks is widely adopted in classification, reconstruction, and segmentation due to its ability preserve spatial agreement of the input data. However, most existing based network approaches are computationally heavy fail at representing entire as a single capsule. We address these limitations by proposing PointCaps, novel convolutional architecture with parameter sharing. Along we propose Euclidean distance routing algorithm class-independent latent representation. The representation captures physically interpretable geometric parameters cloud, dynamic routing, PointCaps well-represents (point-to-part) relationships points. has significantly lower number requires FLOPs while achieving better reconstruction comparable classification accuracy for raw clouds compared state-of-the-art networks.