Capsule networks for 3D pose estimation in computer graphics

Abstract

Pose estimation is an important task for novel engineering applications, such as virtual and augmented reality (VR/AR), pose-based video games, object reconstruction, target tracking, driving assistance and recent sports analytics. Commonly, an efficient pose estimation system depends on the pose visualization given by a 3D configuration of location, orientation, and scaling parameters of the target. In this work we implement Capsule Networks to solve 3D pose estimation in computer graphics of rigid objects using a multi-GPU architecture