Investigations on Output Parameterizations of Neural Networks for Single Shot 6D Object Pose Estimation

TL;DR

This paper introduces novel neural network output parameterizations for single shot 6D object pose estimation, achieving state-of-the-art results and enabling real-world robotic grasping without extra refinement.

Contribution

It proposes new parameterizations for neural network outputs in 6D pose estimation, improving accuracy and practical applicability.

Findings

Achieved state-of-the-art performance on benchmark datasets.

Demonstrated successful robotic grasping without ICP refinement.

Proposed novel output parameterizations for neural networks.

Abstract

Single shot approaches have demonstrated tremendous success on various computer vision tasks. Finding good parameterizations for 6D object pose estimation remains an open challenge. In this work, we propose different novel parameterizations for the output of the neural network for single shot 6D object pose estimation. Our learning-based approach achieves state-of-the-art performance on two public benchmark datasets. Furthermore, we demonstrate that the pose estimates can be used for real-world robotic grasping tasks without additional ICP refinement.