Hybrid Bayesian Eigenobjects: Combining Linear Subspace and Deep Network Methods for 3D Robot Vision

  • 2018-06-20 17:57:56
  • Benjamin Burchfiel, George Konidaris
  • 8

Abstract

We introduce Hybrid Bayesian Eigenobjects (HBEOs), a novel representation for3D objects designed to allow a robot to jointly estimate the pose, class, andfull 3D geometry of a novel object observed from a single viewpoint in a singlepractical framework. By combining both linear subspace methods and deepconvolutional prediction, HBEOs efficiently learn nonlinear objectrepresentations without directly regressing into high-dimensional space. HBEOsalso remove the onerous and generally impractical necessity of input datavoxelization prior to inference. We experimentally evaluate the suitability ofHBEOs to the challenging task of joint pose, class, and shape inference onnovel objects and show that, compared to preceding work, HBEOs offerdramatically improved performance in all three tasks along with several ordersof magnitude faster runtime performance.

 

Quick Read (beta)

loading the full paper ...