Learning Orientation Distributions for Object Pose Estimation
Brian Okorn
Mengyun Xu
Martial Hebert
David Held
Multi-modal distributions estimated by our Learned Comparison Histogram approach. Distributions are generated for the tuna can (multi-modal), bowl (symmetric) and sugar box (uni-modal) using PoseCNN featurizations of the top right image.


For robots to operate robustly in the real world, they should be aware of their uncertainty. However, most methods for object pose estimation return a single point estimate of the object's pose. In this work, we propose two learned methods for estimating a distribution over an object's orientation. Our methods take into account both the inaccuracies in the pose estimation as well as the object symmetries. Our first method, which regresses from deep learned features to an isotropic Bingham distribution, gives the best performance for orientation distribution estimation for non-symmetric objects. Our second method learns to compare deep features and generates a non-parameteric histogram distribution. This method gives the best performance on objects with unknown symmetries, accurately modeling both symmetric and non-symmetric objects, without any requirement of symmetry annotation. We show that both of these methods can be used to augment an existing pose estimator. Our evaluation compares our methods to a large number of baseline approaches for uncertainty estimation across a variety of different types of objects.

Teaser Video


The base pose estimator generates an orientation and a featurization φ of the input, one or both of which are used to estimate a uncertainty distribution over possible poses. We render this distribution as a heat map in axis angle space, lower right, with each orientation being plotted as point in the directions of the axis of rotation and at a distance away form the origin equal to the angle of rotation.

Paper and Supplementary Material

B. Okorn*, Q. Gu*, M. Hebert, D. Held.
Learning Orientation Distributions for Object Pose Estimation.
In IROS, 2020.
(Main Paper, Supplemental)



This template was originally made by Phillip Isola and Richard Zhang for a colorful ECCV project; the code can be found here.