Efficient initialization of mixtures of experts for human pose estimation

Huazhong Ning, Yuxiao Hu, Thomas Huang

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

This paper addresses the problem of recovering 3D human pose from a single monocular image. In the literature, Bayesian Mixtures of Experts (BME) was successfully used to represent the multimodal image-to-pose distributions. However, the expectation-maximization (EM) algorithm that learns the BME model may converge to a suboptimal local maximum. And the quality of the final solution depends largely on the initial values. In this paper, we propose an efficient initialization method for BME learning. We first partition the training set so that each subset can be well modeled by a single expert and the total regression error is minimized. Then each expert and gate of BME model is initialized on a partition subset. Our initialization method is tested on both a quasi-synthetic dataset and a real dataset (HumanEva). Results show that it greatly reduces the computational cost in training while improves testing accuracy.

Original languageEnglish (US)
Title of host publication2008 IEEE International Conference on Image Processing, ICIP 2008 Proceedings
Pages2164-2167
Number of pages4
DOIs
StatePublished - 2008
Event2008 IEEE International Conference on Image Processing, ICIP 2008 - San Diego, CA, United States
Duration: Oct 12 2008Oct 15 2008

Publication series

NameProceedings - International Conference on Image Processing, ICIP
ISSN (Print)1522-4880

Other

Other2008 IEEE International Conference on Image Processing, ICIP 2008
CountryUnited States
CitySan Diego, CA
Period10/12/0810/15/08

Keywords

  • Bayesian Mixtures of Experts
  • Human pose estimation
  • Initialization

ASJC Scopus subject areas

  • Software
  • Computer Vision and Pattern Recognition
  • Signal Processing

Fingerprint Dive into the research topics of 'Efficient initialization of mixtures of experts for human pose estimation'. Together they form a unique fingerprint.

Cite this