DocumentCode :
1872812
Title :
Efficient initialization of Mixtures of Experts for human pose estimation
Author :
Ning, Huazhong ; Hu, Yuxiao ; Huang, Thomas
Author_Institution :
ECE Department, U. of Illinois at Urbana-Champaign, 61801, USA
fYear :
2008
fDate :
12-15 Oct. 2008
Firstpage :
2164
Lastpage :
2167
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.
Keywords :
Bayesian methods; Computational efficiency; Costs; Humans; Image converters; Kernel; Partitioning algorithms; Robustness; Testing; Videos; Bayesian Mixtures of Experts; Human Pose Estimation; Initialization;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Image Processing, 2008. ICIP 2008. 15th IEEE International Conference on
Conference_Location :
San Diego, CA, USA
ISSN :
1522-4880
Print_ISBN :
978-1-4244-1765-0
Electronic_ISBN :
1522-4880
Type :
conf
DOI :
10.1109/ICIP.2008.4712217
Filename :
4712217
Link To Document :
بازگشت