Loading...
Please wait, while we are loading the content...
Similar Documents
Efficient initialization of mixtures of experts for human pose estimation.
| Content Provider | CiteSeerX |
|---|---|
| Author | Ning, Huazhong Hu, Yuxiao Huang, Thomas |
| 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 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. |
| File Format | |
| Access Restriction | Open |
| Subject Keyword | Bme Model Partition Subset Bayesian Mixture Total Regression Error First Partition Single Expert Initial Value Suboptimal Local Maximum Real Dataset Quasi-synthetic Dataset Final Solution Human Pose Efficient Initialization Method Computational Cost Em Algorithm Single Monocular Image Multimodal Image-to-pose Distribution Initialization Method |
| Content Type | Text |
| Resource Type | Article |
