Loading...
Please wait, while we are loading the content...
Similar Documents
MODEC: Multimodal Decomposable Models for Human Pose Estimation
| Content Provider | CiteSeerX |
|---|---|
| Author | Taskar, Ben |
| Abstract | We propose a multimodal, decomposable model for articulated human pose estimation in monocular images. Most approaches for this problem use a single linear model, which make it difficult to capture the wide range of appearance present in realistic, unconstrained images. In this paper, we propose a model of human pose that explicitly captures a variety of pose modes. Unlike other multimodal models, our approach includes both holistic and local cues and uses a convex objective and joint training for mode selection and pose estimation. We also employ a cascaded mode selection step which controls the trade-off between speed and accuracy, yielding a 5x speedup in inference and learning. Our model significantly outperforms state-of-theart approaches across the accuracy-speed trade-off curve for several pose datasets. This includes our newly-collected dataset of people in movies, which contains an order of magnitude more labeled data for training and testing than existing datasets. 1. |
| File Format | |
| Access Restriction | Open |
| Subject Keyword | Pose Mode Convex Objective Several Pose Datasets Unconstrained Image Labeled Data Decomposable Model Pose Estimation Monocular Image Cascaded Mode Selection Step Wide Range Mode Selection Problem Use State-of-theart Approach Single Linear Model Articulated Human Pose Estimation Multimodal Decomposable Model Multimodal Model Accuracy-speed Trade-off Curve Newly-collected Dataset Local Cue Joint Training Human Pose Estimation Appearance Present |
| Content Type | Text |
