June 11, 2014Open Access
Joint Training of a Convolutional Network and a Graphical Model for Human Pose Estimation
Key Points
- This research aims to develop a hybrid model for human pose estimation using deep learning and graphical models.
- Created a hybrid architecture combining a convolutional network and a Markov random field.
- Applied the architecture to articulate human pose estimation in monocular images.
- Utilized joint training to enhance model performance based on structural domain constraints.
- Achieved significant improvements in pose estimation accuracy compared to existing state-of-the-art methods.
- Demonstrated the effectiveness of joint training in utilizing geometric relationships among body joints.
Abstract
This paper proposes a new hybrid architecture that consists of a deep Network and a Markov Random Field. We show how this architecture successfully applied to the challenging problem of articulated human pose in monocular images. The architecture can exploit structural domain such as geometric relationships between body joint locations. We that joint training of these two model paradigms improves performance and us to significantly outperform existing state-of-the-art techniques.
Connected Papers
Building similarity graph...
Analyzing shared references across papers
Discussion
Loading...
Authors
J
Jonathan Tompson
Supélec
A
Arjun Jain
Linde (United States)
Yann LeCun
Courant Institute of Mathematical Sciences
Actions
Institutions
Supélec
University of Applied Sciences and Arts of Southern Switzerland
Shandong University of Political Science and Law
References and Citations
Connected Papers
Building similarity graph...
Analyzing shared references across papers
Discussion
Loading...