3D pose estimation for behavior analysis with deep learning

Event details
Date | 19.07.2022 |
Hour | 14:00 › 16:00 |
Speaker | Haozhe Qi |
Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Pascal Fua
Thesis advisor: Prof. Alexander Mathis
Co-examiner: Prof. Amir Zamir
Abstract
For the automated behavioral analysis, 3D pose estimation provides an ideal representation of animals. However, 3D pose estimation is less well-studied due to the lack of annotated data sets, as well as the higher computational burden compared to the well-established 2D pose estimation. To tackle those challenges, during my Ph.D study, I will focus on three sub-areas of the 3D pose estimation: multi-view single individual pose estimation, multi-view multi-individual pose estimation and egocentric pose estimation. By doing so, I will make sure that those innovations are suitable for various behavioral analysis applications (mouse behavior analysis, fish behavior analysis and measurement of patients' functional impairment respectively). Thus, I can create methods that are robust and universal and thus strongly impact to the computational ethology and neuroscience.
Background papers
1. Title: Cross View Fusion for 3D Human Pose Estimation
Authors: Haibo Qiu, Chunyu Wang, Jingdong Wang, Naiyan Wang, Wenjun Zeng.
Link: https://openaccess.thecvf.com/content_ICCV_2019/papers/Qiu_Cross_View_Fusion_for_3D_Human_Pose_Estimation_ICCV_2019_paper.pdf
2. Title: H2O: Two Hands Manipulating Objects for First Person Interaction Recognition
Authors: Taein Kwon, Bugra Tekin, Jan Stuhmer, Federica Bogo, Marc Pollefeys.
Link: https://openaccess.thecvf.com/content/ICCV2021/papers/Kwon_H2O_Two_Hands_Manipulating_Objects_for_First_Person_Interaction_Recognition_ICCV_2021_paper.pdf
3. Title: Interacting Attention Graph for Single Image Two-Hand Reconstruction
Authors: Mengcheng Li, Liang An, Hongwen Zhang, Lianpeng Wu, Feng Chen, Tao Yu, Yebin Liu.
Link: https://arxiv.org/pdf/2203.09364.pdf
Exam president: Prof. Pascal Fua
Thesis advisor: Prof. Alexander Mathis
Co-examiner: Prof. Amir Zamir
Abstract
For the automated behavioral analysis, 3D pose estimation provides an ideal representation of animals. However, 3D pose estimation is less well-studied due to the lack of annotated data sets, as well as the higher computational burden compared to the well-established 2D pose estimation. To tackle those challenges, during my Ph.D study, I will focus on three sub-areas of the 3D pose estimation: multi-view single individual pose estimation, multi-view multi-individual pose estimation and egocentric pose estimation. By doing so, I will make sure that those innovations are suitable for various behavioral analysis applications (mouse behavior analysis, fish behavior analysis and measurement of patients' functional impairment respectively). Thus, I can create methods that are robust and universal and thus strongly impact to the computational ethology and neuroscience.
Background papers
1. Title: Cross View Fusion for 3D Human Pose Estimation
Authors: Haibo Qiu, Chunyu Wang, Jingdong Wang, Naiyan Wang, Wenjun Zeng.
Link: https://openaccess.thecvf.com/content_ICCV_2019/papers/Qiu_Cross_View_Fusion_for_3D_Human_Pose_Estimation_ICCV_2019_paper.pdf
2. Title: H2O: Two Hands Manipulating Objects for First Person Interaction Recognition
Authors: Taein Kwon, Bugra Tekin, Jan Stuhmer, Federica Bogo, Marc Pollefeys.
Link: https://openaccess.thecvf.com/content/ICCV2021/papers/Kwon_H2O_Two_Hands_Manipulating_Objects_for_First_Person_Interaction_Recognition_ICCV_2021_paper.pdf
3. Title: Interacting Attention Graph for Single Image Two-Hand Reconstruction
Authors: Mengcheng Li, Liang An, Hongwen Zhang, Lianpeng Wu, Feng Chen, Tao Yu, Yebin Liu.
Link: https://arxiv.org/pdf/2203.09364.pdf
Practical information
- General public
- Free