|
Xinghan Wang 王星瀚 I am a Ph.D. candidate at Wangxuan Institute of Computer Technology in Peking University, advised by Prof. Yadong Mu. Prior to that, I received my B.S. degree (Summa Cum Laude) in artificial intelligence (Turing Class) from Peking University. |
|
ResearchMy research interests lie in human motion analysis, temporal modeling and multi-modal learning. |
|
Text-controlled Motion Mamba: Text-Instructed Temporal Grounding of Human Motion
Xinghan Wang, Zixi Kang, Yadong Mu arXiv, 2024 paper / bibtex We introduce the text-based human motion grounding task, along with a new text-motion dataset with temporal and textual annotations. A model called TM-Mamba is proposed which can selectively propagates information using a text-controlled mechanism within the Mamba framework. |
|
Localized Linear Temporal Dynamics for Self-supervised Skeleton Action Recognition
Xinghan Wang, Yadong Mu IEEE Transactions on Multimedia, 2024 paper / bibtex A temporal contrastive learning framework for self-supervised skeleton action recognition that leverages segment-level localized linear dynamics to acquire discriminative temporal evolution patterns. |
|
Neural Koopman pooling: Control-inspired Temporal Dynamics Encoding for Skeleton-based Action Recognition
Xinghan Wang, Xin Xu, Yadong Mu CVPR, 2023 paper / code / bibtex A novel parameterized high-order temporal pooling method based on Koopman theory for skeleton action recognition. |
|
Learning Temporal Co-attention Models for Unsupervised Video Action Localization
Guoqiang Gong, Xinghan Wang, Yadong Mu, Qi Tian CVPR, 2020 paper / code / bibtex A fully unsupervised framework for video action localization based on temporal co-attention, clustering and iterative optimization. |
Education
|
Experiences
|
Academic Services
|
|
Template from Jon Barron |