JointContrast: Skeleton-Based Mutual Action Recognition with Contrastive Learning
Paper
Paper/Presentation Title | JointContrast: Skeleton-Based Mutual Action Recognition with Contrastive Learning |
---|---|
Presentation Type | Paper |
Authors | Jia, Xiangze, Zhang, Ji, Wang, Zhen, Luo, Yonglong, Chen, Fulong and Xiao, Jing |
Journal or Proceedings Title | Proceedings of the 19th Pacific Rim International Conference on Artificial Intelligence (PRICAI 2022) |
Journal Citation | 13631, pp. 478-489 |
Number of Pages | 12 |
Year | 2022 |
Publisher | Springer |
Place of Publication | Switzerland |
ISBN | 9783031208683 |
9783031208676 | |
Digital Object Identifier (DOI) | https://doi.org/10.1007/978-3-031-20868-3_35 |
Web Address (URL) of Paper | https://link.springer.com/chapter/10.1007/978-3-031-20868-3_35 |
Web Address (URL) of Conference Proceedings | https://link.springer.com/book/10.1007/978-3-031-20868-3 |
Conference/Event | 19th Pacific Rim International Conference on Artificial Intelligence (PRICAI 2022) |
Event Details | 19th Pacific Rim International Conference on Artificial Intelligence (PRICAI 2022) Parent Pacific Rim International Conference on Artificial Intelligence Delivery In person Event Date 10 to end of 13 Nov 2022 Event Location Shanghai, China |
Abstract | Skeleton-based action recognition relies on skeleton sequences to detect certain categories of human actions. In skeleton-based action recognition, it is observed that many scenes are mutual actions characterized by more than one subject, and the existing works deal with subjects independently or use the pooling layer for feature fusion leading to ineffective learning and fusion of different subjects. In this paper, we propose a novel framework, JointContrast, for Skeleton-based action recognition to deal with these challenges. Our JointContrast includes two innovative components. One is the pre-training process with a fine-grained contrastive loss that effectively enhances the representation ability of the model, and the other is an Interactive Graph (IG) representation for skeletal sequences that contributes to the fusion of features between subjects. We validate our JointContrast in the popular SBU and NTU RGB-D datasets, and experimental results show that our model outperforms other baseline methods in terms of recognition accuracy. |
Keywords | Pre-training ; Action recognition; Interactive graph ; Contrastive learning |
ANZSRC Field of Research 2020 | 460999. Information systems not elsewhere classified |
Public Notes | Files associated with this item cannot be displayed due to copyright restrictions. |
Series | Lecture Notes in Computer Science |
Byline Affiliations | Nanjing University of Aeronautics and Astronautics, China |
University of Southern Queensland | |
Zhejiang Lab, China | |
Anhui Normal University, China |
https://research.usq.edu.au/item/z58yy/jointcontrast-skeleton-based-mutual-action-recognition-with-contrastive-learning
47
total views1
total downloads4
views this month0
downloads this month