Anomalous Action Recognition via Spatio-temporal Relation and Key Patch Selection
Keywords: Anomalous Action Recognition; Spatio-temporal Relation;Key Patch Selection
TL;DR: The Spatio-Temporal Key Patch Selection Network (SKPS-Net) is proposed for anomalous action recognition, which dynamically selects key patches with spatio-temporal information.
Abstract: For providing timely warnings and preventing potential damages, it is crucial to detect anomalous actions that threaten public safety through surveillance cameras. Compared to normal actions, anomalous actions often occupy only a small portion of surveillance videos and exhibit more complex manifestations in terms of time and space. Considering that normal action recognition methods fail to highlight crucial information from small-sized patches, resulting in imprecise anomaly modeling, we propose the Spatio-Temporal Key Patch Selection Network (SKPS-Net). To tackle the challenge of detecting anomalous behaviors that manifest in small and inconspicuous areas, we design a spatial adaptive key patch selection module to select small but informative patches on input videos. Furthermore, the long-short feature map spatio-temporal relation module is devised to make the key patch effectively capture the continuous dynamic changes of anomalous actions. Finally, we propose a spatio-temporal refined loss to reinforce fine-grained feature learning. Experiments conducted on the HMDB51, Kinetics, and UCF-Crime v2 datasets demonstrate that our SKPS-Net achieves state-of-the-art performance in few-shot action recognition, outperforming the most competitive methods by 1.2% on the anomalous action dataset UCF-Crime v2.
Primary Area: transfer learning, meta learning, and lifelong learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 2522
Loading