GlanceVAD: Exploring Glance Supervision for Label-efficient Video Anomaly Detection
Huaxin Zhang,Xiang Wang,Xiaohao Xu,Xiaonan Huang,Chuchu Han,Yuehuan Wang,Changxin Gao,Shanjun Zhang,Nong Sang
DOI: https://doi.org/10.48550/arxiv.2403.06154
2024-01-01
Abstract:In recent years, video anomaly detection has been extensively investigated inboth unsupervised and weakly supervised settings to alleviate costly temporallabeling. Despite significant progress, these methods still suffer fromunsatisfactory results such as numerous false alarms, primarily due to theabsence of precise temporal anomaly annotation. In this paper, we present anovel labeling paradigm, termed "glance annotation", to achieve a betterbalance between anomaly detection accuracy and annotation cost. Specifically,glance annotation is a random frame within each abnormal event, which can beeasily accessed and is cost-effective. To assess its effectiveness, we manuallyannotate the glance annotations for two standard video anomaly detectiondatasets: UCF-Crime and XD-Violence. Additionally, we propose a customizedGlanceVAD method, that leverages gaussian kernels as the basic unit to composethe temporal anomaly distribution, enabling the learning of diverse and robustanomaly representations from the glance annotations. Through comprehensiveanalysis and experiments, we verify that the proposed labeling paradigm canachieve an excellent trade-off between annotation cost and model performance.Extensive experimental results also demonstrate the effectiveness of ourGlanceVAD approach, which significantly outperforms existing advancedunsupervised and weakly supervised methods. Code and annotations will bepublicly available at https://github.com/pipixin321/GlanceVAD.