MCN: Magnitude-Continuity Network for Video Anomaly Detection under Weak Supervision
Xiaohu Luan, Hong-min ZHANG · 2023
Video anomaly detection is a complex task that combines anomaly detection and video image processing. Video anomaly detection aims to identify abnormal behavior in video and is widely used in the real world. In existing works, Feature-Magnitude have shown good performance in weakly supervised video anomaly detection. However, Feature Magnitude-based methods is easily affected by the environment, resulting in error detection. Motivated by the fact that the abnormal events in real video are often continuous, we propose a Magnitude-Contiuity Network. In addition, in order to further improve the robustness of existing Feature Magnitude-based methods, the dynamic Feature-Magnitude learning loss function is used. Experimental results on UCF-Crime and ShanghaiTech datasets show that the robustness and performance of the proposed method are significantly improved.