Self-attention Pyramidal Convolutional Network for Weakly-supervised Video Anomaly Detection

Video anomaly detection refers to detecting and recognising abnormal performance in videos that deviate from normal behaviour. The anomaly detection performance in weakly supervised video anomaly detection degrades due to the lack of attention to temporal information in the video features extracted by the pre-trained network. To address this problem, we propose a weakly supervised video anomaly detection method based on a self-attention pyramidal convolutional network (SAP-net), which includes a redesigned multi-scale module with a self-attention mechanism. Experimental results show the SAP-net outperforms the state-of-the-art method in the UCF-Crime dataset.