Distilling Aggregated Knowledge for Weakly-Supervised Video Anomaly Detection

2024-06-05

arXiv.org

AI Summary

To address the challenges of limited training samples, video-level supervision only, and extreme class imbalance in weakly supervised video anomaly detection, this paper proposes a two-tier knowledge distillation framework coupled with a decoupled cross-attention feature aggregation network. The method enables the first efficient transfer of representations from multiple teacher backbones to a single student model: the upper tier mitigates label sparsity via video-level knowledge distillation, while the lower tier fuses multi-source features and suppresses background interference through decoupled cross-attention. Evaluated on UCF-Crime, ShanghaiTech, and XD-Violence, the approach achieves absolute AUC improvements of 1.36%, 0.78%, and 7.02%, respectively—setting new state-of-the-art performance. It significantly enhances anomaly localization accuracy and generalization capability under few-shot, weakly supervised settings.

Technology Category

• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Computer Vision: Video Understanding & Activity Analysis
• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Video anomaly detection aims to develop automated models capable of identifying abnormal events in surveillance videos. The benchmark setup for this task is extremely challenging due to: i) the limited size of the training sets, ii) weak supervision provided in terms of video-level labels, and iii) intrinsic class imbalance induced by the scarcity of abnormal events. In this work, we show that distilling knowledge from aggregated representations of multiple backbones into a single-backbone Student model achieves state-of-the-art performance. In particular, we develop a bi-level distillation approach along with a novel disentangled cross-attention-based feature aggregation network. Our proposed approach, DAKD (Distilling Aggregated Knowledge with Disentangled Attention), demonstrates superior performance compared to existing methods across multiple benchmark datasets. Notably, we achieve significant improvements of 1.36%, 0.78%, and 7.02% on the UCF-Crime, ShanghaiTech, and XD-Violence datasets, respectively.

Problem

• Video Anomaly Detection
• Limited Training Data
• Rare Abnormal Events

Innovation

• DAKD
• Dual-Layer Learning
• Video Anomaly Detection