Rethinking Metrics and Benchmarks of Video Anomaly Detection
This work addresses evaluation challenges in VAD, providing tools to improve model assessment and reduce biases, which is incremental but important for researchers and practitioners in video analysis.
The paper identifies three limitations in current Video Anomaly Detection (VAD) evaluation metrics and benchmarks, such as single annotation bias and lack of early detection rewards, and proposes new metrics like Prob-AUC/AP and LaAP, along with hard normal benchmarks, to address these issues, reporting performance comparisons of ten state-of-the-art methods.
Video Anomaly Detection (VAD), which aims to detect anomalies that deviate from expectation, has attracted increasing attention in recent years. Existing advancements in VAD primarily focus on model architectures and training strategies, while devoting insufficient attention to evaluation metrics and benchmarks. In this paper, we rethink VAD evaluation methods through comprehensive analyses, revealing three critical limitations in current practices: 1) existing metrics are significantly influenced by single annotation bias; 2) current metrics fail to reward early detection of anomalies; 3) available benchmarks lack the capability to evaluate scene overfitting of fully/weakly-supervised algorithms. To address these limitations, we propose three novel evaluation methods: first, we establish probabilistic AUC/AP (Prob-AUC/AP) metrics utlizing multi-round annotations to mitigate single annotation bias; second, we develop a Latency-aware Average Precision (LaAP) metric that rewards early and accurate anomaly detection; and finally, we introduce two hard normal benchmarks (UCF-HN, MSAD-HN) with videos specifically designed to evaluate scene overfitting. We report performance comparisons of ten state-of-the-art VAD approaches using our proposed evaluation methods, providing novel perspectives for future VAD model development. We release our data and code in https://github.com/Kamino666/RethinkingVAD.