Open-Vocabulary Video Anomaly Detection

TL;DR

This work introduces open-vocabulary video anomaly detection (OVVAD), which jointly detects and categorizes seen and unseen anomalies using a two-task framework. It leverages pre-trained vision-language models through a nearly weight-free Temporal Adapter (TA), a Semantic Knowledge Injection (SKI) module, and a Novel Anomaly Synthesis (NAS) module to exploit cross-modal priors and generate pseudo novel anomalies for training. The approach is trained with a two-stage objective that combines detection and categorization losses, aided by pseudo anomalies to improve generalization to novel categories. Experiments on three benchmarks show state-of-the-art OVVAD performance and strong cross-dataset generalization, highlighting the practical impact for informed video surveillance in open-world settings.

Abstract

Video anomaly detection (VAD) with weak supervision has achieved remarkable performance in utilizing video-level labels to discriminate whether a video frame is normal or abnormal. However, current approaches are inherently limited to a closed-set setting and may struggle in open-world applications where there can be anomaly categories in the test data unseen during training. A few recent studies attempt to tackle a more realistic setting, open-set VAD, which aims to detect unseen anomalies given seen anomalies and normal videos. However, such a setting focuses on predicting frame anomaly scores, having no ability to recognize the specific categories of anomalies, despite the fact that this ability is essential for building more informed video surveillance systems. This paper takes a step further and explores open-vocabulary video anomaly detection (OVVAD), in which we aim to leverage pre-trained large models to detect and categorize seen and unseen anomalies. To this end, we propose a model that decouples OVVAD into two mutually complementary tasks -- class-agnostic detection and class-specific classification -- and jointly optimizes both tasks. Particularly, we devise a semantic knowledge injection module to introduce semantic knowledge from large language models for the detection task, and design a novel anomaly synthesis module to generate pseudo unseen anomaly videos with the help of large vision generation models for the classification task. These semantic knowledge and synthesis anomalies substantially extend our model's capability in detecting and categorizing a variety of seen and unseen anomalies. Extensive experiments on three widely-used benchmarks demonstrate our model achieves state-of-the-art performance on OVVAD task.

Figures (4)

• Figure 1: Comparison of different VAD tasks.
• Figure 2: Overview of our proposed framework.
• Figure 3: Qualitative results of our model on testing videos. Colored window denotes ground-truth anomalous region.
• Figure 4: Confusion matrices of anomaly categorization.