VADTree: Explainable Training-Free Video Anomaly Detection via Hierarchical Granularity-Aware Tree

TL;DR

VADTree introduces a hierarchical, training-free approach for video anomaly detection that adaptively samples videos and leverages large pre-trained models for explainable and efficient anomaly detection across various datasets.

Contribution

The paper proposes VADTree, a novel hierarchical structure that improves training-free video anomaly detection by adaptive sampling and multi-model integration, enhancing accuracy and interpretability.

Findings

Achieves state-of-the-art performance on three datasets.

Reduces the number of sampled video segments significantly.

Provides explainable anomaly detection results.

Abstract

Video anomaly detection (VAD) focuses on identifying anomalies in videos. Supervised methods demand substantial in-domain training data and fail to deliver clear explanations for anomalies. In contrast, training-free methods leverage the knowledge reserves and language interactivity of large pre-trained models to detect anomalies. However, the current fixed-length temporal window sampling approaches struggle to accurately capture anomalies with varying temporal spans. Therefore, we propose VADTree that utilizes a Hierarchical Granularityaware Tree (HGTree) structure for flexible sampling in VAD. VADTree leverages the knowledge embedded in a pre-trained Generic Event Boundary Detection (GEBD) model to characterize potential anomaly event boundaries. Specifically, VADTree decomposes the video into generic event nodes based on boundary confidence, and performs adaptive coarse-fine hierarchical structuring and redundancy removal to construct the HGTree. Then, the multi-dimensional priors are injected into the visual language models (VLMs) to enhance the node-wise anomaly perception, and anomaly reasoning for generic event nodes is achieved via large language models (LLMs). Finally, an inter-cluster node correlation method is used to integrate the multi-granularity anomaly scores. Extensive experiments on three challenging datasets demonstrate that VADTree achieves state-of-the-art performance in training-free settings while drastically reducing the number of sampled video segments. The code will be available at https://github.com/wenlongli10/VADTree.