Paper Overview
- Field: Computer Vision
- Authors: Wenti Yin, Xiang Wang, Huaxin Zhang
- arXiv: 2508.03800
- Paper: https://arxiv.org/abs/2508.03800
Abstract
Video anomaly detection (VAD) aims to identify and temporally localize abnormal events in videos. Supervised methods learn anomaly decision boundaries from target-domain annotations but require substantial in-domain data. Existing training-free methods leverage the rich semantic knowledge and reasoning capabilities of pretrained models to interpret visual content, yet these capabilities do not directly define an anomaly decision criterion: richer anomaly descriptions better capture hazard resemblance without resolving abnormality.
Method: CEAVAD
To address this, the authors propose Contrastive Event Adjudication for training-free Video Anomaly Detection (CEAVAD), which shifts the unit of inference from isolated anomaly concepts to falsifiable event hypotheses and establishes an inference-time explanatory boundary through interaction between competing explanations and video evidence.
The approach works in three stages:
1. Hazard-benign event contrast construction: Using public safety knowledge, CEAVAD pairs each hazard mechanism with both a generic normal description and a mechanism-specific benign counterpart.
2. Contrastive boundary proposal: It determines whether the target interval better supports the hazardous explanation or its benign competing explanation, yielding a correctable contrastive boundary proposal for the target.
3. Adjudication: CEAVAD adjudicates between competing explanations to determine whether the hazard hypothesis withstands the video evidence, supporting both temporally localized anomaly detection and evidence-based explanation.
Results
Experiments on three widely used VAD benchmarks demonstrate that CEAVAD achieves state-of-the-art performance in the training-free paradigm.