Beyond Hazard Resemblance: Contrastive Event Adjudication for Training-Free Video Anomaly Detection

Forum topic · 小凯 · 2026-08-11

Summary

This paper introduces CEAVAD (Contrastive Event Adjudication for training-free Video Anomaly Detection), a new approach to video anomaly detection (VAD) that identifies and temporally localizes abnormal events without requiring in-domain training data. While supervised VAD methods learn anomaly decision boundaries from target-domain annotations, they demand substantial in-domain data. Existing training-free methods rely on pretrained models' semantic knowledge and reasoning, but richer anomaly descriptions only capture hazard resemblance without resolving actual abnormality. CEAVAD 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. It first constructs hazard-benign event contrasts using public safety knowledge, pairing each hazardous mechanism with a generic normal description and a mechanism-specific benign counterpart. It then determines whether a target interval better supports the hazardous or benign explanation, producing a revisable contrastive boundary proposal, and finally adjudicates between competing explanations to yield both temporal localization and evidence-based explanations. Experiments on three widely used VAD benchmarks show state-of-the-art performance in the training-free paradigm.

Paper Overview

Research Area: Computer Vision Authors: Wenti Yin, Xiang Wang, Huaxin Zhang Published: 2026-08-11 arXiv: 2508.03800

Summary

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.

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 the interaction between competing explanations and video evidence.

Key Points

Hazard-benign event contrasts: CEAVAD first uses public safety knowledge to construct hazard-benign event contrasts, pairing each hazardous mechanism with a generic normal description and a mechanism-specific benign counterpart.
Contrastive boundary proposals: It determines whether a target interval better supports the hazardous explanation or its benign competing explanation, producing a revisable contrastive boundary proposal.
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 explanations.
Results: Experiments on three widely used VAD benchmarks demonstrate that CEAVAD achieves state-of-the-art performance under the training-free paradigm.

Beyond Hazard Resemblance: Contrastive Event Adjudication for Training-Free Video Anomaly Detection

Summary

Paper Overview

Summary

Key Points

Links

Tags