English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original
智柴网 中文原文

CEAVAD: Contrastive Event Adjudication for Training-Free Video Anomaly Detection

Forum topic · 小凯 · 2026-08-11

Summary

This paper introduces CEAVAD, a training-free framework for video anomaly detection (VAD) that identifies and temporally localizes abnormal events without relying on in-domain annotations. The authors argue that prior training-free methods leverage pretrained models' semantic knowledge for visual interpretation but fail to establish a true anomaly decision criterion, since richer descriptions only capture hazard resemblance rather than abnormality itself. CEAVAD reframes inference around falsifiable event hypotheses, using public-safety knowledge to construct hazard-vs-benign event contrasts, where each hazard mechanism is paired with a generic normal description and a mechanism-specific benign counterpart. It first proposes a revisable contrastive boundary for a target interval, then adjudicates between competing hazard and benign explanations against the video evidence. The approach yields both temporally localized anomaly decisions and evidence-based explanations. Experiments on three widely used VAD benchmarks show that CEAVAD achieves state-of-the-art performance within the training-free paradigm.

Paper Overview

  • Field: Computer Vision (CV)
  • Authors: Wenti Yin, Xiang Wang, Huaxin Zhang
  • Release Date: 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 this end, 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.

Method

1. Hazard–Benign Event Contrast Construction — Using public-safety knowledge, CEAVAD pairs each hazard mechanism with a generic normal description and a mechanism-specific benign counterpart. 2. Contrastive Boundary Proposal — It determines whether a target interval better supports a hazard interpretation or its competing benign interpretation, producing a revisable contrastive boundary proposal. 3. Event Adjudication — CEAVAD adjudicates between competing explanations to test whether the hazard hypothesis withstands the video evidence, enabling 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 under the training-free paradigm.

Original 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. To this end, we 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 thr...

Tags

#video-anomaly-detection#training-free#contrastive-learning#event-adjudication#computer-vision#vad-benchmark#arxiv#multimodal-reasoning

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178633352