CoReVAD: A Contextual Reasoning Framework for Training-Free Video Anomaly Detection

Hyeongmuk Lim, Youngbum Hur

arXiv:2605.23116 · 2026-05-25 공개 · arXiv · PDF

vlm training-free contextual-reasoning video-anomaly-detection xd-violence ucf-crime local-response-cleaning softmax-refinement

Abstract

Existing Video Anomaly Detection (VAD) methods typically rely on task-specific training, leading to strong domain dependency and high training costs. Moreover, most existing methods output only scalar anomaly scores, providing limited insight into why specific events are considered abnormal. Recent advances in Vision-Language Models (VLMs) have enabled both anomaly detection and human-interpretable reasoning. However, many VLM-based approaches still require additional training steps (e.g., instruction tuning or verbalized learning) or external Large Language Models (LLMs), incurring further training costs and inference overhead. To address these challenges, we propose CoReVAD, a contextual reasoning framework for training-free video anomaly detection that operates with a single frozen VLM. CoReVAD directly generates anomaly scores and temporal descriptions from the VLM. To mitigate noise in generative outputs, we introduce a Local Response Cleaning (LRC) module based on local vision-text alignment. Furthermore, global temporal context and progression are incorporated through softmax-based refinement, Gaussian smoothing, and position weighting. Experiments on UCF-Crime and XD-Violence demonstrate that CoReVAD achieves competitive performance among training-free methods while providing reliable and interpretable explanations. Our official code is available at: https://github.com/Muk-00/CoReVAD

한국어 요약

📋 한 줄 요약

**[비지도 비디오 이상 탐지 / VLM]** CoReVAD가 단일 frozen VLM만으로 anomaly score·시간 description 직접 생성, Local Response Cleaning·softmax refinement·Gaussian smoothing·position weighting 결합, UCF-Crime·XD-Violence에서 training-free SOTA급.

🎯 핵심 기여도

💡 핵심 아이디어

VLM의 추론 능력을 비디오 이상 탐지에 활용할 때 추가 학습·외부 LLM 없이도, local alignment 기반 response cleaning과 softmax·smoothing·position weighting의 시간 context 통합만으로 신뢰성 있고 interpretable한 anomaly score를 얻을 수 있다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: Training-free VAD의 explainability·실용성 동시 달성, VLM 기반 시간 추론 일반 패턴, LRC·smoothing·weighting의 modular 조합 레시피. **한계**: VLM 자체 성능에 종속, frozen 사용으로 특정 도메인 anomaly에 fine-grained 적응 어려움, UCF-Crime·XD-Violence 외 long-form video 일반화 검증 필요.

🚀 실용적 활용