Vad-R1: Towards Video Anomaly Reasoning via Perception-to-Cognition Chain-of-Thought

Wei Wang (Beijing University of Post and Telecommunication) · Chao Huang (Sun Yat-Sen University) · Xiaochun Cao (SUN YAT-SEN UNIVERSITY) · Li Shen (Sun Yat-Sen University) · Benfeng Wang (SUN YAT-SEN UNIVERSITY) · Jie Wen (Harbin Institute of Technology, Shenzhen) · Chengliang Liu (Shenzhen University)

anomaly recognitionava-grpodeep reasoning end-to-end framework experimental resultslimited annotationsmultimodal large language modelsperception-to-cognition chain-of-thoughtproprietary models reinforcement learning algorithmself-verification mechanismstructured reasoning superior performancevad-reasoning datasetvideo anomaly detectionvideo anomaly reasoning

Recent advancements in reasoning capability of Multimodal Large Language Models (MLLMs) demonstrate its effectiveness in tackling complex visual tasks. However, existing MLLM-based Video Anomaly Detection (VAD) methods remain limited to shallow anomaly descriptions without deep reasoning. In this paper, we propose a new task named Video Anomaly Reasoning (VAR), which aims to enable deep analysis and understanding of anomalies in the video by requiring MLLMs to think explicitly before answering. To this end, we propose Vad-R1, an end-to-end MLLM-based framework for VAR. Specifically, we design a Perception-to-Cognition Chain-of-Thought (P2C-CoT) that simulates the human process of recognizing anomalies, guiding the MLLMs to reason about anomalies step-by-step. Based on the structured P2C-CoT, we construct Vad-Reasoning, a dedicated dataset for VAR. Furthermore, we propose an improved reinforcement learning algorithm AVA-GRPO, which explicitly incentivizes the anomaly reasoning capability of MLLMs through a self-verification mechanism with limited annotations. Experimental results demonstrate that Vad-R1 achieves superior performance, outperforming both open-source and proprietary models on VAD and VAR tasks.