QVAD uses iterative LLM-guided query refinement to unlock state-of-the-art video anomaly detection from lightweight VLMs on UCF-Crime, XD-Violence, and UBNormal with low memory and high speed.
3.Model Locks:Thread-safe locks (vlm lock,llm lock) serialize access to the VLM and LLM during generation
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
QVAD: A Question-Centric Agentic Framework for Efficient and Training-Free Video Anomaly Detection
QVAD uses iterative LLM-guided query refinement to unlock state-of-the-art video anomaly detection from lightweight VLMs on UCF-Crime, XD-Violence, and UBNormal with low memory and high speed.