O-VAD Framework for Object-Centric Industrial Video Anomaly Detection
July 19, 2026
O-VAD is a training-free agentic framework designed for industrial settings with strict physics and procedural constraints. It tracks spatial-temporal dynamics and object state evolution to detect anomalies without requiring domain-specific training.
HOW THIS AFFECTS YOU
●
builderYou can deploy training-free anomaly detection in manufacturing environments with complex object transformations.
●
researcherYou can leverage object-centric tracking to improve VLM reasoning in physics-constrained domains.