‘GPT-4o’

VANE-Bench: Video Anomaly Evaluation Benchmark for Conversational LMMs

VANE-Bench is a benchmark designed to assess Video-LMMs’ ability to detect and localize anomalies in videos, revealing their limitations in subtle anomaly detection through a visual question-answering challenge.