Abstract: The advancement of Large Language Models (LLMs) with vision capabilities in recent years has elevated video analytics applications to new heights. To address the limited computing and ...