VLM: The More You Tell it, The Less it Sees.

Defining Anchoring Bias in Visual Language Models (VLMs) & an engineer’s field notes. [~_~] One of the lingering challenges in computer vision, the video search problem, has been around as long as surveillance systems have existed. These systems generate hundreds of hours of video data with barebones analytics. Searching through such recordings is a huge and mostly manual undertaking. As a computer vision engineer, it has always appealed to me to solve it. It was only natural that in the era of VLMs & AI, the video search problem should be revisited with fresh tools.