how to use GPT4V for visual grounding tasks

[xiang-xiang-zhu]

I see in the paper you guys have a picture comparing the effects of gpt4v with the query and results. I would like to know how to accomplish such an effect using gpt4v, thanks!