Closed Caroline-6 closed 4 months ago
Hi, thank you for your feedback. The relate code you're asking is the file datasets/gpt4v_prompt_gen.py
. As described in the paper, we guide GPT4V with a prompt (which you can find both in the file and in the Appendix), the original image, and an image with all the bounding boxes of each object (code at datasets/mark_img.py
).
oh,ok. I see. thank you!
Hi, this a great work. I am very interested in the part of using the GPT-4V to generate hard negative captions. Would you like to release the related code or describe more details about the process?