How do you obtain high-quality text descriptions corresponding to images?

[shams2023]

This is a great piece of work that has given me some inspiration. Thank you very much for your contributions!
You described a paragraph in the text, as shown in the picture below. What I would like to ask is how you fine tuned BLIP (because it also relates to my captioning operations on the dataset I collected). I am very interested in knowing some details of your fine tuning, and even if it is possible to share this part of the code without damaging your knowledge.
I would greatly appreciate it and look forward to your reply!
Thank you all!