How to identify specific roles by fine tune BLIP Image Captioning with custom dataset

[swe1210]

Hi, thanks for your impressive work.
I have a question:
How to identify specific roles by fine tune BLIP Image Captioning with custom dataset
My dataset:each of role has this role picture
I try to give each picture a unique caption, either simply giving the character name or the character name and the character's individual appearance features.
However, the effect of this text description is not obvious, and the model cannot recognize the character name.
Are there any recommended alternative text descriptions or other ways to tweak the model?
Thank tou very much!