luogen1996 / LaVIN

[NeurIPS 2023] Official implementations of "Cheap and Quick: Efficient Vision-Language Instruction Tuning for Large Language Models"
502 stars 36 forks source link

GPT-4 evaluation #6

Closed WizardMx closed 1 year ago

WizardMx commented 1 year ago

How do you evaluate your response via GPT-4? Feed him the image and its response, then ask him to give a score out of 10?

luogen1996 commented 1 year ago

We use the GPT-assisted evaluation from llava, which feeds GPT4 with the object categories, boxes and image captions, and asks it to evaluate the given responses. See the details in LLaVA.