mlfoundations / visit-bench Goto Github PK

Hi VisIT-Bench team,

Thanks for the great work! We are trying to use the dataset to evaluate our model, and we have two questions:

• it seems 50% of the examples in the dataset doesn't have a reference answer. Do you know why and when we could have full reference answers?
• Will the conditioned image caption be provide sometime in the future?

Thanks in advance!

Can you provide some samples in prediction.csv?

Firstly, I really appreciate your outstanding contribution to the LVLM field. Now we are doing some work on the LVLM model and it's imperative for us to evaluate our model's performance in time. As the performance score in VisIT-Bench is obtained by a GPT-4 evaluator, will you have a plan to release this score calculation code?