search

Dongping-Chen / MLLM-Judge

[ICML 2024 Oral] Official code repository for MLLM-as-a-Judge.
https://mllm-judge.github.io/
46 stars 4 forks source link

About the dataset #5

Open tyxiong23 opened 1 month ago

tyxiong23 commented 1 month ago

Hi authors, in the paper you mentioned that you included image-question pairs from CC3M. But I didn't find them in Dataset/Benchmark/score.jsonl. Instead, I found data from llava_bench. Are they the same?

Dongping-Chen commented 1 month ago

Yes, they are the same. We first chose llava_bench (-wild), but its size does not meet our expectations. Therefore, we use CC3M.