Clarification on Fine-tuning of LLaVA-Tri and Baselines on VQA Datasets

[xw-wj]

Hi, thanks for releasing this great work!

I have a question about the VQA experiments. In the paper, it is mentioned:

"The model is fine-tuned for three epochs on each of the three VQA datasets and evaluated accordingly."

I would like to clarify:

• Does this 3-epoch fine-tuning apply only to LLaVA-Tri, or did you also fine-tune the baseline models (e.g., LLaVA-Med, BLIP-2, PMC-VQA, etc.) for three epochs on the same datasets?
• Or are the baseline results directly taken from the original papers without additional fine-tuning?

This clarification would really help me better understand the fairness of the comparison and how to reproduce the results.

Thanks a lot!