Request for Evaluation Details and Code for Multiple Datasets

[YuHengsss]

Thank you for your excellent work on this paper. I'm currently trying to reproduce your reported results, but I've encountered some challenges with the evaluation process.

Missing Evaluation Code

I noticed that the evaluation code for several datasets is not provided in the repository:

• VisWiz
• TextVQA
• MMMU
• MME

Specific Information Needed

Could you please provide:

1. Hyperparameters used for each dataset:

   • CLIP layer index
   • Kernel size
   • Enhancement coefficient (enhance_coe)
   • Grayscale settings

2. Evaluation methodology details:

   • I see that for these datasets, performance is evaluated "using matching accuracy with the ground truth response" as mentioned in the paper
   • Could you share the specific implementation for calculating this accuracy?

3. Evaluation code:

   • The actual code used to evaluate model outputs against ground truth for these datasets

This information would be extremely helpful for ensuring accurate reproduction of your results.

Thank you for your assistance!