Clarification on image quality criteria

[ifwang2001]

Thanks for your great work on MMBench!

I’d like to ask how the image quality subtask (e.g. which image is brighter, sharper, more colorful, higher contrast) is evaluated.

Are the ground-truth answers based on human annotations or objective metrics?

If objective, could you share the definitions or formulas for these four metrics?

This will help us understand and improve our model’s performance on this subtask.

Thanks!