Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Questions about the T2I-CompBench results #58

Open
SparklingH opened this issue Oct 28, 2024 · 0 comments
Open

Questions about the T2I-CompBench results #58

SparklingH opened this issue Oct 28, 2024 · 0 comments

Comments

@SparklingH
Copy link

Thanks for your excellent work!

I tried to evaluate RPG on the T2I-CompBench but got different results than those in Table 1. So, I want to ask for the details about the evaluation of RPG on the T2I-CompBench.

  1. About the MLLMs: GPT-4o or other MLLMs.
  2. About the Base Ratio: As you said in your paper, the Base Ratio is important in RPG which affects the generation results. So I want to know the value of the Base Ratio you use to evaluate.
  3. About the results: I use RPG to generate one image for each prompt in T2I-CompBench with GPT-4o-mini (300 images for each metric). However, the results are quite different from those in Table 1. Some results are as follows:
  • color -- 0.4917, shape -- 0.3864, texture -- 0.4573, spatial -- 0.1042, non-spatial -- 0.2877, complex -- 0.2997 (Base Ratio = 0.2)
  • color -- 0.4464, shape -- 0.3863, texture -- 0.4289, spatial -- 0.1132, non-spatial -- 0.2950, complex -- 0.3012 (Base Ratio = 0.5)
    I also evaluate SD v2 on T2I-CompBench and the difference between 300 images and 3000 images is around 0.05. So I don't think it is caused by the number of generated images. Moreover, I use GPT-4o as MLLMs and the results are similar to those using GPT-4o-mini.
  1. About the code: When I use RPG on the T2I-CompBench, some bugs also happen.
  • When the 'Final split ratio' is '1': An error occurred: local variable 'split_ratio2' referenced before assignment.

Hope to see your reply.

Best regards.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant