Datasets:
Inquiries about real-world data
Hi, thanks for your amazing work for the image generation community!
In your paper, you mentioned that you incorporated both real and synthetic images for the MultiBanana benchmark. But as I looked up the dataset on hugging face, I couldn't find a clear distinction between real and synthetic image pairs. The image pairs are mainly categorized by their reference and combination types. So I'm wondering whether there is a convenient way to quickly identify the real images.
Thank you again for your great work! And I really look forward to your help in solving the problem about real-world images!
Thank you for your interest in MultiBanana!
You're right that the current structure doesn't explicitly label real vs. synthetic pairs. We are working on adding these labels now and will release an updated version in a week.
This update will allow you to quickly identify whether the image sources are real or synthetic. Thanks for your patience!
Thanks a lot for your timely reply! Really looking forward to your update!
Thank you for your patience!
We have now added from_where.csv, which indicates the source of each image (real or synthetic).
Again, we appreciate your patience and interest in our work!