iamwangyabin
Question about the OpenSDID Dataset
Thanks to the authors for a really forward thinking design for the OpenSDID dataset.
It seems the dataset doesn't include the originals for the subset of manipulated (or "partials" in the HF dataset card), or the text prompts that were used to drive generation. Am I missing something where this information is currently available? If not, are there any thoughts to make it available?
In other words, for a given filename in the dataset (e.g., partial/flux/fake/000201886.jpg), with an image and mask pair (and label), is it possible to find out what the text prompt used was and what the original image looked like?
Thank you.
Actually, all the data is available on Hugging Face, but I haven't had the time to clean and organize it recently. I'll likely get it all sorted out after I graduate. However, if you're in a real hurry, you can access all the raw data at this link and process it yourself (almost 2 TB raw data... most of them are useless files, but you can find what you need according the file names). But if you can wait a few weeks, I should be able to provide you with the clean data.
