Closed KevinLight831 closed 1 week ago
Thanks for the reminder. We didn't release them since one can sample benign prompts from the original LAION dataset to achieve a similar result. Now we've put them under Datasets
directory.
Thank you for the reply. However, the column "evaluation_seed" of this benign prompt csv file is missing. is the same specific "evaluation_seed" used?
Yes, the same evaluation seeds are applied to all methods. And they are in the last (third) column of the .csv
file.
Did the author forget to release the subset with 500 benign prompts from LAION for the out-of-prompt CLIP score?