Open huizhg opened 1 month ago
Hello,
Thank you for your interest in our work. We'll give the example code for the untargeted attack soon. Or you can just change the tar_label from line 223 in ldm/models/diffusion/ddim_adv.py to the ground truth label and use negative sampling. However, we do not recommend directly using the untargeted attack version as they can harm the generation quality.
Thanks for the quick reply. I just want to check the difference between the samples generated by the two different sampling methods.
Hi,
Thank you for sharing the code for your great work. I am curious how you implemented the untargeted attack you mentioned in your paper. There's no explanation in the paper either.