Clarification on the Use of CelebA-HQ Dataset in Your Paper

[parham1998]

Dear Yuhao,

I have a question regarding the experimental setup involving the CelebA-HQ dataset, as there appear to be some differences between your protocol and the experimental protocols used in other studies, such as AMT-GAN and Clip2Protect.
While it is clearly stated that images in the LADN dataset were divided into four groups, it is not explicitly mentioned whether the same grouping was applied to the CelebA-HQ dataset.
I reached out to the author of Clip2Protect for clarification on this point, and they indicated that they use all 1000 images for each target, which differs from your approach.
In your paper, you mentioned "For CelebA-HQ, we select a subset of 1000 images and divide them into four groups, each of which has a target identity.

Thank you for your time and consideration.

[fangxuehouwuming]

Dear Yuhao,

I have a question regarding the experimental setup involving the CelebA-HQ dataset, as there appear to be some differences between your protocol and the experimental protocols used in other studies, such as AMT-GAN and Clip2Protect. While it is clearly stated that images in the LADN dataset were divided into four groups, it is not explicitly mentioned whether the same grouping was applied to the CelebA-HQ dataset. I reached out to the author of Clip2Protect for clarification on this point, and they indicated that they use all 1000 images for each target, which differs from your approach. In your paper, you mentioned "For CelebA-HQ, we select a subset of 1000 images and divide them into four groups, each of which has a target identity.

Thank you for your time and consideration.

Hello, I also found this problem, in addition, I found that both DiffAM and Clip2Protect just directly use the result from AMT-GAN, in AMT-GAN, the author divided 1000 celebahq into four subsets, each pair printed a target face, the same is true for LADN. But I have only found the 4 target faces so far, and I have not found how the author divided the data set and the correspondence between the source and the target. If you find out, I hope you can let me know, thank you very much