Purdue-M2
commented
1 month ago Hi, thanks for your interest in our work. (1) it would be c23 version. (2) The more manipulation methods would be more beneficial for our method to extract common forgery features. Thus, we considered Faceshifter. Furthermore, in order to evaluate fairness, we need demographic annotations, which we obtained from 'Analyzing Fairness in Deepfake Detection With Massively Annotated Databases', it also included Faceshiter, so we also used it.
Thank you for sharing your interesting work. I have a few questions regarding the dataset that I hope you can help clarify: (1) I couldn't find specific information on the compression type of the FF++ dataset utilized in your experiments. As there are three types of compression available for FF++, could you please confirm if all experiments were conducted using FF++ c40? (2) I noticed that Faceshifter was employed in FF++, whereas typically, FF++ contains only four types of forgeries. Could you please explain the reason for including Faceshifter in your experiments? I appreciate your time and assistance in addressing these queries. Thank you very much, and I look forward to your response.