Question about paper

[freshman2001]

https://ojs.aaai.org/index.php/AAAI/article/view/19990

Hi! Thank you for your amazing work.
In this paper FrePGAN, in Table3, your baseline method's performance be writed as 50.4, in Acc, which is almost equal with random classification in binary classification task.

Of course, it's possible that I have misunderstood this result, but I'm curious to know if you have any new insights regarding this result? Looking forward to your response!

[PeterWang512]

Hi, our model generalizes to other methods in terms of the ordering of the real/fake prediction, which is shown in terms of high average precision (AP). However, it was still challenging for our method to find the right cutoff for real/fake prediction scores without seeing the unknown method. I hope this helps!