About acc

[peylnog]

Dear Dr.andrew
I'm sorry to bother you. I'm quite interested in following your work after reading your paper, "CNN-generated pictures are surprisingly easy to recognize... for now." But I ran into some issues.
1. How accurate is ResNet50's baseline performance without data augment? (In the paper, I just see AP; I don't seeACC.)
2. Do you aware of any CNNdetection follow-up work? Is this field's SOTA CNNdetection?
I will be appreciated for your reply, thank you very much!
Best wishes

[PeterWang512]

1. I'm not sure whether I have the accuracy for the model without data augmentation at hand. However, I believe it will perform roughly at chance performance for unseen methods.
2. Yes, one of our coauthors has a follow-up paper on this: https://arxiv.org/abs/2310.15150

I hope this helps!