Low AUC in FF-DF

[WeinanGuan]

Hi! I re-train a new PCL model by following your codes. The training data is only I2G data. However, the AUC of FF-DF is only 0.5. It seems like that the trained model has no ability of generalization. I have checked the log file, which shows that loss_D is 1.1+ and acc_D is 1.0 in the training phase, but loss_D and acc_D in the validation phase are 17+ and 0.5, respectively. I have no idea about the reason of that. Would you like to give me some suggestions about this problem? Maybe the generated I2G data or some possible errors? And would you like to provide your trained weight of PCL? My email is weinan.guan@gmail.com. Thanks you a lot~

[jhaotingc]

Hi,

I didn't get the result shown in the original paper yet by only using I2G images. The original paper did not release their code on generating the I2G images. There must be some features they've kept or some error in my I2G code. This was stated in some previous issues.
Some rather good results are achieved by mixing I2G fake images, "real fake" images (fake data in FF), and real images.

I haven't got time to maintain this project recently. Please feel free to make a pull request or colaborate if you make some breakthrough.

[WeinanGuan]

Got it! Thanks for your response~
I will try to mix FF and I2G data for a new training process~

[CJL0114]

知道了！谢谢你的回复~ 我会尝试混合FF和I2G数据进行新的训练过程~

请问如何对FF数据进行新的训练过程。