Can not achieve good results on NT/F2F/FS while training on raw

[Nerdary]

I follow the settings in 4.1. Implementation Details and try to train a detector on FF+ raw with only real faces and self-blend fake faces. However when testing on FF+ raw it can only perform good on Deepfakes(138/140 correct), the results on NeuralTextures(≈50% acc), Face2Face(≈60% acc) and Faceswape(≈60% acc). I also tried the released pretrain weights to test on FF++ raw and everything goes well with the data in paper. Such problem makes me really confused, have anyone meet similar issue?

[WangLedi]

The same goes for me. The model trained on raw only had an accuracy of 82% when tested on the CDF data set, but I didn't change anything except the data set path. Did you reproduce it later?

[molokanov50]

I trained the model on FF++ c23 and got 82% auc on CDF

[Elahe-sd]

The same goes for me. The model trained on raw only had an accuracy of 82% when tested on the CDF data set, but I didn't change anything except the data set path. Did you reproduce it later?

Hey I also have the same problem. How did you manage to solve it? Would you help me, please?

[Elahe-sd]

I trained the model on FF++ c23 and got 82% auc on CDF
Hey, Could you achieve the reported result?

[gbljdgb]

I trained the model on FF++ c23 and got 82% auc on CDF

Hello, I would like to ask whether the result of your test is frame level or video level ?