explanation of the split jsons about faceforensics HOT 7 OPEN

Re: negative instances, I suspect it works like this: If train.json contains ["123", "456"], this means the training set includes the original videos 123 and 456, and the manipulated videos 123_456 and 456_123. This is just a guess, though. It would be nice to see this spelled out explicitly.

from faceforensics.

what about original videos? (for negative instances for deepfake detection) what is the split for that?

from faceforensics.

yes I thought of that. this means that it is also works the other way round, since there are 140 numbers in the test split so for each pair, the first number would be the source for the second and vice versa.

Indeed I found these correspond to videos in the data. For example the first pair of test.json is 953, 974. If you look under DeepFakes directory you will find two videos 953_974.mp4 and 974_953.mp4. The same goes for the other subsets.

from faceforensics.

same question

from faceforensics.

I found something. The pairs mean the source and target corresponding in the manipulated sequences.

For example, the pair with '210_219' folder means the background video is from 210 while the face is from 219.

from faceforensics.

@ondyari can u explain how to load these json and make them dataloaders?

from faceforensics.

@ondyari can u explain how to load these json and make them dataloaders?
I have the same question ,do you already know how to get it ?

from faceforensics.