Does not work on fake videos.

[mirfan899]

I'm trying to detect fake video and it does not work. Here is the video I'm using for fake detection https://youtu.be/r1jng79a5xc Here is output

Dealing with video  Jennifer Lawrence-Buscemi on her favorite housewives [Deepfake].mp4
Face extraction warning :  0 - found face in full frame [(173, 797, 334, 636)]
Face extraction warning :  1444 - found face in full frame [(156, 690, 316, 529)]
Face extraction report of not_found : 2
Face extraction report of no_face : 0
Predicting  Jennifer Lawrence-Buscemi on her favorite housewives [Deepfake].mp4
2020-03-13 12:18:34.535467: W tensorflow/core/framework/cpu_allocator_impl.cc:81] Allocation of 86507520 exceeds 10% of system memory.
2020-03-13 12:18:34.554189: W tensorflow/core/framework/cpu_allocator_impl.cc:81] Allocation of 86507520 exceeds 10% of system memory.
2020-03-13 12:18:34.586315: W tensorflow/core/framework/cpu_allocator_impl.cc:81] Allocation of 86507520 exceeds 10% of system memory.
`Jennifer Lawrence-Buscemi on her favorite housewives [Deepfake]` video class prediction : 1.0

[AnniRen]

Me neither

[mirfan899]

any update?

[pcmg12138]

me neither, did you solved?

[hzhumsc]

Hi, guys! Did you find the reason why it returns the trivial value for manipulated videos?

[hzhumsc]

I think it's a matter of the weights. The author's weights are trained w.r.t. the specific manipulation.