search

shaoanlu / faceswap-GAN

A denoising autoencoder + adversarial losses and attention mechanisms for face swapping.
3.36k stars 844 forks source link

Your take on the ReCycle-GAN #118

Open Neltherion opened 5 years ago

Neltherion commented 5 years ago

I know this is not an issue but I wanted to know your take on the ReCycle-GAN approach for deepfakes. Could it be better than the current approaches or does it have ups and downs which makes it useless for deepfakes?

Here's a link to a YouTube video and here's the paper's link.

iperov commented 5 years ago

I am trying to implement RecycleGAN any advice welcome https://github.com/iperov/DeepFaceLab/issues/82

iperov commented 5 years ago

nothing succesful with RecycleGAN .

look at my SAE model, it morphs face and can reconstruct occlusions without any masks and GAN https://www.youtube.com/watch?v=xi1Qme-uB2Y

iperov commented 5 years ago

no gans, no masks python_2019-01-17_19-40-49

Neltherion commented 5 years ago

@iperov Thanks for the new information. Then let me ask what your take is on the InstaGAN Instance-aware Image-to-Image Translation approach?

iperov commented 5 years ago

@Neltherion I have no success with GANs at all

claudetheboof commented 5 years ago

cant we just use FANSEG to mask the part we need for training and apply the model on the editted mask?