How does vid2vid Pose To Body compare to the "Everbody Dance Now"?

NVIDIA / vid2vid

Pytorch implementation of our method for high-resolution (e.g. 2048x1024) photorealistic video-to-video translation.

Other

8.6k stars 1.2k forks source link

How does vid2vid Pose To Body compare to the "Everbody Dance Now"? #38

Closed zoink closed 4 years ago

zoink commented 6 years ago

I know the code for the Berkeley paper hasn't been released, but any observations/comments on differences between the (a) temporal smoothing process and (b) the results would be appreciated. Thanks so much!

petergerten commented 6 years ago

It seems that "Everybody Dance Now" uses a separate network for the face. I would also be interested in further differences. Hopefully they will release the code as well.

fantasy-fish commented 5 years ago

What's more, the Everyboday Dance Now paper only uses two consecutive frames in their temporal consistency loss while vid2vid uses more frames. And vid2vid also has the optical flow loss.