the video generated by image2video is shaking .
rexxar-liang opened this issue · 10 comments
Hello,
Our repository is Amazing！We can swap a whole video in 1 min!
When we use face_swap_image2video.cpp, we have a problem and hope your optimize opinion.
We found that when the video is playing, the face will shake somehow. Is there any good way to remove this sway?
Many thanks!
This repository is mainly for image-to-image face swapping.
You should wait for our follow up paper's code which will support video face swapping: https://nirkin.com/fsgan
Great! Hope to try the video face swapping when the code is ready.
Thanks!
@YuvalNirkin That is great. When will the code be ready? I can't wait to try it out.
@YuvalNirkin Wow! fsgan is great, img blending is difficult and complex. I think that is a great tech.
@YuvalNirkin will FSGAN code be released before or after iccv 2019?
Hello,
Our repository is Amazing！We can swap a whole video in 1 min!
When we use face_swap_image2video.cpp, we have a problem and hope your optimize opinion.
We found that when the video is playing, the face will shake somehow. Is there any good way to remove this sway?Many thanks!
Can you share this file?
fatal error: face_swap/face_swap_export.h: No such file or directory
@YuvalNirkin Hi any plan for fsgan ?
@YuvalNirkin
Hi, there is a new faces waping paper come out, ”FaceShifter: Towards High Fidelity And Occlusion Aware Face Swapping“[https://arxiv.org/pdf/1912.13457.pdf].
The faces show in the paper looks good, but I think for video scenario， this new paper will not work fine(may shaking)。
FSGAN support video scenario which is much more attracting me.
Any new plan for open source? I eager to try it out.
Thanks.
@rexxar-liang Yes, I also suspect that similar to other attribute-based algorithms such as RSGAN and FSNet, this method won't be able to generate temporally consistent results suitable for video face swapping. It's a very interesting work nonetheless.
The FSGAN code is ready but unfortunately we have some legal issues concerning the release.
Sorry for the inconvenience,
Yuval
@rexxar-liang Yes, I also suspect that similar to other attribute-based algorithms such as RSGAN and FSNet, this method won't be able to generate temporally consistent results suitable for video face swapping. It's a very interesting work nonetheless.
The FSGAN code is ready but unfortunately we have some legal issues concerning the release.
Sorry for the inconvenience,
Yuval
So, do you mean that the code will not be released?