MRzzm / DINet

The source code of "DINet: deformation inpainting network for realistic face visually dubbing on high resolution video."
895 stars 167 forks source link

Openface problem "detected landmarks" #80

Open NickVitali opened 6 months ago

NickVitali commented 6 months ago

File "inference.py", line 59, in raise ('video frames are misaligned with detected landmarks')

how can I fix this?

Inferencer commented 6 months ago

answered here https://github.com/MRzzm/DINet/issues/76#issuecomment-1854432204

NickVitali commented 6 months ago

Thanks so much mate. Now it works. I have another question, what can I do to improve quality?

Inferencer commented 6 months ago

Thanks so much mate. Now it works. I have another question, what can I do to improve quality?

You can use a person without facial hair, Use a American male voice, You can use a person that's closer to the camera, Make sure the person is fully facing the camera like how you would expect a news reader to look, You can remove the mask/ bounding box in post processing using a video editor.

Inferencer commented 6 months ago

Should end up looking something like this https://streamable.com/s6wgqr

NickVitali commented 6 months ago

How do I do with video editor? Any tutorial?

Inferencer commented 6 months ago

How do I do with video editor? Any tutorial?

https://streamable.com/rme7wf It's a bad tut as I did it in a rush ages ago

There's also other methods to get better quality such as changing the code to select the best reference frames etc