Inferencing on custom videos

[amil-rp-work]

Hey @AliaksandrSiarohin
Thanks a lot for sharing such awesome work. All the best for the conference!

I had a hard time replicating the work and was wondering if you could share some inputs here

• Parameters for selecting source image compatible with driving video
  • Body Visibility?
  • Initial pose?
  • Gender/Age
• For head re-enactment (similar to First order motion model), is the procedure same as before?

[AliaksandrSiarohin]

Have you tried Ted-YouTube checkpoint usually it performs better. Ideally body should be visible and legs should be cutted similarly like in Readme images. There is no requirement for initial pose and age gender. For head you may specify relative animation mode.

[amil-rp-work]

@AliaksandrSiarohin
So for Ted Youtube, the source, and driving need to be similar to the examples shown in Readme?

And for head re-enactment -

• Which model to use?
• How cropped the inputs should be? Should they only include face?

[AliaksandrSiarohin]

Right they need to be cropped like in Readme.
Vox255.pth, they need to be cropped similar to first order.

[surfingnirvana]

How do i specify the relative animation mode?