HenryKang1

Did you figure it out? I am also wondering about it.

> HenryKang1, > > I pivoted my application to be mobile web-based, this has allowed me to utilize the phone's accelerometer with JavaScript instead of relying on data from the...

It has less artifact based on my test. As far as I know, author does not share the training strategy anymore.

I am not the author however what I would like to recommend the Apply the SR algorithm for this. The input video is cropped face basically. After that blend with...

> He's referring to a super resolution model such as codeformer or gfpgan which would upscale the face at some cost to skin fidelity although I don't see one implemented...