[HELP] SadTalker inference.py. Cannot detect face in large image.

[Zeelyne]

Hi guys,

Object:
SadTalker inference.py. Failure to detect face in large image.

I cannot detect the face in a high-resolution image (large size). In my case the image is 4233*4924. It is a clear front portrait of the head part of Mona Lisa. So I suppose the failure is not related to the face too small in the image.

Description of the bug:

To reproduce:

1. put an image of resolution higher than 4000*4000 and any audio file
2. any setting (face model resolution, preprocess mode, still mode and batch size)

Other remarks:

1. With images of size no too far away from 512, it works perfectly well for me.
2. Assuming it is the render part of the face too small for my image of large size, I tried to add – size 2000 at the end of my command.

However, another bug appears:

I have been blocked by this for a couple of days. Any suggestions for dealing with the images whose size is very different from 512 or 256?
Thank you very much!