TencentARC/VQFR

Resolution is low

leopardhe opened this issue · 4 comments

I am not sure why after running VQFR, the mouth part of the face has lower resolution than the one from GFPGAN. I used the default parameters.
VQFR
image
GFPGAN
image

Any parameter I can change to make the resolution higher?

Hello, may I see the LQ input?

Here is the input
image

VQFR tries to keep the fidelity (identity, like the mouth) of the extracted feature of input. GFP-GAN changes the mouth shape. We are now working on releasing another model in recent weeks, emphasizing more quality than fidelity for real-world applications. Please stay tuned for our project.

ok, sounds great. Looking forward to it.