GFPGAN icon indicating copy to clipboard operation
GFPGAN copied to clipboard

unexpected glasses on restored images

Open chky opened this issue 2 years ago • 0 comments

great job!

The performance is quite good on many blurred face images, however, when I test the inference code on other datasets, some unexpected glasses emerge on restored image. Is there any parameters I can adjust to avoid this situation or I have to retrain the model on my dataset? I have loaded model v1.3 and some examples are as follows, 1 2 3

Thank you.

chky avatar Aug 24 '22 08:08 chky