GFPGAN
GFPGAN copied to clipboard
TencentARC
New argument to force full processing, even when CUDA is not available.
Thanks, this got RealESRGAN to work with my GFPGAn for the first time ever!!!
One thing, for me, at least, as written it will not force ESRGAN, but if you change line 62 to True then it will work. Was that the intent? (Also, I had to get rid of Else BG_Upsampler=none, but I might have an older inference file)
Thanks, this got RealESRGAN to work with my GFPGAn for the first time ever!!!
One thing, for me, at least, as written it will not force ESRGAN, but if you change line 62 to True then it will work. Was that the intent? (Also, I had to get rid of Else BG_Upsampler=none, but I might have an older inference file)
You shouldn't need to make that change on line 62? If you make a call to inference_gfpgan.py and include the --force_esrgan argument, e.g.:
sudo python3 inference_gfpgan.py -i ./inputs --force_esrgan
Then execution will fall through to the else:, and bg_upsampler will be assigned an instance of RealESRGANer.
That makes sense to me now and you are absolutely 100% correct. I was trying to make it default to use ESRGAN all the time without using an argument. That was especially lazy of me since I usually use a .bin and could just have added it there. Sorry for the confusion.
That makes sense to me now and you are absolutely 100% correct. I was trying to make it default to use ESRGAN all the time without using an argument. That was especially lazy of me since I usually use a .bin and could just have added it there. Sorry for the confusion.
No worries! Glad we got it straightened out.
