yinboc / liif

Learning Continuous Image Representation with Local Implicit Image Function, in CVPR 2021 (Oral)
https://yinboc.github.io/liif/
BSD 3-Clause "New" or "Revised" License
1.27k stars 145 forks source link

Not the result I expected #16

Closed yaseryacoob closed 3 years ago

yaseryacoob commented 3 years ago

In trying super resolution on faces fr 0 om CELEBHQ, I get poor results (using both pretrained models). See the 1Kx1K output below. output

xinzi2018 commented 3 years ago

i have the same problem!

yinboc commented 3 years ago

It should work if you follow the instruction and generate the celebAHQ png files, the pretrained model may work better. E.g. try this png image: 128

I hypothesize this is because the potential noise in your jpg file makes the input out of the training domain, as the model is trained for png files without much noise.