wkim97 / fsr Goto Github PK

Thanks for your great work.

I found all the biases are set to False in your Separation and Recalibration network.
Is there any reason you did?

If I take the four outputs out, r_outputs, nr_outputs, rec_outputs from the back of your model and give them in order to just use out, and then do the adversarial training separately afterward, will the results have a big impact?

thanks!~
1.but i want to know why my GPU just run with no more than 10% used?
2.is there any way to improve this as it is too slow?
maybe the Q above is too basic as i am a beginner ...
be grateful and look forward to your Answer!~