tensorflow-SRGAN
tensorflow-SRGAN copied to clipboard
What is motivation of using large kernel size at input and output?
What is motivation of using large kernel size at input and output?
https://github.com/trevor-m/tensorflow-SRGAN/blob/master/srgan.py#L43
https://github.com/trevor-m/tensorflow-SRGAN/blob/master/srgan.py#L59
As far as I know, the paper doesn’t give any justification for the larger kernel sizes. I would be surprised if it actually helped much. If you decide to experiment would be interested in hearing the results!