What is motivation of using large kernel size at input and output?

[mrgloom]

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

[trevor-m]

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!