Concerns about the time result in Table.2

[Bob-Yeah]

I'm wondering how you measure the time for EDSR and EDSR-APE?
It seems fair to me if full image forward is for EDSR and patched forward is for EDSR-APE and the time for EDSR-APE is added up with crop and combine operations. Because I'm concerned that a patched forward mode is causing extra overhead in PyTorch computation.
But in your code, if I haven't misunderstood, patched forward is both used for measuring time of EDSR and EDSR-APE.

Please check Line 123 and Line 160 in template.py.