About shuffling operation in RCAN training

[ZahraFan]

Hello, I came across your paper RealSR and I found it to be very impressive. I became curious about the description "To keep the network structures of SRResNet and RCAN unchanged, the input images were shuffled with factor for the three scaling factors respectively".

Do you mean you shuffle the hw image into 16h/4w/4 and get 16h*w output, then take the mean as the HR result?

Could you please provide more details about this process and how it contributed to the final high-resolution output? Thank you.