Wasn't there any artifact due to discontinuity at the border of upsampled parameters?

Since the nearest neighbor method is used for the parameter upsampling, it seems like the discontinuity at the border of parameters is inevitable.
However, according to the qualitative results reported in the paper, those 16x16 grid artifacts are not observed.

Wasn't there any worry or concern about this when you design the network?