about implementation details

[fido20160817]

Hi, I am not clear about some details during implementations: 1024 locations are sampled from style and content images respectively (so is the output image X_(t) sampled accordingly?), then how to make sure every location of the output image is optimized? because of the resampling after each step of RMSprop? (if enough resampling is adopted, then the output image could be optimized?) can you help me figure it out?