predictions don't match?

[josegarciacarreno]

Hello, thanks for this, great job!

I installed the dockerfile and out of the box I tried your script for the sample images you provide in the docs (the ones starting with the beach sunset and ending with the boring hotel room) I expected them to match exactly, but they differ quite a lot, for example:
sunset 6.52 stated --> 5.37 predicted for aesthetic_0.07
houses 4.92 ----------> 5.54

what is the reason? Do scores belong to different training stages?

thnx for releasing all this code!!