How to reproduce the results in the paper

[yueanga]

Thank you very much for open-sourcing your work. I tried to reproduce the results, but why is the generated output like this? There are very inconsistent texture variations (I used your code exactly, just changed the batch size to 2 due to limited memory)