Fail to reproduce Inception Score on CIFAR10

[taufikxu]

Dear authors:

Thanks for your code and SN-GAN is a really impressive work.

However, I found that I fail to reproduce the inception score of 8.22 on CIFAR10. I re-run the script and the Inception score for the final model, i.e., 50000 iterations, only achieved about 8.0. Indeed, the highest IS during the training process is 8.28, which is even higher than 8.22. So I want to know that whether the number reported in the paper is the highest IS during the training process or just the IS in the final?

Thanks again!