Hi,

The previous version of code (before synchronous BN) had evaluation while training (after each epoch). It seems that the new version does not have it. Is it because of Sync BN training/evaluation difference?

Thanks