I have utilized your repository for training multiple models with various hyper-parameter changes, however, one thing that is common in all of my trainings is that the secret loss (cross entropy) isnt being optimized in all of them or there is a minute optimization resulting in a really low str_acc.
Do you know what could be the possible issue for that?
Hi Tancik,
I have utilized your repository for training multiple models with various hyper-parameter changes, however, one thing that is common in all of my trainings is that the secret loss (cross entropy) isnt being optimized in all of them or there is a minute optimization resulting in a really low str_acc.
Do you know what could be the possible issue for that?