voxceleb_trainer
voxceleb_trainer copied to clipboard
softmaxproto loss for binary classification
Hi @joonson, I am trying to adapt the code to train a binary classifier. I am getting a validation accuracy of up to 99%, but the loss values are very very high (in the order of 10^9). Is it normal? Is it a good idea to use softmaxproto loss for binary classification problem? if not, why?