will you share the pretrained model which used amsoftmax, i trained myself use amsoftmax but very hard to convergence

WeidiXie / VGG-Speaker-Recognition

Utterance-level Aggregation For Speaker Recognition In The Wild

364 stars 97 forks source link

will you share the pretrained model which used amsoftmax, i trained myself use amsoftmax but very hard to convergence #23

Closed mmxuan18 closed 5 years ago

mmxuan18 commented 5 years ago

i trained the model use the same backbone and ghost pooling, with softmax and amsoftmax, but the amsoftmax acc is very low, any tricks for training amsoftmax?

WeidiXie commented 5 years ago

Don't train amsoftmax from scratch. Try to initialise the model with pretrained softmax, and the amsoftmax I trained is not as good as softmax version, so I won't post it, you can fine-tune on what I provided if you want.