search

clovaai / voxceleb_trainer

In defence of metric learning for speaker recognition
MIT License
1.02k stars 272 forks source link

Lossfunction issue #123

Closed makimon123 closed 2 years ago

makimon123 commented 3 years ago

when i tyoe the command“python ./trainSpeakerNet.py --model ResNetSE34V2 --log_input True --encoder_type SAP --trainfunc angleproto --save_path exps/exp1 --nClasses 5994 --batch_size 2 --margin 0.3”, I'll get the result below, "IndexError: Dimension out of range (expected to be in range of [-1, 0], but got 1)“
but "--trainfunc amsoftmax" not

makimon123 commented 3 years ago

error "nloss = self.criterion(cos_sim_matrix, label) "

Jungjee commented 2 years ago

You have to set nPerSpeaker to 2 or a higher value to use angleproto as it is a metric learning loss function.

speaker-lover commented 2 years ago

Excuse me, what is the result of using angleproto loss function? I got eer=2.6 minDCF=0.2. Is this normal? Does it belong to self-supervised learning method or supervised learning method?

makimon123 commented 2 years ago

您必须设置为 2 或更高的值才能使用，因为它是指标学习损失函数。nPerSpeaker``angleproto

thanks a bunch.