It is hard to find the right loss function for the metric learning without any hints. The solution is the DiscriminativeLoss which is a custom function in the attached utils folder. In future versions I suggest we introduce this function in the main code, or at least give a strong hint that one of the loss functions will be the DiscriminativeLoss
It is hard to find the right loss function for the metric learning without any hints. The solution is the DiscriminativeLoss which is a custom function in the attached utils folder. In future versions I suggest we introduce this function in the main code, or at least give a strong hint that one of the loss functions will be the DiscriminativeLoss