Hi! Thanks for the code. I have been using your repository for quite a while and had cited your works several times.

You have provided the weight links in the README. However, I can't reproduce your reported results using those pretrained weights. For example, I can only get 56% for protonet, despite adding in the --use_euclidean flags and so on. I wonder whether these pretrained weights are already converged, or I need further training? Or it's because I didn't use the right hyperparameters for testing?

Thanks again.