Can not replicate the results of MAML in pytorch

I think everything is the same as official tf impl of MAML, but we cannot replicate the results of MiniImageNet. We got 45% in pytorch, 48%+ in tf.

Is there anyone to succeed this impl of MAML in pytorch?

have a look at https://github.com/katerakelly/pytorch-maml

As she wrote in README, she was successful in one setting, not in other setting… I think there is no perfect impl of MAML in pytorch yet.

I have a PyTorch implementation, and performance there is slightly lagging the TF version as well.

There are lots and lots of hyperparameters to keep track of. I assume that’s the issue.

I was wondering if you could share your implementation…

@patrickhop Could u kindly share the source link?