miniImagenet experiments about maml-pytorch HOT 4 CLOSED

hi, maybe you need enlarge the default epoches. I trained it and got 46% with no modification.

Since it involves 2nd order derivative, so the gpu memory occupy is extremely demanding. Maybe you can scale the task number down to try larger channels. But on the other hand, any changes will harm the performance.

MAML is really sensitive.

from maml-pytorch.

I also cannot reproduce the result using the default para.
The result for 5 way 1 shot is 42.94%. How can I address this issue.

from maml-pytorch.

Hi, my performance report in readme file is true. just enlarge training epoch.

from maml-pytorch.

hi，The results of my running are as follows：

“Test acc: [0.2012 0.5845 0.6074 0.613 0.6143 0.6147]”

I ran the code for six hours in Tesla k80 ，miniImageNet-5-way-1-shot

from maml-pytorch.