Difference between your ProtoNet and Chen et al's ProtoNet

[ashok-arjun]

Hi!

In your paper Table 3 (miniImageNet 5-way 5-shot), you have reported a result of 75.2 for Ours (ProtoNet), and 73.7 for Chen et al's ProtoNet for an image size of 224 x 224. However, there is no difference between both of the ProtoNet models, as seen from the paper as well as the code.

I ran the code for multiple seeds, for a good number of epochs, both with and without tracking stats in BatchNorm, but I could get only 70.6 for this setup.

Can you please elaborate on the result of 75.2 that you have obtained?

Thanks.