Obtained test accuracy lower than reported for miniImageNet

[kikyou123]

Hello , thank you for the implementation of your work. For the first stage, I used the Conv128CosineClassifier configuration. Later with Attention based 5way 1-shot training(untouched configuration), I get 55.9 on novel class and 70.3 on base class for the test set. It is lower than your report. I wonder whether the result I got is reasonable or not?

[asdfqwer2015]

I got same issue too, even lower. But I tested Conv64 in my test, which got novel: 55.47% and base: 67.98% for posted novel: 55.86% and base: 68.43% by the author.