Problem of the “full context embeddings” implement

[cultivater]

First thank you for your code. But I am puzzled about your “full context embeddings” implementation. In the paper, the process of g' and f' are different, but you just stacked g' and f' and put them into a LSTM. Perhaps this is why the accuray is low under miniImagenet ,I think?

[sunbear616]

I face the same question, I don't know why, but most pytorch code, they all do not implemente the same both g' and f', just use g I think

[sunbear616]

Do you implement this part? may I see your code?

[AugusYaoI]

I also want to ask the same question. I doubt if this code is consistent with the paper. I hope the author of this code can answer this question.