Abstract: One effective approach for solving few-shot classification is learning deep representations that measure the similarity between query images and a few support images of specific categories.