Reviews: Delta-encoder: an effective sample synthesis method for few-shot object recognition

It is also encouraging that the reported results compare favourably to the recent paper by Gidaris et al. Finally, the provided explanation fully clarify my understanding of the algorithm, and I feel it's a nice and simple idea. I've raised my score to 7 as a result of all the above. One thing that would be important to clarify is that the feature extractor is trained exclusively on the meta-training set of each dataset considered. Another reviewer misunderstood this fact, and I remember also misunderstanding this during my first pass of the paper.