How do you reproduce the NN model in the baseline

[a-home-bird]

tag extension how to be used in few-shot. Could you please describe as detailed as possible? Thank you very much

[rtmaww]

Hello. The original experiments in the NNShot paper are source->target setting, which means they first train the model on source domain training data and then test the model on the target domain N-shot support set. In our experiments, we adopted the source-free experimental setting. So we trained the NNShot (or StructShot) model directly on the N-shot support set (here is the training set) in the target domain and tested with the same N-shot support set (which means we also use the N-shot training data to calculate the nearest neighbor for testing).