Few-Shot Classification With Feature Map Reconstruction Networks

Reformulate few-shot classification as latent space reconstruction  Approximate query features as weighted sums of support features; assign class based on reconstruction quality  Simple, closed-formed ridge regression is more performant and efficient than previous approaches  Across-the-board superiority on four fine-grained datasets and the cross-domain setup  Competitive results on mini-ImageNet and tiered-ImageNet with minimal bells and whistles