Improved Few-Shot Learning for Images Classification

Abstract

Few-shot learning is an approach that classify unseen classes with limited labeled samples. We propose improved networks of Relation Network to classify images with small samples. The improved networks is ECA Relation Network (ECA-RNET). The accuracy of ECA-RNET is 52.24% and 67.85% on 5-way 1-shot and 5-way 5-shot of mini-ImageNet dataset, respectively.