Large Margin Mechanism and Pseudo Query Set on Cross-Domain Few-Shot Learning

In recent years, few-shot learning problems have received a lot of attention.\nWhile methods in most previous works were trained and tested on datasets in one\nsingle domain, cross-domain few-shot learning is a brand-new branch of few-shot\nlearning problems, where models handle datasets in different domains between\ntraining and testing phases. In this paper, to solve the problem that the model\nis pre-trained (meta-trained) on a single dataset while fine-tuned on datasets\nin four different domains, including common objects, satellite images, and\nmedical images, we propose a novel large margin fine-tuning method (LMM-PQS),\nwhich generates pseudo query images from support images and fine-tunes the\nfeature extraction modules with a large margin mechanism inspired by methods in\nface recognition. According to the experiment results, LMM-PQS surpasses the\nbaseline models by a significant margin and demonstrates that our approach is\nrobust and can easily adapt pre-trained models to new domains with few data.\n