Multi-level Metric Learning for Few-Shot Image Recognition

Few-shot learning devotes to training a model on few samples. Most of these approaches learn based pixel-level or global-level feature representation. However, using global features may lose local information, and the contextual semantics image. Moreover, such works can only measure their relations single level, which is not comprehensive effective. And if query images simultaneously be well classified via three distinct level similarity metrics, within class more tightly distributed in smaller space, generating discriminative maps. Motivated by this, we propose novel Part-level Embedding Adaptation with Graph (PEAG) method generate task-specific features. Multi-level Metric Learning (MML) proposed, calculates part-level but also considers metrics. Extensive experiments popular few-shot image recognition datasets prove effectiveness our compared state-of-the-art methods. Our code available at: https://github.com/chenhaoxing/M2L .