Hardness-Aware Deep Metric Learning

About

This paper presents a hardness-aware deep metric learning (HDML) framework. Most previous deep metric learning methods employ the hard negative mining strategy to alleviate the lack of informative samples for training. However, this mining strategy only utilizes a subset of training data, which may not be enough to characterize the global geometry of the embedding space comprehensively. To address this problem, we perform linear interpolation on embeddings to adaptively manipulate their hard levels and generate corresponding label-preserving synthetics for recycled training, so that information buried in all samples can be fully exploited and the metric is always challenged with proper difficulty. Our method achieves very competitive performance on the widely used CUB-200-2011, Cars196, and Stanford Online Products datasets.

Wenzhao Zheng, Zhaodong Chen, Jiwen Lu, Jie Zhou · 2019

Related benchmarks

Task	Dataset	Result
Image Retrieval	CUB-200-2011 (test)	Recall@153.7	251
Image Retrieval	Stanford Online Products (test)	Recall@168.7	231
Image Retrieval	CARS196 (test)	Recall@179.1	134
Deep Metric Learning	CUB200 2011 (test)	Recall@153.7	129
Image Retrieval	Stanford Online Products	Recall@168.7	64
Image Retrieval	CARS 196 (test)	Recall@179.1	56
Deep Metric Learning	CARS196 (test)	R@179.1	56
Deep Metric Learning	CARS196	Recall@179.1	50
Deep Metric Learning	SOP (test)	Recall@168.7	32
Clustering	CARS 196 (test)	NMI69.7	21