Abstract:The objective of deep metric learning (DML) is to learn embeddings that can capture semantic similarity and dissimilarity information among data points. Existing pairwise or tripletwise loss functions used in DML are known to suffer from slow convergence due to a large proportion of trivial pairs or triplets as the model improves. To improve this, ranking-motivated structured losses are proposed recently to incorporate multiple examples and exploit the structured information among them. They converge faster and achieve state-of-the-art performance. In this work, we unveil two limitations of existing ranking-motivated structured losses and propose a novel ranked list loss to solve both of them. First, given a query, only a fraction of data points is incorporated to build the similarity structure. Consequently, some useful examples are ignored and the structure is less informative. To address this, we propose to build a set-based similarity structure by exploiting all instances in the gallery. The learning setting can be interpreted as few-shot retrieval: given a mini-batch, every example is iteratively used as a query, and the rest ones compose the gallery to search, i.e., the support set in few-shot setting. The rest examples are split into a positive set and a negative set. For every mini-batch, the learning objective of ranked list loss is to make the query closer to the positive set than to the negative set by a margin. Second, previous methods aim to pull positive pairs as close as possible in the embedding space. As a result, the intraclass data distribution tends to be extremely compressed. In contrast, we propose to learn a hypersphere for each class in order to preserve useful similarity structure inside it, which functions as regularisation. Extensive experiments demonstrate the superiority of our proposal by comparing with the state-of-the-art methods on the fine-grained image retrieval task. Our source code is available online: https://github.com/XinshaoAmosWang/Ranked-List-Loss-for-DML.

Relationship constraint deep metric learning

Deep Relational Metric Learning.

Deep Metric Learning with Adaptively Composite Dynamic Constraints

Ranked List Loss for Deep Metric Learning

Deep metric learning for image retrieval in smart city development

Towards Improved Proxy-based Deep Metric Learning via Data-Augmented Domain Adaptation

Potential Field Based Deep Metric Learning

Exponential Discriminative Metric Embedding in Deep Learning

Deep Factorized Metric Learning

Deep Metric Learning Based on Meta-Mining Strategy With Semiglobal Information

Deep Compositional Metric Learning

Improved Diversity-Promoting Collaborative Metric Learning for Recommendation

Face Recognition Based on LDA and Improved Pairwise-Constrained Multiple Metric Learning Method.

Anti-Collapse Loss for Deep Metric Learning Based on Coding Rate Metric

Distance Metric Learning with Joint Representation Diversification

Metric Learning with Adaptive Density Discrimination

Deep Metric Learning with Soft Orthogonal Proxies

Improving Deep Metric Learning Via Self-Distillation and Online Batch Diffusion Process

Constrained Empirical Risk Minimization Framework for Distance Metric Learning.

A multi-birth metric learning framework based on binary constraints

A Distributed Approach Toward Discriminative Distance Metric Learning