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.

Deep Metric Learning with Soft Orthogonal Proxies

Multi-proxy based deep metric learning

A Non-isotropic Probabilistic Take on Proxy-based 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

A Hierarchical Multi-Proxy Loss with Dynamic Main-Proxy for Deep Metric Learning

A Unified Multiple Proxy Deep Metric Learning Framework Embedded With Distribution Optimization for Fine-Grained Ship Classification in Remote Sensing Images

Potential Field Based Deep Metric Learning

Ranked List Loss for Deep Metric Learning

Proxy Synthesis: Learning with Synthetic Classes for Deep Metric Learning

Relationship constraint deep metric learning

Hierarchical multiple proxy loss for deep metric learning

Proxy-based Robust Deep Metric Learning in the Presence of Label Noise

Scalable Similarity-Consistent Deep Metric Learning for Face Recognition.

Learnable dynamic margin in deep metric learning

Deep Metric Learning with Adaptively Composite Dynamic Constraints

Exponential Discriminative Metric Embedding in Deep Learning

Learning Deep Optimal Embeddings with Sinkhorn Divergences

Fewer is More: A Deep Graph Metric Learning Perspective Using Fewer Proxies.

Deep Metric Learning with Graph Consistency

Visual Embedding Augmentation in Fourier Domain for Deep Metric Learning