Ranked List Loss for Deep Metric Learning

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.

Accepted to T-PAMI. Therefore, to read the offical version, please go to IEEE Xplore. Fine-grained image retrieval task. Our source code is available online: https://github.com/XinshaoAmosWang/Ranked-List-Loss-for-DML

References in corpus (19)

• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• Prototypical Networks for Few-shot Learning
• Caffe: Convolutional Architecture for Fast Feature Embedding
• In Defense of the Triplet Loss for Person Re-Identification
• Going Deeper with Convolutions
• SphereFace: Deep Hypersphere Embedding for Face Recognition
• Learning Deep Embeddings with Histogram Loss
• Multi-Similarity Loss with General Pair Weighting for Deep Metric Learning
• Circle Loss: A Unified Perspective of Pair Similarity Optimization
• Learning Fine-grained Image Similarity with Deep Ranking
• SoftTriple Loss: Deep Metric Learning Without Triplet Sampling
• Embedding Deep Metric for Person Re-identication A Study Against Large Variations
• No Fuss Distance Metric Learning using Proxies
• Smart Mining for Deep Metric Learning
• Classification is a Strong Baseline for Deep Metric Learning
• MIC: Mining Interclass Characteristics for Improved Metric Learning
• Divide and Conquer the Embedding Space for Metric Learning
• Unbiased Evaluation of Deep Metric Learning Algorithms
• Instance Cross Entropy for Deep Metric Learning

Cited by in corpus (1)

• Spherical Feature Transform for Deep Metric Learning