Deep Metric Learning: A Survey

Abstract

Metric learning aims to measure the similarity among samples while using an optimal distance metric for learning tasks. Metric learning methods, which generally use a linear projection, are limited in solving real-world problems demonstrating non-linear characteristics. Kernel approaches are utilized in metric learning to address this problem. In recent years, deep metric learning, which provides a better solution for nonlinear data through activation functions, has attracted researchers' attention in many different areas. This article aims to reveal the importance of deep metric learning and the problems dealt with in this field in the light of recent studies. As far as the research conducted in this field are concerned, most existing studies that are inspired by Siamese and Triplet networks are commonly used to correlate among samples while using shared weights in deep metric learning. The success of these networks is based on their capacity to understand the similarity relationship among samples. Moreover, sampling strategy, appropriate distance metric, and the structure of the network are the challenging factors for researchers to improve the performance of the network model. This article is considered to be important, as it is the first comprehensive study in which these factors are systematically analyzed and evaluated as a whole and supported by comparing the quantitative results of the methods.

Enhancing Multi-tissue and Multi-scale Cell Nuclei Segmentation with Deep Metric Learning

“…The concept of deep metric learning was introduced when deep learning and metric learning were combined [34]. As such, the performance of a deep metric learning model depends not only on the quality of image dataset and the deep network architecture, but also on the metric loss function used in the learning.…”

“…Loss function is one of the most important functions in deep feature learning. In general, the models presented in the the domain of deep metric learning can be categorised as minimising either the contrastive loss or the triplet loss [33,34]. If the contrastive loss L C [35] function is selected, the model is penalised differently based on whether the classes of the samples are the same or different.…”

“…For the case considered in this paper, this function refers to the results of a 12th layer. Another metric considered in the current work is the triplet loss with semi-hard negative mining [20,34]. Instead of calculating the loss based on two examples, triplet loss L T involves three inputs: an anchor example X, similar to anchor input X S and dissimilar to anchor input X D .…”

Similarity Retention Loss (SRL) Based on Deep Metric Learning for Remote Sensing Image Retrieval

“…With the development of research, a number of loss functions have been proposed. Kaya M et al [21] combined with recent research results, revealed the importance of deep metric learning and summarized the current problems dealt with in this filed. For instance, the contrastive loss [22,23] captures the similarity or dissimilarity between pairwise of samples, while the triplet-based loss [12,24] describes the relationship among the triple samples.…”

Mitosis Detection Under Limited Annotation: A Joint Learning Approach

“…However, no effort has been dedicated to tackling the issue of limited training data. Recent advancements in meta-learning and metric learning [7] have paved the way for addressing the data scarcity issue [8]. Training HPFs 35 311 Test HPFs 15 295 Training mitoses 226 550 Test mitoses 101 533 In this work, we investigate deep metric learning and propose a classification method for mitosis detection utilizing both class label information and local spatial distribution information between training samples, to be able to learn from fewer annotations.…”