Deep Metric Learning using Similarities from Nonlinear Rank Approximations

2019 IEEE 21st International Workshop on Multimedia Signal Processing (MMSP '19)

TL;DR

We introduce a metric learning algorithm that enhances image similarity search by focusing on feature vectors that most impact retrieval quality. By computing normalized approximated ranks and using a nonlinear transfer function in a new loss function, our approach significantly improves deep feature embeddings across multiple datasets.

Abstract

In recent years, deep metric learning has achieved promising results in learning high dimensional semantic feature embeddings where the spatial relationships of the feature vectors match the visual similarities of the images. Similarity search for images is performed by determining the vectors with the smallest distances to a query vector. However, high retrieval quality does not depend on the actual distances of the feature vectors, but rather on the ranking order of the feature vectors from similar images. In this paper, we introduce a metric learning algorithm that focuses on identifying and modifying those feature vectors that most strongly affect the retrieval quality. We compute normalized approximated ranks and convert them to similarities by applying a nonlinear transfer function. These similarities are used in a newly proposed loss function that better contracts similar and disperses dissimilar samples. Experiments demonstrate significant improvement over existing deep feature embedding methods on the CUB-200-2011, Cars196, and Stanford Online Products data sets for all embedding sizes.

BibTeX

If you use our work in your research, please cite our publication:

@INPROCEEDINGS{8901815,
author={Schall, Konstantin and Barthel, Kai Uwe and Hezel, Nico and Jung, Klaus},
booktitle={2019 IEEE 21st International Workshop on Multimedia Signal Processing (MMSP)},
title={Deep Metric Learning using Similarities from Nonlinear Rank Approximations},
year={2019},
volume={},
number={},
pages={1-6},
keywords={Computer vision;Image retrieval;Content-based retrieval;Feature extraction;Machine learning algorithms;Nearest neighbor searches},
doi={10.1109/MMSP.2019.8901815}
}