Learning Non-Metric Visual Similarity for Image Retrieval

Noa Garcia, George Vogiatzis

Abstract: Measuring visual (dis)similarity between two or more instances within a data distribution is a fundamental task in many applications, specially in image retrieval. Theoretically, non-metric distances are able to generate a more complex and accurate similarity model than metric distances, provided that the non-linear data distribution is precisely captured by the similarity model. In this work, we analyze a simple approach for deep learning networks to be used as an approximation of non-metric similarity functions and we study how these models generalize across different image retrieval datasets.

TL;DR: Similarity network to learn a non-metric visual similarity estimation between a pair of images

Keywords: image retrieval, visual similarity, non-metric learning

Data: [Oxford5k](https://paperswithcode.com/dataset/oxford5k)