Unsupervised Natural Image Patch Learning

TL;DR

This paper introduces an unsupervised method for learning a natural image patch similarity metric using spatial proximity, eliminating the need for annotated data and enabling effective domain adaptation.

Contribution

It presents a novel unsupervised deep embedding approach based on spatial proximity, outperforming supervised methods and facilitating self-supervised domain adaptation.

Findings

Unsupervised embedding outperforms supervised counterparts.

The method effectively captures patch similarity without annotations.

Enables efficient self-supervised domain adaptation.

Abstract

Learning a metric of natural image patches is an important tool for analyzing images. An efficient means is to train a deep network to map an image patch to a vector space, in which the Euclidean distance reflects patch similarity. Previous attempts learned such an embedding in a supervised manner, requiring the availability of many annotated images. In this paper, we present an unsupervised embedding of natural image patches, avoiding the need for annotated images. The key idea is that the similarity of two patches can be learned from the prevalence of their spatial proximity in natural images. Clearly, relying on this simple principle, many spatially nearby pairs are outliers, however, as we show, the outliers do not harm the convergence of the metric learning. We show that our unsupervised embedding approach is more effective than a supervised one or one that uses deep patch representations. Moreover, we show that it naturally leads itself to an efficient self-supervised domain adaptation technique onto a target domain that contains a common foreground object.