Learning the Roots of Visual Domain Shift

TL;DR

This paper introduces a method to localize and leverage spatial domain shift information within images to improve domain adaptation performance, achieving state-of-the-art results on benchmark datasets.

Contribution

It presents a novel approach to learn spatial domain shift maps using CNN visualization techniques, enhancing domain adaptation by incorporating local domain specificity features.

Findings

Improved classification accuracy on the Office dataset.

Localization of domain shift within images.

Enhanced domain adaptation performance with spatial features.

Abstract

In this paper we focus on the spatial nature of visual domain shift, attempting to learn where domain adaptation originates in each given image of the source and target set. We borrow concepts and techniques from the CNN visualization literature, and learn domainnes maps able to localize the degree of domain specificity in images. We derive from these maps features related to different domainnes levels, and we show that by considering them as a preprocessing step for a domain adaptation algorithm, the final classification performance is strongly improved. Combined with the whole image representation, these features provide state of the art results on the Office dataset.