Deep Domain Adaptation by Geodesic Distance Minimization

TL;DR

This paper introduces Deep LogCORAL, a novel unsupervised domain adaptation method that minimizes geodesic distances between source and target domain feature distributions, improving over previous Euclidean-based approaches.

Contribution

The paper proposes using Riemannian geodesic distances with Log-Euclidean approximation in deep domain adaptation, incorporating both mean and covariance alignment for better performance.

Findings

Deep LogCORAL outperforms Deep CORAL on the Office dataset.

Incorporating first and second order information yields further improvements.

The method achieves state-of-the-art results in unsupervised domain adaptation.

Abstract

In this paper, we propose a new approach called Deep LogCORAL for unsupervised visual domain adaptation. Our work builds on the recently proposed Deep CORAL method, which proposed to train a convolutional neural network and simultaneously minimize the Euclidean distance of convariance matrices between the source and target domains. We propose to use the Riemannian distance, approximated by Log-Euclidean distance, to replace the naive Euclidean distance in Deep CORAL. We also consider first-order information, and minimize the distance of mean vectors between two domains. We build an end-to-end model, in which we minimize both the classification loss, and the domain difference based on the first and second order information between two domains. Our experiments on the benchmark Office dataset demonstrate the improvements of our newly proposed Deep LogCORAL approach over the Deep CORAL method, as well as further improvement when optimizing both orders of information.