Single Domain Generalization via Normalised Cross-correlation Based Convolutions

TL;DR

This paper introduces XCNorm, a novel linear operator based on normalized cross-correlation, which enhances the robustness of deep neural networks to domain shifts in single domain generalization tasks without relying on data augmentation.

Contribution

The paper proposes XCNorm, a new operator that improves domain shift robustness by being invariant to affine and energy changes, eliminating the need for non-linear activations in deep networks.

Findings

XCNorm achieves competitive performance on S-DG benchmarks.

Deep networks with XCNorm are robust to semantic distribution shifts.

The method reduces reliance on data augmentation techniques.

Abstract

Deep learning techniques often perform poorly in the presence of domain shift, where the test data follows a different distribution than the training data. The most practically desirable approach to address this issue is Single Domain Generalization (S-DG), which aims to train robust models using data from a single source. Prior work on S-DG has primarily focused on using data augmentation techniques to generate diverse training data. In this paper, we explore an alternative approach by investigating the robustness of linear operators, such as convolution and dense layers commonly used in deep learning. We propose a novel operator called XCNorm that computes the normalized cross-correlation between weights and an input feature patch. This approach is invariant to both affine shifts and changes in energy within a local feature patch and eliminates the need for commonly used non-linear activation functions. We show that deep neural networks composed of this operator are robust to common semantic distribution shifts. Furthermore, our empirical results on single-domain generalization benchmarks demonstrate that our proposed technique performs comparably to the state-of-the-art methods.