CycleMix: Mixing Source Domains for Domain Generalization in Style-Dependent Data

TL;DR

CycleMix tackles domain generalization in style-dependent data by using CycleGANs to learn domain-specific styles and then randomly mixing these styles to synthesize novel training samples. This augmentation encourages a feature extractor to focus on style-invariant, class-relevant information, improving performance on unseen domains. Evaluated on the PACS benchmark, CycleMix outperforms strong baselines by several percentage points, demonstrating the practical value of style-translation augmentation for DG. The approach highlights a trade-off between expanded style diversity and computational cost, pointing to future work in multi-domain translation and broader dataset evaluation.

Abstract

As deep learning-based systems have become an integral part of everyday life, limitations in their generalization ability have begun to emerge. Machine learning algorithms typically rely on the i.i.d. assumption, meaning that their training and validation data are expected to follow the same distribution, which does not necessarily hold in practice. In the case of image classification, one frequent reason that algorithms fail to generalize is that they rely on spurious correlations present in training data, such as associating image styles with target classes. These associations may not be present in the unseen test data, leading to significant degradation of their effectiveness. In this work, we attempt to mitigate this Domain Generalization (DG) problem by training a robust feature extractor which disregards features attributed to image-style but infers based on style-invariant image representations. To achieve this, we train CycleGAN models to learn the different styles present in the training data and randomly mix them together to create samples with novel style attributes to improve generalization. Experimental results on the PACS DG benchmark validate the proposed method.

Figures (2)

• Figure 1: Illustration of the proposed CycleMix augmentation method. Before passing through a feature extractor (e.g ResNet-50), the styles of each domain source domain are mixed together in order to create samples from novel domains. As the magniutude of each style component is random in each minibatch, the model is constantly provided with previously unseen data samples during training.
• Figure 2: Indicative examples of the trained CycleGAN translations between domains in the PACS dataset. Despite the relatively small size of the samples, the CycleGAN models are able to capture the style attributes in each domain.