Abstract:Although great success has been achieved in various computer vision tasks, deep neural networks (DNNs) suffer dramatic performance degradation when evaluated on out-of-distribution data. Domain generalization (DG) is proposed to handle this problem by learning domain-agnostic information from multiple source domains to generalize well on unseen target domains. Several methods resort to Fourier transform due to its simplicity and efficiency. They argue that amplitude spectra imply domain-specific information, which should be suppressed, while phase counterparts imply domain-agnostic information, which should be preserved. However, these methods only suppress the domain-specific information in source domains and neglect the relationship with target domains, leading to the persistence of the domain gap. Besides, these methods preserve domain-agnostic information by keeping phase components unchanged, causing them to be underutilized. In this paper, we propose Dual Branch Augmentation Module (DBAM) by leveraging Fourier transform and taking advantage of both amplitude and phase spectra. For the amplitude branch, we propose Inner-domain Amplitude Distribution Rectification (IADR) and Cross-domain Amplitude Dirichlet Mixup (CADM) to stabilize the training process and explore more feature space. In addition, we propose Test-time Amplitude Prototype Calibration (TAPC) to construct the connection between source and target domains during evaluation to further mitigate the domain gap. For the phase branch, we propose Random Symmetric Phase Perturbation (RSPP) to enhance the robustness for recognizing domain-agnostic information. With the combined contributions of the two branches, DBAM significantly surpasses other state-of-the-art (SOTA) methods. Extensive experiments on four benchmarks and further analysis demonstrate the effectiveness of DBAM.

Fourier-based augmentation with applications to domain generalization

A Fourier-based Framework for Domain Generalization

Cross-Domain Feature Augmentation for Domain Generalization

Fourier-basis Functions to Bridge Augmentation Gap: Rethinking Frequency Augmentation in Image Classification

Domain Generalization with Fourier Transform and Soft Thresholding

Domain generalization via geometric adaptation over augmented data

It takes two: Dual Branch Augmentation Module for domain generalization

A Simple Feature Augmentation for Domain Generalization

A Fourier Transform Framework for Domain Adaptation

Fourier-Based Augmentations for Improved Robustness and Uncertainty Calibration

Untapped Potential of Data Augmentation: A Domain Generalization Viewpoint

Improving Stereo Matching Generalization Via Fourier-Based Amplitude Transform

Towards Combating Frequency Simplicity-biased Learning for Domain Generalization

Visual Embedding Augmentation in Fourier Domain for Deep Metric Learning

Fourier-Based Frequency Space Disentanglement and Augmentation for Generalizable Face Anti-Spoofing

Inter-Class and Inter-Domain Semantic Augmentation for Domain Generalization

Learning to Augment via Implicit Differentiation for Domain Generalization

Semantic Data Augmentation based Distance Metric Learning for Domain Generalization

FAMLP: A Frequency-Aware MLP-Like Architecture For Domain Generalization

Augmentation-based Domain Generalization for Semantic Segmentation

Progressive Fourier Adversarial Domain Adaptation for Object Classification and Retrieval