Fourier-Attentive Representation Learning: A Fourier-Guided Framework for Few-Shot Generalization in Vision-Language Models

TL;DR

This paper introduces FARL, a Fourier-based framework that disentangles structural and stylistic features in images to improve few-shot generalization in vision-language models, showing strong results across multiple datasets.

Contribution

FARL explicitly separates visual structure and style using Fourier analysis and a dual cross-attention mechanism, enhancing model robustness and generalization.

Findings

Improved few-shot learning performance on 15 datasets.

Effective disentanglement of structure and style features.

Enhanced vision-language alignment through deep token injection.

Abstract

Large-scale pre-trained Vision-Language Models (VLMs) have demonstrated strong few-shot learning capabilities. However, these methods typically learn holistic representations where an image's domain-invariant structure is implicitly entangled with its domain-specific style. This presents an opportunity to further enhance generalization by disentangling these visual cues. In this paper, we propose Fourier-Attentive Representation Learning (FARL), a novel framework that addresses this by explicitly disentangling visual representations using Fourier analysis. The core of our method is a dual cross-attention mechanism, where learnable representation tokens separately query an image's structural features (from the phase spectrum) and stylistic features (from the amplitude spectrum). This process yields enriched, disentangled tokens that are then injected deep into the VLM encoders to guide adaptation. Our design, which includes an asymmetric injection strategy, forces the model to learn a more robust vision-language alignment. Extensive experiments on 15 datasets demonstrate the effectiveness of our approach.