SRasP: Self-Reorientation Adversarial Style Perturbation for Cross-Domain Few-Shot Learning
This work provides an incremental improvement in model robustness and transferability for researchers and practitioners working on Cross-Domain Few-Shot Learning, by stabilizing style-based perturbation methods.
This paper addresses the problem of gradient instability and convergence to sharp minima in existing style-based perturbation methods for Cross-Domain Few-Shot Learning (CD-FSL). The authors propose SRasP, a novel crop-global style perturbation network that leverages global semantic guidance to identify incoherent crops, reorients and aggregates their style gradients, and uses a multi-objective optimization function to maximize visual discrepancy while enforcing semantic consistency. This approach leads to consistent improvements over state-of-the-art methods on multiple CD-FSL benchmarks.
Cross-Domain Few-Shot Learning (CD-FSL) aims to transfer knowledge from a seen source domain to unseen target domains, serving as a key benchmark for evaluating the robustness and transferability of models. Existing style-based perturbation methods mitigate domain shift but often suffer from gradient instability and convergence to sharp minima.To address these limitations, we propose a novel crop-global style perturbation network, termed Self-Reorientation Adversarial \underline{S}tyle \underline{P}erturbation (SRasP). Specifically, SRasP leverages global semantic guidance to identify incoherent crops, followed by reorienting and aggregating the style gradients of these crops with the global style gradients within one image. Furthermore, we propose a novel multi-objective optimization function to maximize visual discrepancy while enforcing semantic consistency among global, crop, and adversarial features. Applying the stabilized perturbations during training encourages convergence toward flatter and more transferable solutions, improving generalization to unseen domains. Extensive experiments are conducted on multiple CD-FSL benchmarks, demonstrating consistent improvements over state-of-the-art methods.