Rectifying the Shortcut Learning of Background for Few-Shot Learning
This addresses a domain-specific issue in Few-Shot Learning by mitigating background shortcuts to improve generalization, representing an incremental advance.
The paper tackled the problem of shortcut learning from image backgrounds in Few-Shot Learning, which hinders generalization across categories, and proposed a framework called COSOC that extracts foreground objects without extra supervision, achieving effectiveness in inductive FSL tasks.
The category gap between training and evaluation has been characterised as one of the main obstacles to the success of Few-Shot Learning (FSL). In this paper, we for the first time empirically identify image background, common in realistic images, as a shortcut knowledge helpful for in-class classification but ungeneralizable beyond training categories in FSL. A novel framework, COSOC, is designed to tackle this problem by extracting foreground objects in images at both training and evaluation without any extra supervision. Extensive experiments carried on inductive FSL tasks demonstrate the effectiveness of our approaches.