Timezone: »

 
Poster
Rectifying the Shortcut Learning of Background for Few-Shot Learning
Xu Luo · Longhui Wei · Liangjian Wen · Jinrong Yang · Lingxi Xie · Zenglin Xu · Qi Tian

Thu Dec 09 12:30 AM -- 02:00 AM (PST) @
in Poster Session 5 »

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.

Keywords: Machine Learning · Contrastive Learning · Few Shot Learning

Author Information

Xu Luo (University of Electronic Science and Technology of China)
Longhui Wei (University of Science and Technology of China)
Liangjian Wen (Huawei Technologies Ltd.)
Jinrong Yang (Huazhong University of Science and Technology)
Lingxi Xie (Huawei Noah's Ark Lab)
Zenglin Xu (University of Electronic Science and Technology of China)
Qi Tian (University of Texas, San Antonio)

More from the Same Authors