Dual class representation learning for few-shot image classification

作者：

Highlights：

• Proposes dual class representation learning (DCRL) for few-shot image classification.

• Jointly trains network on dual classes to incorporate class invariant information.

• Proposes novel approach for generating dual class images.

• Demonstrates superiority of DCRL over single class representation learning.

• Achieves state-of-the-art performance on multiple few-shot benchmark datasets.

摘要

•Proposes dual class representation learning (DCRL) for few-shot image classification.•Jointly trains network on dual classes to incorporate class invariant information.•Proposes novel approach for generating dual class images.•Demonstrates superiority of DCRL over single class representation learning.•Achieves state-of-the-art performance on multiple few-shot benchmark datasets.

论文关键词：Few-shot learning,Image classification,Deep learning,Representation learning

论文评审过程：Received 12 July 2021, Revised 25 November 2021, Accepted 27 November 2021, Available online 12 December 2021, Version of Record 1 January 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2021.107840