FocusNet: Classifying better by focusing on confusing classes

作者：

Highlights：

• Showing the relationship between network attention and confusing classes.

• Proposing FocusNet and the focus-picking loss function to address the class-confusion issue.

• Focusing more on confusing classes and difficult samples during training.

• Accuracy improvement on multiple classification datasets.

摘要

•Showing the relationship between network attention and confusing classes.•Proposing FocusNet and the focus-picking loss function to address the class-confusion issue.•Focusing more on confusing classes and difficult samples during training.•Accuracy improvement on multiple classification datasets.

论文关键词：Image classification,Inter-class correlations,Confusing classes

论文评审过程：Received 9 August 2021, Revised 14 March 2022, Accepted 10 April 2022, Available online 13 April 2022, Version of Record 21 April 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.108709