Prototype adjustment for zero shot classification

作者：

Highlights：

• The proposed method jointly adjusting the prototypes and learning mapping function.

• Adjust the prototypes to make them accurate, separable and discriminative.

• A generalizable mapping function is learnt to tackle the domain shift problem.

• The proposed method shows promising results on ZSL datasets.

摘要

•The proposed method jointly adjusting the prototypes and learning mapping function.•Adjust the prototypes to make them accurate, separable and discriminative.•A generalizable mapping function is learnt to tackle the domain shift problem.•The proposed method shows promising results on ZSL datasets.

论文关键词：Zero shot classification,Prototype adjustment,Separation,Mapping function

论文评审过程：Received 9 September 2018, Revised 1 January 2019, Accepted 25 February 2019, Available online 12 March 2019, Version of Record 13 March 2019.

论文官网地址：https://doi.org/10.1016/j.image.2019.02.011