Learning discriminative and meaningful samples for generalized zero shot classification

作者：

Highlights：

• Introducing class consistency to GAN to make the generated samples discriminative.

• Introducing semantic consistency to GAN to make the generated samples meaningful.

• Selecting confident unseen samples to augment the training sets to alleviate the bias problem.

• The proposed method outperforms existing methods on GZSC tasks.

摘要

•Introducing class consistency to GAN to make the generated samples discriminative.•Introducing semantic consistency to GAN to make the generated samples meaningful.•Selecting confident unseen samples to augment the training sets to alleviate the bias problem.•The proposed method outperforms existing methods on GZSC tasks.

论文关键词：Generalized zero shot classification,Generative adversarial network,Class consistency,Semantic consistency

论文评审过程：Received 14 November 2019, Revised 11 June 2020, Accepted 11 June 2020, Available online 20 June 2020, Version of Record 22 June 2020.

论文官网地址：https://doi.org/10.1016/j.image.2020.115920