Learning domain invariant unseen features for generalized zero-shot classification

作者：

Highlights：

• Learn the support seen class sets for unseen classes to generate unseen samples.

• Learn domain invariant unseen features to minimize domain shift problem.

• Select confident target samples to retrain the classifier to alleviate the bias problem.

• Extensive experimental results show astonishing results on GZSC tasks.

摘要

•Learn the support seen class sets for unseen classes to generate unseen samples.•Learn domain invariant unseen features to minimize domain shift problem.•Select confident target samples to retrain the classifier to alleviate the bias problem.•Extensive experimental results show astonishing results on GZSC tasks.

论文关键词：Generalized zero shot classification,Domain invariant unseen features,Support seen class sets,Maximum Mean Discrepancy

论文评审过程：Received 26 February 2020, Revised 2 July 2020, Accepted 5 August 2020, Available online 8 August 2020, Version of Record 12 August 2020.

论文官网地址：https://doi.org/10.1016/j.knosys.2020.106378