Multi-expert human action recognition with hierarchical super-class learning

作者：

Highlights：

• Still image action recognition is cast as a fine-grained classification.

• The proposed SCLAR method relaxes the requirement of auxiliary supervision.

• The proposed GCS algorithm promotes inter-class variation among different classes.

• Data imbalance issue in still image action recognition is addressed.

• The IHAR dataset with long-tailed distribution is introduced in action recognition.

摘要

•Still image action recognition is cast as a fine-grained classification.•The proposed SCLAR method relaxes the requirement of auxiliary supervision.•The proposed GCS algorithm promotes inter-class variation among different classes.•Data imbalance issue in still image action recognition is addressed.•The IHAR dataset with long-tailed distribution is introduced in action recognition.

论文关键词：Action recognition,Still images,Super-class learning,Long-tailed classification

论文评审过程：Received 6 November 2021, Revised 16 May 2022, Accepted 17 May 2022, Available online 27 May 2022, Version of Record 9 June 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2022.109091