Hierarchical complementary learning for weakly supervised object localization

作者：

Highlights：

• Weakly supervised object localization aims to localize objects using image labels.

• HCLNet hierarchically generates different class activation maps, and fuses them.

• The addition strategy and the l1-norm strategy have been introduced to fuse the CAMs.

• Extensive experiments show that HCLNet achieves a new state-of-the-art performance.

摘要

•Weakly supervised object localization aims to localize objects using image labels.•HCLNet hierarchically generates different class activation maps, and fuses them.•The addition strategy and the l1-norm strategy have been introduced to fuse the CAMs.•Extensive experiments show that HCLNet achieves a new state-of-the-art performance.

论文关键词：Weakly supervised object localization,Class activation map,Complementary map,Fusion strategy

论文评审过程：Received 5 September 2020, Revised 1 August 2021, Accepted 22 September 2021, Available online 8 October 2021, Version of Record 1 November 2021.

论文官网地址：https://doi.org/10.1016/j.image.2021.116520