Co-saliency detection with intra-group two-stage group semantics propagation and inter-group contrastive learning

作者：

Highlights：

• We employ a two-stage group semantics propagation mechanism (TGSP).

• TGSP includes image-to-group propagation (IGP) and group-to-pixel propagation (GPP).

• IGP captures image interactions and GPP redistributes them to each pixel.

• We propose a contrastive learning module to minimize the noise object interference.

• Experimental results show that our TopicNet outperforms previous competitors.

摘要

•We employ a two-stage group semantics propagation mechanism (TGSP).•TGSP includes image-to-group propagation (IGP) and group-to-pixel propagation (GPP).•IGP captures image interactions and GPP redistributes them to each pixel.•We propose a contrastive learning module to minimize the noise object interference.•Experimental results show that our TopicNet outperforms previous competitors.

论文关键词：Co-salient object detection,Group semantics extraction,Noise object suppression,Contrastive learning

论文评审过程：Received 19 April 2022, Revised 28 June 2022, Accepted 28 June 2022, Available online 1 July 2022, Version of Record 11 July 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2022.109356