Group-wise co-salient object detection via multi-view self-labeling novel class discovery

1 Introduction Co-salient object detection(CoSOD)aims to extract the salient object(s)that are common across a group of relevant images[1].Group-wise clue plays a crucial role in accurately predicting the co-salient regions.Therefore,numerous groupwise deep models have been proposed by exploring consistency across images in unsupervised clustering manners[2-4]or the semantic connections guidance information[5].