As a newly emerging and significant topic in computer vision community, co-saliency detection aims at discovering the common salient objects in multiple related images. The existing methods often generate the co-saliency map through a direct forward pipeline which is based on the designed cues or in