Recent progress on salient object detection (SOD) mainly benefits from
multi-scale learning, where the high-level and low-level features collaborate
in locating salient objects and discovering fine details, respectively.
However, most efforts are devoted to low-level feature learning b