CAA-Net: Class-Aware Attention Network for Selective Target Label Salient Object Detection

This paper aims to accomplish salient-object detection (SOD) by considering a selective-focus object class. In realistic applications of saliency prediction, people may focus on a particular class of objects. Few studies consider both objectives simultaneously to achieve the requirements for the above applications. Thus, we propose a novel class-aware SOD network that predicts salient objects within the specific class of human interest. Experiments reveal that the proposed network can effectively predict salient objects of the selected target class and can be applied to other saliency-based applications.