LAUSR

Abstract In recent years, visual saliency has witnessed tremendous progress through using deep convolutional neural networks (CNNs). For effective salient object detection, contextual information has been widely employed since the global context can tell different objects apart while the local context can distinguish salient ones from the background. Inspired by this, in this paper we propose a novel Multi-scale Pyramid Pooling Network (MPPNet) by exploiting global and local context in a unified way. This is achieved by incorporating hierarchical local information and global pyramid pooling representation. Particularly, the integration of multi-scale pyramid pooling proves its capacity to produce high-quality prediction map through the use of multiple pooling variables. Quantitative and qualitative experiments demonstrate the effectiveness of the proposed framework. Our method can significantly improve the performance based on four popular benchmark datasets.