Dual Pyramid Network for Salient Object Detection

Publication date: Available online 27 September 2019Source: NeurocomputingAuthor(s): Xuemiao Xu, Jiaxing Chen, Huaidong Zhang, Guoqiang HanAbstractSaliency detection is a fundamental yet challenging task in computer vision, aiming at identifying the most visually distinctive objects in an image. Despite promising results from recent deep learning based methods, existing works are still not sufficient to incorporate momentous global context to detect salient objects. In this work, we present a dual pyramid network (DPNet) for salient object detection by formulating self-attention mechanism into the sub-regions based contexts. Our key idea is to aggregate different sub-regions based contexts into the enhanced global context without sacrificing the resolution of deep features. To achieve this, we design the spatial pyramid module and channel-wise pyramid module. The proposed modules can enhance the global context by incorporating self-attention mechanism into the sub-regions based contexts in the spatial and channel-wise dimension, respectively. Compared with 23 state-of-the-art methods on six RGB saliency benchmark datasets, experimental results show that our method performs favorably over all the others, both quantitatively and visually. The superior performance of proposed DPNet than others on three RGB-D saliency benchmark datasets further demonstrates the powerful generalization ability of our network.
Source: Neurocomputing - Category: Neuroscience Source Type: research