Multi-scale feature aggregation and boundary awareness network for salient object detection

Abstract

Salient object detection aims to detect the most visually distinctive objects in an image. Although existing FCN-based methods have shown strong advantages in this field, scale variation and complex boundary are still great challenges. In this paper, we propose a multi-scale feature aggregation and boundary awareness network to overcome the problems. Multi-scale feature aggregation module is proposed to integrate adjacent hierarchical features and the multiple aggregation strategy solves the problem of scale variation. To obtain more effective multi-scale features from integrated features, a cross feature refinement module is proposed to compose the decoder. For the issue of complex boundary, we design a boundary pixel awareness loss function to enable the network to acquire boundary information and generate high-quality saliency maps with better boundary. Experiments on five benchmark datasets show that our network outperforms recent state-of-the-art detectors quantitatively and qualitatively.