FGNet: Fixation guidance network for salient object detection

In challenging scenarios (e.g., small objects and cluttered backgrounds), most existing algorithms suffer from inconsistent results with human visual attention. Since fixation prediction can better model the human visual attention mechanism and has a strong correlation with salient objects. Inspired by this, we proposed a fixation guidance network (FGNet) for salient object detection, which innovatively used fixation prediction to guide both salient object detection and edge detection. Firstly, a multi-branch network structure was designed to achieve multi-task detection. Each branch unit significantly learned the extracted features to accomplish the correct prediction. Secondly, given the strong correlation between the fixation and salient objects, a fixation guidance module was employed to guide salient object detection and edge detection for obtaining more accurate detection results. Finally, to full use the complementary relationship between salient features and edge features, we proposed a multi-resolution feature interaction module to achieve mutual optimization within the same feature and between the different features for suppressing noise and enhancing their representations. The experimental results show that our proposed method performed better in challenging scenes and outperformed existing state-of-the-art algorithms in several metrics on four public benchmark datasets.