Search

Existing RGB-D Salient Object Detection (SOD) methods take advantage of depth cues to improve the detection accuracy, while pay insufficient attention to the quality of depth information. In practice, a depth map is often with uneven quality and sometimes suffers from distractors, due to various factors in the acquisition procedure. In this article, to mitigate distractors in depth maps and highlight salient objects in RGB images, we propose a Hierarchical Alternate Interactions Network (HAINet) for RGB-D SOD. Specifically, HAINet consists of three key stages: feature encoding, cross-modal alternate interaction, and saliency reasoning. The main innovation in HAINet is the Hierarchical Alternate Interaction Module (HAIM), which plays a key role in the second stage for cross-modal feature interaction. HAIM first uses RGB features to filter distractors in depth features, and then the purified depth features are exploited to enhance RGB features in turn. The alternate RGB-depth-RGB interaction proceeds in a hierarchical manner, which progressively integrates local and global contexts within a single feature scale. In addition, we adopt a hybrid loss function to facilitate the training of HAINet. Extensive experiments on seven datasets demonstrate that our HAINet not only achieves competitive performance as compared with 19 relevant state-of-the-art methods, but also reaches a real-time processing speed of 43 fps on a single NVIDIA Titan X GPU. The code and results of our method are available at https://github.com/MathLee/HAINet. [ABSTRACT FROM AUTHOR]

With the supplement of texture and geometry cues in depth maps, some difficult scenes of salient object detection (SOD) in 2D images can be overcome. However, some distractors in the depth maps with relatively poor quality may interfere with SOD. Thus, how to suppress the interference of depth maps and extract valuable depth cues, is a critical issue to serve as effective complements to RGB cues. Aiming at addressing this issue, we propose a predict-refine scheme based Circular Complement Network (CCNet), which consists of a prediction subnetwork and a refinement subnetwork. On one hand, since RGB images generally contain more essential information for SOD, we propose a strategy which employs higher-level RGB feature maps to suppress the interference of depth feature maps. With this strategy, a novel Circular Feature Complement (CFC) module is specifically designed to enhance depth feature maps as well as to promote mutual complementarity between RGB feature maps and depth feature maps. The CFC modules are embedded into two subnetworks to achieve the cross-modal interactions at three levels. On the other hand, for the sake of the integration of two subnetworks, a Transmission Bridge (TB) module is proposed to effectively transfer the feature maps of the prediction subnetwork to the refinement subnetwork. The non-salient regions are thus further suppressed in the TB module. Comprehensive experiments on six benchmark datasets show that the proposed CCNet outperforms 13 state-of-the-art models. [ABSTRACT FROM AUTHOR]