A Novel Edge-Inspired Depth Quality Evaluation Network for RGB-D Salient Object Detection

Recently, the pair of RGB images and depth images, which is denoted as RGB-D images, are introduced to improve their performances of Salient Object Detection(SOD), because one of the pairs may stand out at least in one modality. However, most existing State-Of-The-Art(SOTA) methods still suffer from dilemma. Firstly, the edges of predicted salient object are blurry. Secondly, how to integrate RGB-D images effectively still needs be explored. Thirdly, the quality of depth images have a strong impact on the performance of SOD so that the selection of depth images is worthy of exploring. To address the problems, in this paper, we propose a Edge-Inspired Depth Quality Evaluation Network(EDQNet), which evaluates the quality of the depth images based on the edge of ground truth, the edge of depth images and the predicted edge. More specifically, the Depth Quality Evaluation Module(DQEM) includes two parts: the depth decider and the depth aggregator. The former judges the quality of the depth images while the latter produces the weighted depth features. Then, the edge Detection Module(EDM) is proposed to predict the edge of salient object, producing the edge features. In addition, features from VGG backbone, edge features and the depth features are integrated by our Multi-Modality Feature Fusion(MMF) by using a series of composition of Hybrid Dilated Convolution(HDC). Moreover, the integrated features are fused by the Three-feature Interactive Module(TIM), Double-feature Interactive Module(DIM) and deconvolution to predict the final salient map. Our experiments on four RGB-D datasets demonstrate that our proposed EDQNet outperforms previous SOTA RGB-D SOD.