Modal-Adaptive Gated Recoding Network for RGB-D Salient Object Detection

The multi-modal salient object detection model based on RGB-D information has better robustness in the real world. However, it remains nontrivial to adaptively balance effective feature fusion phase. In this letter, we propose a novel gated recoding network (GRNet) evaluate validity of two modes, and their influence. Our framework is divided into three phases: perception phase, mixing phase integration First, A encoder adopted extract multi-level single-modal features, which lays foundation for semantic comparative analysis. Then, modal-adaptive gate unit (MGU) proposed suppress invalid transfer modal features mixer hybrid branch decoder. responsible balanced information. Finally, decoder completes under guidance an optional edge stream (OEGS). Experiments analysis eight popular benchmarks verify that our performs favorably against 9 state-of-art methods.