MambaSOD: Dual Mamba-driven cross-modal fusion network for RGB-D Salient Object Detection

The purpose of RGB-D Salient Object Detection (SOD) is to pinpoint the most visually conspicuous areas within images accurately. Numerous conventional models heavily rely on CNN and overlook the long-range contextual dependencies, subsequent transformer-based models have addressed the issue to some extent but introduce quadratic computational complexity. Moreover, incorporating spatial information from depth maps has been proven effective for this task and the primary challenge is how to effectively fuse the complementary information from RGB and depth. Recent advancements in Mamba, particularly its superior ability to perform long-range modeling within linear efficiency, have motivated our exploration of its potential in the RGB-D SOD task. In this paper, we propose a dual Mamba-driven cross-modal fusion network for RGB-D SOD, named MambaSOD, which effectively leverages Mamba's long-range dependency modeling capability. Specifically, we employ a dual Mamba-driven feature extractor to process RGB and depth inputs to obtain features with global contextual information. Then, we design a cross-modal fusion Mamba to perform modality-specific feature enhancement and model the inter-modal correlation between the RGB and depth features. To the best of our knowledge, this work is an innovative attempt to explore the potential of the pure Mamba in the RGB-D SOD task, offering a novel perspective. Numerous experiments conducted on seven prevailing datasets demonstrate our method's superiority over eighteen state-of-the-art RGB-D SOD models. The source code will be released at https://github.com/YueZhan721/MambaSOD.