CMA-SOD: cross-modal attention fusion network for RGB-D salient object detection

Salient object detection (SOD) aims to identify and highlight the most visually prominent objects within an image. With the help of depth images, performance is pushed to the top in SOD. However, most of the current methods do not take into account capturing global information and modeling long-range dependencies of both modalities. Therefore, we introduce a novel cross-modal attention fusion network (CMA-SOD) for RGB-D salient object detection. By integrating Sequential Attention-Driven Integration, Multi-modal Feature Enhancement, and Cross-Attention Fusion Modules, CMA-SOD effectively suppresses redundant feature data and captures intra- and inter-modal saliency cues. Experimental results on seven benchmark datasets demonstrate the superiority of CMA-SOD, achieving state-of-the-art performance with mean absolute error (MAE) as low as 0.015 and F-measure up to 0.964 on NLPR dataset and achieving the best performance results on the challenging ReDWeb-S and COME15K datasets. Our method effectively handles complex scenarios and low-quality depth maps, highlighting its robustness and generalization capabilities. Our code can be found at this link: https://github.com/Corgislam/CMA-SOD/