Siamese Network for RGB-D Salient Object Detection and Beyond

Existing RGB-D salient object detection (SOD) models usually treat RGB and depth as independent information design separate networks for feature extraction from each. Such schemes can easily be constrained by a limited amount of training data or over-reliance on an elaborately designed process. Inspired the observation that modalities actually present certain commonality in distinguishing objects, novel joint learning densely cooperative fusion (JL-DCF) architecture is to learn both inputs through shared network backbone, known Siamese architecture. In this paper, we propose two effective components: (JL), (DCF). The JL module provides robust saliency exploiting cross-modal via network, while DCF introduced complementary discovery. Comprehensive experiments using 5 popular metrics show framework yields detector with good generalization. As result, JL-DCF significantly advances SOTAs average ~2.0% (F-measure) across 7 challenging datasets. addition, readily applicable other related multi-modal tasks, including RGB-T SOD video SOD, achieving comparable better performance.