Quality-Driven Dual-Branch Feature Integration Network for Video Salient Object Detection

Video salient object detection has attracted growing interest in recent years. However, some existing video saliency models often suffer from the inappropriate utilization of spatial and temporal cues insufficient aggregation different level features, leading to remarkable performance degradation. Therefore, we propose a quality-driven dual-branch feature integration network majoring adaptive fusion multi-modal sufficient multi-level spatiotemporal features. Firstly, employ (QMFF) module combine Particularly, quality scores estimated each level’s are not only used weigh two modal features but also adaptively integrate coarse predictions into guidance map, which further enhances Secondly, deploy dual-branch-based (DMFA) where branches including progressive decoder branch direct concatenation sufficiently explore cooperation In particular, order provide an for outputs branches, design (DF) unit, channel weight output can be learned jointly outputs. The experiments conducted on four datasets clearly demonstrate effectiveness superiority our model against state-of-the-art models.