WBNet: Weakly-supervised salient object detection via scribble and pseudo-background priors

Yi Wang, Ruili Wang, Xiangjian He, Chi Lin, Tianzhu Wang, Qi Jia, Xin Fan

Abstract: Highlights•A multi-source weakly supervised SOD network by scribble and pseudo-background labels.•Transformer-based network with feature aggregation and an auxiliary edge detector.•Salient pseudo-mask generator using multi-source self-learning features.•Points- and box-prompted pseudo-masks by Segment Anything Model (SAM).•The proposed method outperforms state-of-the-art methods by a large margin.