Scribble-based complementary graph reasoning network for weakly supervised salient object detection

Shuang Liang, Zhiqi Yan, Chi Xie, Hongming Zhu, Jiewen Wang

Abstract: Highlights•A novel complementary graph reasoning network (CGRNet) and cross graph interaction structure (GCU) is proposed, which can capture long-distance dependencies between pixels, making it suitable for energy propagation in weakly supervised learning.•Our proposed EOM and DFS extract and aggregate multi-source information in a simple and effective way, improving the boundary accuracy of the model.•Experimental results show that our proposed method achieves state-of-the-art performance on five challenging datasets.