Group Collaborative Learning for Co-Salient Object Detection

We present a novel group collaborative learning framework (GCoNet) capable ofdetecting co-salient objects in real time (16ms), by simultaneously miningconsensus representations at group level based on the two necessary criteria:1) intra-group compactness to better formulate the consistency among co-salientobjects by capturing their inherent shared attributes using our novel groupaffinity module; 2) inter-group separability to effectively suppress theinfluence of noisy objects on the output by introducing our new groupcollaborating module conditioning the inconsistent consensus. To learn a betterembedding space without extra computational overhead, we explicitly employauxiliary classification supervision. Extensive experiments on threechallenging benchmarks, i.e., CoCA, CoSOD3k, and Cosal2015, demonstrate thatour simple GCoNet outperforms 10 cutting-edge models and achieves the newstate-of-the-art. We demonstrate this paper's new technical contributions on anumber of important downstream computer vision applications including contentaware co-segmentation, co-localization based automatic thumbnails, etc.