Regularized Densely-Connected Pyramid Network for Salient Instance Segmentation

Much of the recent efforts on salient object detection (SOD) have been devoted to producing accurate saliency maps without being aware their instance labels. To this end, we propose a new pipeline for end-to-end segmentation (SIS) that predicts class-agnostic mask each detected instance. better use rich feature hierarchies in deep networks and enhance side predictions, regularized dense connections, which attentively promote informative features suppress non-informative ones from all pyramids. A novel multi-level RoIAlign based decoder is introduced adaptively aggregate predictions. Such strategies can be well-encapsulated into Mask R-CNN pipeline. Extensive experiments popular benchmarks demonstrate our design significantly outperforms existing state-of-the-art competitors by 6.3% (58.6% vs. 52.3%) terms AP metric. The code available at https://github.com/yuhuan-wu/RDPNet.