We present the first fully convolutional end-to-end solutionfor instance-aware semantic segmentation task. It inheritsall the merits of FCNs for semantic segmentation [29]and instance mask proposal [5]. It detects and segmentsthe object instances jointly and simultanoulsy. By the introductionof position-senstive inside/outside score maps,the underlying convolutional representation is fully sharedbetween the two sub-tasks, as well as between all regionsof interest. The proposed network is highly integrated andachieves state-of-the-art performance in both accuracy andefficiency. It wins the COCO 2016 segmentation competitionby a large margin.