Mix-and-Match Tuning for Self-Supervised Semantic Segmentation

作者：XiaohangZhan,ZiweiLiu,PingLuo,XiaoouTang,ChenChangeLoy摘要：Deepconvolutionalnetworksforsemanticimagesegmentationtypicallyrequirelarge-scalelabeleddata,e.g.ImageNetandMSCOCO,fornetworkpre-training.Toreduceannotationefforts,self-supervisedsemanticsegmentationisrecentlyproposedtopre-trainanetwor