COCOisalarge-scaleobjectdetection,segmentation,andcaptioningdataset.COCOhasseveralfeatures:ObjectsegmentationRecognitionincontextSuperpixelstuffsegmentation330Kimages(>200Klabeled)1.5millionobjectinstances80objectcategories91stuffcategories5captionsperimage250,000peoplewithkeypoints