Pose Partition Network pretrained model

TheproposedPPNisfavorablyfeaturedbylowcomplexityandhighaccuracyofjointdetectionandpartition.Inparticular,PPNperformsdenseregressionsfromglobaljointcandidateswithinaspecificembeddingspace,whichisparameterizedbycentroidsofpersons,toefficientlygeneraterobustpersondetectionandjointpartition