A pytorch implementation of Google's paper <Towards Accurate Multi-person Pose Estimation in the Wild>
[[x1,y1,v1,x2,y2,v2,...],[...]]
where v = 1 (visible), v = 2 (not visible)