Efficient Human Pose Estimation from Single Depth Images

微软研究院的最新大作。Wedescribetwonewapproachestohumanposeestimation.Bothcanquicklyandaccuratelypredictthe3Dpositionsofbodyjointsfromasingledepthimage,withoutusinganytemporalinformation.Thekeytobothapproachesistheuseofalarge,realistic,andhighlyvariedsyntheticsetoftrainingimages.Thisallowsustolearnmodelsthat