A human pose estimation method based on semantic message passing
Aiming at the problem that the feature of adjacent joins can not be fully utilized in the research of human pose estimation based on convolutional neural network(CNN),this paper proposes a semantic message passing based human pose estimation method.The messages are passed among feature maps obtained by full CNN.A message passing strategy with reward and penalty mechanism according to the distance among joints is proposed.The method was validated on the leeds sports pose(LSP)and the frames labeled in cinema(FLIC),the experiment results indicated that this method had a good performance.
NETL
NSTL
万方数据
