Rethinking on Multi-Stage Networks for Human Pose Estimation

Authors:
  • WLWenbo Li
    Wenbo Li
  • ZWZhicheng Wang
    Zhicheng Wang
  • BYBinyi Yin
    Binyi Yin
  • QPQixiang Peng
    Qixiang Peng
  • YDYuming Du
    Yuming Du
  • TXTianzi Xiao
    Tianzi Xiao
  • GYGang Yu
    Gang Yu
  • HLHongtao Lu
    Hongtao Lu
  • YWYichen Wei
    Yichen Wei
  • JSJian Sun
    Jian Sun

Abstract
Existing pose estimation approaches can be categorized into single-stage and multi-stage methods. While a multi-stage architecture is seemingly more suitable for the task, the performance of current multi-stage methods is not as competitive as single-stage ones. This work studies this issue. We argue that the current unsatisfactory performance comes from various insufficient design in current methods. We propose several improvements on the architecture design, feature flow, and loss function. The resulting multi-stage network outperforms all previous works and obtains the best performance on COCO keypoint challenge 2018. The source code will be released.