本文探讨了通过2D姿态估计和3D运动捕捉数据推导三维人体姿态的方法，提出了基于虚拟标记的中间表示和Pose and Mesh Co-Evolution网络，成功解决了从视频中恢复3D人体运动的挑战。研究结果显示，该方法在多个基准数据集上优于现有技术，展现出强大的泛化能力。

Last Friday, I had first time played around in Mocap space. It is Amazing~! The body tracking is super accurate and steady, also the latency is low enough to provide high quality VR experience....