An Unofficial Implementation to Combination of HuMoR and PARE

background

HuMoR can recover a smooth human mesh from videos. The official code has implemented 3d point clouds, 2d keypoints pipeline. However, currently most of human pose estimation method can only output root-relatived results instead of 3d point clouds in real world.

Method

We use the root aligned coordinate as the optimization target to solve this problem. We use PARE as the estimator.

This code is highly dependent on PARE and HuMoR