Self-Supervised Learning of 3D Human Pose using Multi-view Geometry

CVPR 2019 Muhammed KocabasSalih KaragozEmre Akbas

Training accurate 3D human pose estimators requires large amount of 3D ground-truth data which is costly to collect. Various weakly or self supervised pose estimation methods have been proposed due to lack of 3D data... (read more)

PDF Abstract

Evaluation results from the paper

Task Dataset Model Metric name Metric value Global rank Compare
3D Human Pose Estimation Human3.6M Fully supervised EpipolarPose Average MPJPE (mm) 51.8 # 10
3D Human Pose Estimation Human3.6M Self supervised EpipolarPose Average MPJPE (mm) 55.0 # 13