Real-Time ESFP: Estimating, Smoothing, Filtering, and Pose-Mapping

A. SMPL: A Skinned Multi-Person Linear Model The SMPL model ( Skinned Multi-Person Linear model) is a widely adopted statistical representation of the human body that combines a low-dimensional parameter space with linear blend skinning (LBS) to generate realistic, fully differentiable 3-D meshes. It underpins many state-of-the-art pipelines for monocular pose estimation, motion capture, and animation because it offers three essential properties: a compact pose-shape space learned from thousands of laser scans, an articulated skeletal structure compatible with traditional skinning, and analytic gradients with respect to both pose and shape parameters [6].