Workflow from Scientific Research

BodyPressureWnet (BPWnet), a deep network that learns a mapping from depth and gender to pose and contact pressure. Depth, D is encoded with a black-box CNN and outputs SMPL [2] human model parameters Ψ^, which is used to reconstruct a SMPL human mesh M^H. Using white-box image reconstruction components DMR and PMR, it refines the pose estimate and outputs a pressure map. The pressure map features are calibrated with CAL to produce a contact pressure estimate P^. The two module design refine estimates with initial and residual stages; subscripts 1 and 2 indicate estimates from each module, respectively.

DOI