Building High-Fidelity Human Body Models From User-Generated Data

We propose a key point-based approach, refers to as KPhub-PC , estimate high-fidelity human body models from low-quality point clouds acquired with an affordable 3D scanner and variation xmlns:xlink="http://www.w3.org/1999/xlink">KPhub-I that can achieve the same purpose based on low-resolution single images taken by smartphones. In KPhub-PC, sparse set of points is annotated guide deformation parametric model SMPL then explain target built. Besides building clouds, KPhub-I designed accurate 2D images. The fitted joints boundary which are detected using CNN methods automatically. Considering people in stable poses most time, pose prior defined CMU motion capture dataset for further improving accuracy. Extensive experiments demonstrate both types user-generated data, proposed approaches build believable animatable robustly. Our approach outperforms state-of-the-arts accuracy shape estimation.