Relative Pose Estimation for RGB-D Human Input Scans via Implicit Function Reconstruction

Figure 1

Methods pipeline. SV is the abbreviation of single-view, and SV model means our single-view human reconstruction model; similarly, MV means multiview and MV model means our multiview reconstruction model. The proposed framework combines the initialization module and optimization module. The black lines represent the initialization flow, and the red lines are the optimization flow. Given the RGB-D input scans, the initialization module produces a coarse initial pose; then, the optimization module refines the relative pose iteratively.