596953.fig.005
Figure 5: We infer a pose of a model from relative positions and rotations computed from handle translations/rotations.