Text this: Estimating a 3D Human Skeleton from a Single RGB Image by Fusing Predicted Depths from Multiple Virtual Viewpoints.