The Vitruvian manifold: Inferring dense correspondences for one-shot human pose estimation | Synapse