Estimating a mesh from an unordered set of sparse, noisy 3d points is a challenging problem that requires carefully selected priors. Existing hand-crafted priors, such as smoothness regularizers, impose an undesi