Meshlet Priors for 3D Mesh Reconstruction

TL;DR

This paper introduces meshlets, small mesh patches that serve as local shape priors, enabling robust 3D mesh reconstruction from sparse, noisy data across various object classes and poses.

Contribution

The authors propose meshlets as local priors, overcoming class-specific limitations of previous learned priors and improving generalization in 3D mesh reconstruction.

Findings

Meshlets enable pose-invariant, class-agnostic mesh reconstruction.

Learned local priors improve reconstruction quality with noisy, sparse data.

The approach generalizes well to unseen object classes.

Abstract

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 undesirable trade-off between attenuating noise and preserving local detail. Recent deep-learning approaches produce impressive results by learning priors directly from the data. However, the priors are learned at the object level, which makes these algorithms class-specific and even sensitive to the pose of the object. We introduce meshlets, small patches of mesh that we use to learn local shape priors. Meshlets act as a dictionary of local features and thus allow to use learned priors to reconstruct object meshes in any pose and from unseen classes, even when the noise is large and the samples sparse.