Point2Mesh: A Self-Prior for Deformable Meshes

Abstract

In this paper, we introduce Point2Mesh, a technique for reconstructing a surface mesh from an input point cloud. Instead of explicitly specifying a prior that encodes the expected shape properties, the prior is defined automatically using the input point cloud, which we refer to as a self-prior. The self-prior encapsulates reoccurring geometric repetitions from a single shape within the weights of a deep neural network. We optimize the network weights to deform an initial mesh to shrink-wrap a single input point cloud. This explicitly considers the entire reconstructed shape, since shared local kernels are calculated to fit the overall object. The convolutional kernels are optimized globally across the entire shape, which inherently encourages local-scale geometric self-similarity across the shape surface. We show that shrink-wrapping a point cloud with a self-prior converges to a desirable solution; compared to a prescribed smoothness prior, which often becomes trapped in undesirable local minima. While the performance of traditional reconstruction approaches degrades in non-ideal conditions that are often present in real world scanning, i.e., unoriented normals, noise and missing (low density) parts, Point2Mesh is robust to non-ideal conditions. We demonstrate the performance of Point2Mesh on a large variety of shapes with varying complexity.

SIGGRAPH 2020; Project page: https://ranahanocka.github.io/point2mesh/

References in corpus (4)

• Geometric deep learning: going beyond Euclidean data
• The Perception-Distortion Tradeoff
• MeshCNN: A Network with an Edge
• Non-Stationary Texture Synthesis by Adversarial Expansion

Cited by in corpus (23)

• Neural Dual Contouring
• Flexible Isosurface Extraction for Gradient-Based Mesh Optimization
• Learning Skeletal Articulations with Neural Blend Shapes
• Iterative Poisson Surface Reconstruction (iPSR) for Unoriented Points
• Orienting Point Clouds with Dipole Propagation
• Learning to Generate 3D Shapes from a Single Example
• Neural-IMLS: Self-supervised Implicit Moving Least-Squares Network for Surface Reconstruction
• Point Cloud Diffusion Models for Automatic Implant Generation
• Geometric Prior Based Deep Human Point Cloud Geometry Compression
• Scalable Surface Reconstruction with Delaunay-Graph Neural Networks
• SpaceMesh: A Continuous Representation for Learning Manifold Surface Meshes
• SPARF: Large-Scale Learning of 3D Sparse Radiance Fields from Few Input Images
• TetWeave: Isosurface Extraction using On-The-Fly Delaunay Tetrahedral Grids for Gradient-Based Mesh Optimization
• Mesh Density Adaptation for Template-based Shape Reconstruction
• Learning the Geodesic Embedding with Graph Neural Networks
• ANISE: Assembly-based Neural Implicit Surface rEconstruction
• Strokes2Surface: Recovering Curve Networks From 4D Architectural Design Sketches
• GeoCode: Interpretable Shape Programs
• NeuralMeshing: Differentiable Meshing of Implicit Neural Representations
• Learning Self-Prior for Mesh Inpainting Using Self-Supervised Graph Convolutional Networks
• Point2Quad: Generating Quad Meshes from Point Clouds via Face Prediction
• DeMapGS: Simultaneous Mesh Deformation and Surface Attribute Mapping via Gaussian Splatting
• Incomplete Gamma Kernels: Generalizing Locally Optimal Projection Operators