PRNet: Self-Supervised Learning for Partial-to-Partial Registration (2019-10-27T00:00:00.000000Z)

TL;DR

This work uses deep networks to tackle non-convexity of the alignment and partial correspondence problem in partial-to-partial point cloud registration, and shows PRNet predicts keypoints and correspondences consistently across views and objects.

Abstract

We present a simple, flexible, and general framework titled Partial Registration Network (PRNet), for partial-to-partial point cloud registration. Inspired by recently-proposed learning-based methods for registration, we use deep networks to tackle non-convexity of the alignment and partial correspondence problem. While previous learning-based methods assume the entire shape is visible, PRNet is suitable for partial-to-partial registration, outperforming PointNetLK, DCP, and non-learning methods on synthetic data. PRNet is self-supervised, jointly learning an appropriate geometric representation, a keypoint detector that finds points in common between partial views, and keypoint-to-keypoint correspondences. We show PRNet predicts keypoints and correspondences consistently across views and objects. Furthermore, the learned representation is transferable to classification.

Authors

Yue Wang

5 papers

J. Solomon

3 papers

TL;DR

Abstract

Authors

References73 items

Unsupervised Learning of Dense Shape Correspondence

Deep Closest Point: Learning Representations for Point Cloud Registration

KPConv: Flexible and Deformable Convolution for Point Clouds

A Polynomial-time Solution for Robust Registration with Extreme Outlier Rates

PointNetLK: Robust & Efficient Point Cloud Registration Using PointNet

Self-supervised Learning of Dense Shape Correspondence

Rethinking ImageNet Pre-Training

Deep part induction from articulated object pairs

Discovery of Latent 3D Keypoints via End-to-end Geometric Reasoning

BA-Net: Dense Bundle Adjustment Network

Audio-Visual Scene Analysis with Self-Supervised Multisensory Features

The Sound of Pixels

Point convolutional neural networks by extension operators

SPLATNet: Sparse Lattice Networks for Point Cloud Processing

Learning Latent Permutations with Gumbel-Sinkhorn Networks

Dynamic Graph CNN for Learning on Point Clouds

PointCNN: Convolution On X-Transformed Points

Reversible Harmonic Maps between Discrete Surfaces

SplineCNN: Fast Geometric Deep Learning with Continuous B-Spline Kernels

Learning SO(3) Equivariant Representations with Spherical CNNs

GWCNN: A Metric Alignment Layer for Deep Shape Analysis

Attention is All you Need

PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space

Deep Functional Maps: Structured Prediction for Dense Shape Correspondence

Deep Sets

PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation

Split-Brain Autoencoders: Unsupervised Learning by Cross-Channel Prediction

Geometric Deep Learning on Graphs and Manifolds Using Mixture Model CNNs

Geometric Deep Learning: Going beyond Euclidean data

Categorical Reparameterization with Gumbel-Softmax

SE-Sync: A certifiably correct algorithm for synchronization over the special Euclidean group

Cross-Modal Scene Networks

Fast Global Registration

Connecting Generative Adversarial Networks and Actor-Critic Methods

Collaborative 3D Reconstruction Using Heterogeneous UAVs: System and Experiments

Point Clouds Registration with Probabilistic Data Association

Semi-Supervised Classification with Graph Convolutional Networks

An Actor-Critic Algorithm for Sequence Prediction

Point registration via efficient convex relaxation

Entropic metric alignment for correspondence problems

Learning Aligned Cross-Modal Representations from Weakly Aligned Data

Go-ICP: A Globally Optimal Solution to 3D ICP Point-Set Registration

Volumetric and Multi-view CNNs for Object Classification on 3D Data

Colorful Image Colorization

RotationNet: Joint Object Categorization and Pose Estimation Using Multiviews from Unsupervised Viewpoints

Asynchronous Methods for Deep Reinforcement Learning

ShapeNet: An Information-Rich 3D Model Repository

Convolutional Networks on Graphs for Learning Molecular Fingerprints

VoxNet: A 3D Convolutional Neural Network for real-time object recognition

A Review of Point Cloud Registration Algorithms for Mobile Robotics

Multi-view Convolutional Neural Networks for 3D Shape Recognition

Adam: A Method for Stochastic Optimization

Sparse Iterative Closest Point

Estimating or Propagating Gradients Through Stochastic Neurons

Soft Maps Between Surfaces

Functional maps

Blended intrinsic maps

Möbius voting for surface correspondence

Generalized multidimensional scaling: A framework for isometry-invariant partial surface matching

Robust Registration of 2D and 3D Point Sets

Efficient variants of the ICP algorithm

Zippered polygon meshes from range images

A Method for Registration of 3-D Shapes

BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

Learning geometric operators on meshes

Globally Optimal Object Pose Estimation in Point Clouds with Mixed-Integer Programming

3 D ShapeNets : A Deep Representation for Volumetric Shape Modeling

Generalized-ICP

Probabilistic Matching for 3D Scan Registration

Eurographics Symposium on Geometry Processing 2008

Eurographics Symposium on Geometry Processing 2013 Dirichlet Energy for Analysis and Synthesis of Soft Maps

The Stanford 3d scanning repository

Freeman , and Michael Rubinstein . Looking to listen at the cocktail party : A speaker - independent audio - visual model for speech separation

Field of Study

Journal Information

Name