3DFeat-Net: Weakly Supervised Local 3D Features for Point Cloud Registration

Abstract

In this paper, we propose the 3DFeat-Net which learns both 3D feature detector and descriptor for point cloud matching using weak supervision. Unlike many existing works, we do not require manual annotation of matching point clusters. Instead, we leverage on alignment and attention mechanisms to learn feature correspondences from GPS/INS tagged 3D point clouds without explicitly specifying them. We create training and benchmark outdoor Lidar datasets, and experiments show that 3DFeat-Net obtains state-of-the-art performance on these gravity-aligned datasets.

arXiv comment17 pages, 6 figures. Accepted in ECCV 2018

References in corpus (6)

• FaceNet: A Unified Embedding for Face Recognition and Clustering
• PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation
• TILDE: A Temporally Invariant Learned DEtector
• SegMatch: Segment based loop-closure for 3D point clouds
• LIFT: Learned Invariant Feature Transform
• PPFNet: Global Context Aware Local Features for Robust 3D Point Matching

Cited by in corpus (53)

• RPM-Net: Robust Point Matching using Learned Features
• Vision-based Robotic Grasping From Object Localization, Object Pose Estimation to Grasp Estimation for Parallel Grippers: A Review
• A comprehensive survey on point cloud registration
• Learning general and distinctive 3D local deep descriptors for point cloud registration
• R-PointHop: A Green, Accurate, and Unsupervised Point Cloud Registration Method
• DCL-SLAM: A Distributed Collaborative LiDAR SLAM Framework for a Robotic Swarm
• VRNet: Learning the Rectified Virtual Corresponding Points for 3D Point Cloud Registration
• PCRNet: Point Cloud Registration Network using PointNet Encoding
• A New Outlier Removal Strategy Based on Reliability of Correspondence Graph for Fast Point Cloud Registration
• Evaluating Local Geometric Feature Representations for 3D Rigid Data Matching
• When Deep Learning Meets Data Alignment: A Review on Deep Registration Networks (DRNs)
• 3D Point Cloud Processing and Learning for Autonomous Driving
• The Perfect Match: 3D Point Cloud Matching with Smoothed Densities
• RSKDD-Net: Random Sample-based Keypoint Detector and Descriptor
• D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features
• DeepCLR: Correspondence-Less Architecture for Deep End-to-End Point Cloud Registration
• Deep Learning for LiDAR Point Clouds in Autonomous Driving: A Review
• PCAN: 3D Attention Map Learning Using Contextual Information for Point Cloud Based Retrieval
• CAE-LO: LiDAR Odometry Leveraging Fully Unsupervised Convolutional Auto-Encoder for Interest Point Detection and Feature Description
• AlignNet-3D: Fast Point Cloud Registration of Partially Observed Objects
• USIP: Unsupervised Stable Interest Point Detection from 3D Point Clouds
• Learning a Task-specific Descriptor for Robust Matching of 3D Point Clouds
• PointDSC: Robust Point Cloud Registration using Deep Spatial Consistency
• SpinNet: Learning a General Surface Descriptor for 3D Point Cloud Registration
• SOE-Net: A Self-Attention and Orientation Encoding Network for Point Cloud based Place Recognition
• MOPS-Net: A Matrix Optimization-driven Network forTask-Oriented 3D Point Cloud Downsampling
• Correspondence-Free Point Cloud Registration with SO(3)-Equivariant Implicit Shape Representations
• One Framework to Register Them All: PointNet Encoding for Point Cloud Alignment
• Micro-Structures Graph-Based Point Cloud Registration for Balancing Efficiency and Accuracy
• DH3D: Deep Hierarchical 3D Descriptors for Robust Large-Scale 6DoF Relocalization
• Learning to Communicate and Correct Pose Errors
• PREDATOR: Registration of 3D Point Clouds with Low Overlap
• Learning Dense Features for Point Cloud Registration Using a Graph Attention Network
• StickyPillars: Robust and Efficient Feature Matching on Point Clouds using Graph Neural Networks
• PSF-LO: Parameterized Semantic Features Based Lidar Odometry
• DeepMapping: Unsupervised Map Estimation From Multiple Point Clouds
• DMLO: Deep Matching LiDAR Odometry
• Coupled Laplacian Eigenmaps for Locally-Aware 3D Rigid Point Cloud Matching
• PointLoc: Deep Pose Regressor for LiDAR Point Cloud Localization
• End-to-End Learning Local Multi-view Descriptors for 3D Point Clouds
• DeepI2P: Image-to-Point Cloud Registration via Deep Classification
• Fast Gravitational Approach for Rigid Point Set Registration with Ordinary Differential Equations
• Deep Magnification-Flexible Upsampling over 3D Point Clouds
• Multi-person 3D Pose Estimation in Crowded Scenes Based on Multi-View Geometry
• Learning to Fuse Local Geometric Features for 3D Rigid Data Matching
• Be your own Benchmark: No-Reference Trajectory Metric on Registered Point Clouds
• A New Framework for Registration of Semantic Point Clouds from Stereo and RGB-D Cameras
• Self-supervised Geometric Perception
• Robust Point Cloud Based Reconstruction of Large-Scale Outdoor Scenes
• Self-supervised Point Set Local Descriptors for Point Cloud Registration
• Registration Loss Learning for Deep Probabilistic Point Set Registration
• Consistent Two-Flow Network for Tele-Registration of Point Clouds
• Keypoint Autoencoders: Learning Interest Points of Semantics