Fully Convolutional Geometric Features

Extracting geometric features from 3D scans or point clouds is the first step in applications such as registration, reconstruction, and tracking. State-of-the-art methods require computing low-level features as input or extracting patch-based features with limited receptive field. In this work, we present fully-convolutional geometric features, computed in a single pass by a 3D fully-convolutional network. We also present new metric learning losses that dramatically improve performance. Fully-convolutional geometric features are compact, capture broad spatial context, and scale to large scenes. We experimentally validate our approach on both indoor and outdoor datasets. Fully-convolutional geometric features achieve state-of-the-art accuracy without requiring prepossessing, are compact (32 dimensions), and are 600 times faster than the most accurate prior method.

PDF Abstract International Conference 2019 PDF International Conference 2019 Abstract

Datasets


Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Benchmark
Point Cloud Registration 3DLoMatch (10-30% overlap) FCGF (reported in PREDATOR) Recall ( correspondence RMSE below 0.2) 40.1 # 8
Point Cloud Registration 3DMatch (at least 30% overlapped - sample 5k interest points) FCGF (reported in PREDATOR) Recall ( correspondence RMSE below 0.2) 85.1 # 7
Point Cloud Registration 3DMatch Benchmark FCGF + RANSAC Feature Matching Recall 82 # 11
3D Feature Matching 3DMatch Benchmark FCGF Average Recall 0.9578 # 1
Point Cloud Registration 3DMatch (trained on KITTI) FCGF Recall 0.325 # 4
Point Cloud Registration ETH (trained on 3DMatch) FCGF Recall 0.161 # 11
Point Cloud Registration KITTI FCGF Success Rate 96.57 # 5
Point Cloud Registration KITTI (trained on 3DMatch) FCGF Success Rate 24.19 # 6

Results from Other Papers


Task Dataset Model Metric Name Metric Value Rank Source Paper Compare
Point Cloud Registration KITTI (FCGF setting) FCGF Recall (0.6m, 5 degrees) 98.2 # 2

Methods


No methods listed for this paper. Add relevant methods here