Fully Convolutional Geometric Features

By Christopher Bongsoo Choy, Jaesik Park, Vladlen Koltun. IEEE International Conference on Computer Vision, 2019

843 citations recorded by the source corpus.

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 290 times faster than the most accurate prior method.

Subfields: Aerospace Engineering, Computational Mechanics, Geology

View Paper: https://www.semanticscholar.org/paper/985e7d47d3d41753900ddffb1af585221bf3d8ec

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Computational Mechanics

In Geology

Citations in the corpus

Deep Global Registration (2020, methodology): “Recall TE (cm) RE (deg) Time (s) FGR [50] 0.2% 40.7 1.02 1.42 RANSAC [35] 34.2% 25.9 1.39 1.37 FCGF [9] 98.2%”
PREDATOR: Registration of 3D Point Clouds with Low Overlap (2020, methodology): “Following the literature [49, 17, 8], we also report Feature Match Recall (FMR), defined as the fraction of pairs that have >5% ”inlier” matches with <10 cm residual under the ground truth transformation (without checking if the transformation can be recovered from those matches), and Inlier Ratio (”
Geometric Transformer for Fast and Robust Point Cloud Registration (2022, methodology): “Most correspondence-based methods rely on keypoint detection [1,4,7,15].”
CoFiNet: Reliable Coarse-to-fine Correspondences for Robust Point Cloud Registration (2021, methodology): “To address this, FCGF [31] makes the first attempt by using sparse convolutions [32] to compute dense descriptors of the whole point cloud in a single pass, which leads to 600x speed-up while still being able to achieve comparable performance to patch-based methods.”
D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features (2020, methodology): “However, as also observed in FCGF [2], we find that a fully convolutional network (e.”
ASLFeat: Learning Local Features of Accurate Shape and Localization (2020, methodology): “Instead of using the hardest-triplet loss in D2-Net [8], we adopt the hardest-contrastive form in FCGF [3], which we ﬁnd guarantee better convergence when training from scratch and equipping DCN, written as: where D ( · , · ) denotes the Euclidean distance measured be-tween two descriptors, and m p…”
A comprehensive survey on point cloud registration (2021, methodology): “To improve the efﬁciency of volumetric-based descriptor, FCGF [17] uses 1 × 1 × 1 kernel to extract a fast and compact metric features for geometric correspondence.”
PointDSC: Robust Point Cloud Registration using Deep Spatial Consistency (2021, methodology): “Although 3D local features [4, 39, 18, 27, 32] have evolved rapidly, correspondences produced by feature matching are still prone to outliers, especially when the overlap of scene fragments is small.”
PointContrast: Unsupervised Pre-training for 3D Point Cloud Understanding (2020, methodology): “In 2D vision it has led to a surge of interest in finding optimal pretext unsupervised tasks [43, 83, 84, 14, 41, 18, 5, 42, 3, 40, 27, 69, 28, 87, 8, 10].”
SpinNet: Learning a General Surface Descriptor for 3D Point Cloud Registration (2020, methodology): “Note that, several baselines [8, 2] require rotation-based data augmentation for training, whilst ours does not.”
A Metric Learning Reality Check (2020, background): “N o rm al iz e d S o ft m ax [ 7 2 ]”
TEASER: Fast and Certifiable Point Cloud Registration (2020, methodology): “Deep global registration [93] uses FCGF feature descriptors for point cloud registration.”