Papers › Fully Convolutional Geometric Features
Fully Convolutional Geometric Features
Christopher Choy, Jaesik Park, Vladlen Koltun
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.
Code
Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.
Code Syntology ran Syntology
Not run by Syntology. Nothing on this page verifies that the listed code works.
Tasks
Results from the paper archive 2025-07-28
| Task | Dataset | Model | Metric | Value | Rank at snapshot | Leaderboard | Report |
|---|---|---|---|---|---|---|---|
| 3D Feature Matching | 3DMatch Benchmark | FCGF | Average Recall | 0.9578 | #1 of 1 | Archive leaderboard | report |
| Point Cloud Registration | 3DLoMatch (10-30% overlap) | FCGF (reported in PREDATOR) | Recall ( correspondence RMSE below 0.2) | 40.1 | #9 of 13 | Archive leaderboard | report |
| 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 of 11 | Archive leaderboard | report |
| Point Cloud Registration | 3DMatch (trained on KITTI) | FCGF | Recall | 0.325 | #4 of 5 | Archive leaderboard | report |
| Point Cloud Registration | 3DMatch Benchmark | FCGF + RANSAC | Feature Matching Recall | 85 | #11 of 15 | Archive leaderboard | report |
| Point Cloud Registration | ETH (trained on 3DMatch) | FCGF | Feature Matching Recall | 0.161 | #11 of 20 | Archive leaderboard | report |
| Point Cloud Registration | KITTI | FCGF | Success Rate | 96.57 | #5 of 6 | Archive leaderboard | report |
| Point Cloud Registration | KITTI (FCGF setting) | FCGF | Recall (0.6m, 5 degrees) | 98.2 | #3 of 11 | Archive leaderboard | report |
| Point Cloud Registration | KITTI (trained on 3DMatch) | FCGF | Success Rate | 24.19 | #14 of 14 | Archive leaderboard | report |
| Point Cloud Registration | RotKITTI Registration Benchmark | FCGF | RR@(1,0.1) | 3.6 | #6 of 6 | Archive leaderboard | report |
| Point Cloud Registration | RotKITTI Registration Benchmark | FCGF | RR@(1.5,0.3) | 11.6 | #6 of 6 | Archive leaderboard | report |
Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.
Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections