{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/fully-convolutional-geometric-features","title":"Fully Convolutional Geometric Features","arxiv_id":null,"date":"2019-10-27","proceeding":"International Conference on Computer vision 2019 10","authors":["Christopher Choy","Jaesik Park","Vladlen Koltun"],"abstract":"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.","url_abs":"https://github.com/chrischoy/FCGF","url_pdf":"https://node1.chrischoy.org/data/publications/fcgf/fcgf.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"fully-convolutional-geometric-features","repo_url":"https://github.com/chrischoy/FCGF","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"3d-feature-matching","task_name":"3D Feature Matching"},{"task_slug":"3d-point-cloud-matching","task_name":"3D Point Cloud Matching"},{"task_slug":"3d-shape-representation","task_name":"3D Shape Representation"},{"task_slug":"metric-learning","task_name":"Metric Learning"},{"task_slug":"point-cloud-registration","task_name":"Point Cloud Registration"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/3d-feature-matching-on-3dmatch-benchmark","task":"3D Feature Matching","dataset":"3DMatch Benchmark","model":"FCGF","rank_in_archive_order":1,"of":1,"metrics":{"Average Recall":"0.9578"},"uses_additional_data":false},{"leaderboard":"/sota/point-cloud-registration-on-3dlomatch-10-30","task":"Point Cloud Registration","dataset":"3DLoMatch (10-30% overlap)","model":"FCGF (reported in PREDATOR)","rank_in_archive_order":9,"of":13,"metrics":{"Recall ( correspondence RMSE below 0.2)":"40.1"},"uses_additional_data":false},{"leaderboard":"/sota/point-cloud-registration-on-3dmatch-at-least-2","task":"Point Cloud Registration","dataset":"3DMatch (at least 30% overlapped - sample 5k interest points)","model":"FCGF (reported in PREDATOR)","rank_in_archive_order":7,"of":11,"metrics":{"Recall ( correspondence RMSE below 0.2)":"85.1"},"uses_additional_data":false},{"leaderboard":"/sota/point-cloud-registration-on-3dmatch-trained","task":"Point Cloud Registration","dataset":"3DMatch (trained on KITTI)","model":"FCGF","rank_in_archive_order":4,"of":5,"metrics":{"Recall":"0.325"},"uses_additional_data":false},{"leaderboard":"/sota/point-cloud-registration-on-3dmatch-benchmark","task":"Point Cloud Registration","dataset":"3DMatch Benchmark","model":"FCGF + RANSAC","rank_in_archive_order":11,"of":15,"metrics":{"Feature Matching Recall":"85"},"uses_additional_data":false},{"leaderboard":"/sota/point-cloud-registration-on-eth-trained-on","task":"Point Cloud Registration","dataset":"ETH (trained on 3DMatch)","model":"FCGF","rank_in_archive_order":11,"of":20,"metrics":{"Feature Matching Recall":"0.161"},"uses_additional_data":false},{"leaderboard":"/sota/point-cloud-registration-on-kitti","task":"Point Cloud Registration","dataset":"KITTI","model":"FCGF","rank_in_archive_order":5,"of":6,"metrics":{"Success Rate":"96.57"},"uses_additional_data":false},{"leaderboard":"/sota/point-cloud-registration-on-kitti-fcgf","task":"Point Cloud Registration","dataset":"KITTI (FCGF setting)","model":"FCGF","rank_in_archive_order":3,"of":11,"metrics":{"Recall (0.6m, 5 degrees)":"98.2"},"uses_additional_data":false},{"leaderboard":"/sota/point-cloud-registration-on-kitti-trained-on","task":"Point Cloud Registration","dataset":"KITTI (trained on 3DMatch)","model":"FCGF","rank_in_archive_order":14,"of":14,"metrics":{"Success Rate":"24.19"},"uses_additional_data":false},{"leaderboard":"/sota/point-cloud-registration-on-rotkitti","task":"Point Cloud Registration","dataset":"RotKITTI Registration Benchmark","model":"FCGF","rank_in_archive_order":6,"of":6,"metrics":{"RR@(1,0.1)":"3.6","RR@(1.5,0.3)":"11.6"},"uses_additional_data":false}],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}