{"id":44034,"date":"2026-11-13T11:03:09","date_gmt":"2026-11-13T11:03:09","guid":{"rendered":"https:\/\/lamarr-institute.org\/publication\/register-any-point-scaling-3d-point-cloud-registration-by-flow-matching\/"},"modified":"2026-11-13T11:03:09","modified_gmt":"2026-11-13T11:03:09","slug":"register-any-point-scaling-3d-point-cloud-registration-by-flow-matching","status":"publish","type":"publication","link":"https:\/\/lamarr-institute.org\/de\/publication\/register-any-point-scaling-3d-point-cloud-registration-by-flow-matching\/","title":{"rendered":"Register Any Point: Scaling 3D Point Cloud Registration by\u00a0Flow Matching"},"content":{"rendered":"<p>Point cloud registration aligns multiple unposed\u00a0point clouds into a common reference frame and is a core step for\u00a03D reconstruction and robot localization when no initial pose guess\u00a0is available. In this work, we cast point cloud registration\u00a0as conditional generation: a learned, continuous point-wise velocity field transports noisy points to a registered scene, from which\u00a0the pose of each view is recovered. Unlike prior methods that perform correspondence matching to estimate pairwise transformations\u00a0and then optimize a pose graph for multi-view registration, our\u00a0model directly generates the registered point cloud, yielding\u00a0both efficiency and point-level global consistency. By scaling\u00a0the training data and conducting test-time rigidity enforcement,\u00a0our approach achieves state-of-the-art average performance on existing pairwise registration benchmarks and on our proposed cross-domain multi-view registration benchmark. The superior zero-shot performance on this benchmark demonstrates that our method generalizes across view counts, scene scales, and sensor modalities even with low overlap.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Point cloud registration aligns multiple unposed\u00a0point clouds into a common reference frame and is a core step for\u00a03D reconstruction and robot localization when no initial pose guess\u00a0is available. In this work, we cast point cloud registration\u00a0as conditional generation: a learned, continuous point-wise velocity field transports noisy points to a registered scene, from which\u00a0the pose of each view is recovered. Unlike prior methods that perform correspondence matching to estimate pairwise transformations\u00a0and [&hellip;]<\/p>\n","protected":false},"author":14,"featured_media":0,"template":"","meta":{"_acf_changed":false,"footnotes":""},"publication-type":[32],"publication-year":[1680],"class_list":["post-44034","publication","type-publication","status-publish","hentry","publication-type-inproceedings","publication-year-1680"],"acf":[],"publishpress_future_workflow_manual_trigger":{"enabledWorkflows":[]},"_links":{"self":[{"href":"https:\/\/lamarr-institute.org\/de\/wp-json\/wp\/v2\/publication\/44034","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/lamarr-institute.org\/de\/wp-json\/wp\/v2\/publication"}],"about":[{"href":"https:\/\/lamarr-institute.org\/de\/wp-json\/wp\/v2\/types\/publication"}],"author":[{"embeddable":true,"href":"https:\/\/lamarr-institute.org\/de\/wp-json\/wp\/v2\/users\/14"}],"version-history":[{"count":0,"href":"https:\/\/lamarr-institute.org\/de\/wp-json\/wp\/v2\/publication\/44034\/revisions"}],"wp:attachment":[{"href":"https:\/\/lamarr-institute.org\/de\/wp-json\/wp\/v2\/media?parent=44034"}],"wp:term":[{"taxonomy":"publication-type","embeddable":true,"href":"https:\/\/lamarr-institute.org\/de\/wp-json\/wp\/v2\/publication-type?post=44034"},{"taxonomy":"publication-year","embeddable":true,"href":"https:\/\/lamarr-institute.org\/de\/wp-json\/wp\/v2\/publication-year?post=44034"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}