{"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/3dmatch-learning-local-geometric-descriptors","title":"3DMatch: Learning Local Geometric Descriptors from RGB-D Reconstructions","arxiv_id":"1603.08182","date":"2016-03-27","proceeding":"CVPR 2017 7","authors":["Andy Zeng","Shuran Song","Matthias Nießner","Matthew Fisher","Jianxiong Xiao","Thomas Funkhouser"],"abstract":"Matching local geometric features on real-world depth images is a challenging\ntask due to the noisy, low-resolution, and incomplete nature of 3D scan data.\nThese difficulties limit the performance of current state-of-art methods, which\nare typically based on histograms over geometric properties. In this paper, we\npresent 3DMatch, a data-driven model that learns a local volumetric patch\ndescriptor for establishing correspondences between partial 3D data. To amass\ntraining data for our model, we propose a self-supervised feature learning\nmethod that leverages the millions of correspondence labels found in existing\nRGB-D reconstructions. Experiments show that our descriptor is not only able to\nmatch local geometry in new scenes for reconstruction, but also generalize to\ndifferent tasks and spatial scales (e.g. instance-level object model alignment\nfor the Amazon Picking Challenge, and mesh surface correspondence). Results\nshow that 3DMatch consistently outperforms other state-of-the-art approaches by\na significant margin. Code, data, benchmarks, and pre-trained models are\navailable online at http://3dmatch.cs.princeton.edu","url_abs":"http://arxiv.org/abs/1603.08182v3","url_pdf":"http://arxiv.org/pdf/1603.08182v3.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":"3dmatch-learning-local-geometric-descriptors","repo_url":"https://github.com/andyzeng/3dmatch-toolbox","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok","spdx":"BSD-2-Clause"}},{"paper_slug":"3dmatch-learning-local-geometric-descriptors","repo_url":"https://github.com/dengzhi-ustc/a-robust-registration-loss","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"3d-reconstruction","task_name":"3D Reconstruction"},{"task_slug":"point-cloud-registration","task_name":"Point Cloud Registration"}],"methods":[],"datasets_introduced":[{"slug":"3dmatch","name":"3DMatch","full_name":""}],"methods_introduced":[],"results":[{"leaderboard":"/sota/3d-reconstruction-on-scan2cad","task":"3D Reconstruction","dataset":"Scan2CAD","model":"3DMatch","rank_in_archive_order":2,"of":2,"metrics":{"Average Accuracy":"10.29%"},"uses_additional_data":false},{"leaderboard":"/sota/point-cloud-registration-on-3dmatch-benchmark","task":"Point Cloud Registration","dataset":"3DMatch Benchmark","model":"3DMatch + RANSAC","rank_in_archive_order":13,"of":15,"metrics":{"Feature Matching Recall":"66.8"},"uses_additional_data":false},{"leaderboard":"/sota/point-cloud-registration-on-eth-trained-on","task":"Point Cloud Registration","dataset":"ETH (trained on 3DMatch)","model":"3DMatch","rank_in_archive_order":10,"of":20,"metrics":{"Feature Matching Recall":"0.169"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1603.08182","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}