{"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/dense-3d-regression-for-hand-pose-estimation","title":"Dense 3D Regression for Hand Pose Estimation","arxiv_id":"1711.08996","date":"2017-11-24","proceeding":"CVPR 2018 6","authors":["Chengde Wan","Thomas Probst","Luc van Gool","Angela Yao"],"abstract":"We present a simple and effective method for 3D hand pose estimation from a\nsingle depth frame. As opposed to previous state-of-the-art methods based on\nholistic 3D regression, our method works on dense pixel-wise estimation. This\nis achieved by careful design choices in pose parameterization, which leverages\nboth 2D and 3D properties of depth map. Specifically, we decompose the pose\nparameters into a set of per-pixel estimations, i.e., 2D heat maps, 3D heat\nmaps and unit 3D directional vector fields. The 2D/3D joint heat maps and 3D\njoint offsets are estimated via multi-task network cascades, which is trained\nend-to-end. The pixel-wise estimations can be directly translated into a vote\ncasting scheme. A variant of mean shift is then used to aggregate local votes\nwhile enforcing consensus between the the estimated 3D pose and the pixel-wise\n2D and 3D estimations by design. Our method is efficient and highly accurate.\nOn MSRA and NYU hand dataset, our method outperforms all previous\nstate-of-the-art approaches by a large margin. On the ICVL hand dataset, our\nmethod achieves similar accuracy compared to the currently proposed nearly\nsaturated result and outperforms various other proposed methods. Code is\navailable $\\href{\"https://github.com/melonwan/denseReg\"}{\\text{online}}$.","url_abs":"http://arxiv.org/abs/1711.08996v1","url_pdf":"http://arxiv.org/pdf/1711.08996v1.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":"dense-3d-regression-for-hand-pose-estimation","repo_url":"https://github.com/melonwan/denseReg","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"3d-hand-pose-estimation","task_name":"3D Hand Pose Estimation"},{"task_slug":"hand-pose-estimation","task_name":"Hand Pose Estimation"},{"task_slug":"pose-estimation","task_name":"Pose Estimation"},{"task_slug":"regression-1","task_name":"regression"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/hand-pose-estimation-on-icvl-hands","task":"Hand Pose Estimation","dataset":"ICVL Hands","model":"Dense Pixel-wise Estimation","rank_in_archive_order":12,"of":15,"metrics":{"Average 3D Error":"7.3"},"uses_additional_data":false},{"leaderboard":"/sota/hand-pose-estimation-on-msra-hands","task":"Hand Pose Estimation","dataset":"MSRA Hands","model":"Dense Pixel-wise Estimation","rank_in_archive_order":4,"of":11,"metrics":{"Average 3D Error":"7.2"},"uses_additional_data":false},{"leaderboard":"/sota/hand-pose-estimation-on-nyu-hands","task":"Hand Pose Estimation","dataset":"NYU Hands","model":"Dense Pixel-wise Estimation","rank_in_archive_order":13,"of":17,"metrics":{"Average 3D Error":"10.2"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1711.08996","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}