{"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/3d-human-pose-machines-with-self-supervised","title":"3D Human Pose Machines with Self-supervised Learning","arxiv_id":"1901.03798","date":"2019-01-12","proceeding":"arXiv.org 2019 1","authors":["Keze Wang","Liang Lin","Chenhan Jiang","Chen Qian","Pengxu Wei"],"abstract":"Driven by recent computer vision and robotic applications, recovering 3D\nhuman poses has become increasingly important and attracted growing interests.\nIn fact, completing this task is quite challenging due to the diverse\nappearances, viewpoints, occlusions and inherently geometric ambiguities inside\nmonocular images. Most of the existing methods focus on designing some\nelaborate priors /constraints to directly regress 3D human poses based on the\ncorresponding 2D human pose-aware features or 2D pose predictions. However, due\nto the insufficient 3D pose data for training and the domain gap between 2D\nspace and 3D space, these methods have limited scalabilities for all practical\nscenarios (e.g., outdoor scene). Attempt to address this issue, this paper\nproposes a simple yet effective self-supervised correction mechanism to learn\nall intrinsic structures of human poses from abundant images. Specifically, the\nproposed mechanism involves two dual learning tasks, i.e., the 2D-to-3D pose\ntransformation and 3D-to-2D pose projection, to serve as a bridge between 3D\nand 2D human poses in a type of \"free\" self-supervision for accurate 3D human\npose estimation. The 2D-to-3D pose implies to sequentially regress intermediate\n3D poses by transforming the pose representation from the 2D domain to the 3D\ndomain under the sequence-dependent temporal context, while the 3D-to-2D pose\nprojection contributes to refining the intermediate 3D poses by maintaining\ngeometric consistency between the 2D projections of 3D poses and the estimated\n2D poses. We further apply our self-supervised correction mechanism to develop\na 3D human pose machine, which jointly integrates the 2D spatial relationship,\ntemporal smoothness of predictions and 3D geometric knowledge. Extensive\nevaluations demonstrate the superior performance and efficiency of our\nframework over all the compared competing methods.","url_abs":"http://arxiv.org/abs/1901.03798v2","url_pdf":"http://arxiv.org/pdf/1901.03798v2.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":"3d-human-pose-machines-with-self-supervised","repo_url":"https://github.com/Khenu/Computer-Animation","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null},{"paper_slug":"3d-human-pose-machines-with-self-supervised","repo_url":"https://github.com/chanyn/3Dpose_ssl","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}}],"tasks":[{"task_slug":"3d-human-pose-estimation","task_name":"3D Human Pose Estimation"},{"task_slug":"pose-estimation","task_name":"Pose Estimation"},{"task_slug":"self-supervised-learning","task_name":"Self-Supervised Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1901.03798","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}