{"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/graphonomy-universal-human-parsing-via-graph","title":"Graphonomy: Universal Human Parsing via Graph Transfer Learning","arxiv_id":"1904.04536","date":"2019-04-09","proceeding":"CVPR 2019 6","authors":["Ke Gong","Yiming Gao","Xiaodan Liang","Xiaohui Shen","Meng Wang","Liang Lin"],"abstract":"Prior highly-tuned human parsing models tend to fit towards each dataset in a\nspecific domain or with discrepant label granularity, and can hardly be adapted\nto other human parsing tasks without extensive re-training. In this paper, we\naim to learn a single universal human parsing model that can tackle all kinds\nof human parsing needs by unifying label annotations from different domains or\nat various levels of granularity. This poses many fundamental learning\nchallenges, e.g. discovering underlying semantic structures among different\nlabel granularity, performing proper transfer learning across different image\ndomains, and identifying and utilizing label redundancies across related tasks.\n  To address these challenges, we propose a new universal human parsing agent,\nnamed \"Graphonomy\", which incorporates hierarchical graph transfer learning\nupon the conventional parsing network to encode the underlying label semantic\nstructures and propagate relevant semantic information. In particular,\nGraphonomy first learns and propagates compact high-level graph representation\namong the labels within one dataset via Intra-Graph Reasoning, and then\ntransfers semantic information across multiple datasets via Inter-Graph\nTransfer. Various graph transfer dependencies (\\eg, similarity, linguistic\nknowledge) between different datasets are analyzed and encoded to enhance graph\ntransfer capability. By distilling universal semantic graph representation to\neach specific task, Graphonomy is able to predict all levels of parsing labels\nin one system without piling up the complexity. Experimental results show\nGraphonomy effectively achieves the state-of-the-art results on three human\nparsing benchmarks as well as advantageous universal human parsing performance.","url_abs":"http://arxiv.org/abs/1904.04536v1","url_pdf":"http://arxiv.org/pdf/1904.04536v1.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":"graphonomy-universal-human-parsing-via-graph","repo_url":"https://github.com/Gaoyiminggithub/Graphonomy","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"human-parsing","task_name":"Human Parsing"},{"task_slug":"transfer-learning","task_name":"Transfer Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1904.04536","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}