{"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/hybrid-knowledge-routed-modules-for-large","title":"Hybrid Knowledge Routed Modules for Large-scale Object Detection","arxiv_id":"1810.12681","date":"2018-10-30","proceeding":"NeurIPS 2018 12","authors":["Chenhan Jiang","Hang Xu","Xiangdan Liang","Liang Lin"],"abstract":"The dominant object detection approaches treat the recognition of each region\nseparately and overlook crucial semantic correlations between objects in one\nscene. This paradigm leads to substantial performance drop when facing heavy\nlong-tail problems, where very few samples are available for rare classes and\nplenty of confusing categories exists. We exploit diverse human commonsense\nknowledge for reasoning over large-scale object categories and reaching\nsemantic coherency within one image. Particularly, we present Hybrid Knowledge\nRouted Modules (HKRM) that incorporates the reasoning routed by two kinds of\nknowledge forms: an explicit knowledge module for structured constraints that\nare summarized with linguistic knowledge (e.g. shared attributes,\nrelationships) about concepts; and an implicit knowledge module that depicts\nsome implicit constraints (e.g. common spatial layouts). By functioning over a\nregion-to-region graph, both modules can be individualized and adapted to\ncoordinate with visual patterns in each image, guided by specific knowledge\nforms. HKRM are light-weight, general-purpose and extensible by easily\nincorporating multiple knowledge to endow any detection networks the ability of\nglobal semantic reasoning. Experiments on large-scale object detection\nbenchmarks show HKRM obtains around 34.5% improvement on VisualGenome (1000\ncategories) and 30.4% on ADE in terms of mAP. Codes and trained model can be\nfound in https://github.com/chanyn/HKRM.","url_abs":"http://arxiv.org/abs/1810.12681v1","url_pdf":"http://arxiv.org/pdf/1810.12681v1.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":"hybrid-knowledge-routed-modules-for-large","repo_url":"https://github.com/chanyn/HKRM","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"object","task_name":"Object"},{"task_slug":"object-detection","task_name":"Object Detection"},{"task_slug":"object-detection-1","task_name":"object-detection"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1810.12681","atlas_url":"https://app.syntology.ai/?focus=1810.12681","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}