{"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/composition-loss-for-counting-density-map","title":"Composition Loss for Counting, Density Map Estimation and Localization in Dense Crowds","arxiv_id":"1808.01050","date":"2018-08-02","proceeding":"ECCV 2018 9","authors":["Haroon Idrees","Muhmmad Tayyab","Kishan Athrey","Dong Zhang","Somaya Al-Maadeed","Nasir Rajpoot","Mubarak Shah"],"abstract":"With multiple crowd gatherings of millions of people every year in events\nranging from pilgrimages to protests, concerts to marathons, and festivals to\nfunerals; visual crowd analysis is emerging as a new frontier in computer\nvision. In particular, counting in highly dense crowds is a challenging problem\nwith far-reaching applicability in crowd safety and management, as well as\ngauging political significance of protests and demonstrations. In this paper,\nwe propose a novel approach that simultaneously solves the problems of\ncounting, density map estimation and localization of people in a given dense\ncrowd image. Our formulation is based on an important observation that the\nthree problems are inherently related to each other making the loss function\nfor optimizing a deep CNN decomposable. Since localization requires\nhigh-quality images and annotations, we introduce UCF-QNRF dataset that\novercomes the shortcomings of previous datasets, and contains 1.25 million\nhumans manually marked with dot annotations. Finally, we present evaluation\nmeasures and comparison with recent deep CNN networks, including those\ndeveloped specifically for crowd counting. Our approach significantly\noutperforms state-of-the-art on the new dataset, which is the most challenging\ndataset with the largest number of crowd annotations in the most diverse set of\nscenes.","url_abs":"http://arxiv.org/abs/1808.01050v1","url_pdf":"http://arxiv.org/pdf/1808.01050v1.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":[],"tasks":[{"task_slug":"crowd-counting","task_name":"Crowd Counting"},{"task_slug":"management","task_name":"Management"},{"task_slug":"visual-crowd-analysis","task_name":"Visual Crowd Analysis"}],"methods":[],"datasets_introduced":[{"slug":"ucf-qnrf","name":"UCF-QNRF","full_name":""}],"methods_introduced":[],"results":[{"leaderboard":"/sota/crowd-counting-on-ucf-qnrf","task":"Crowd Counting","dataset":"UCF-QNRF","model":"Idrees et al.","rank_in_archive_order":16,"of":23,"metrics":{"MAE":"132"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1808.01050","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}