{"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/beyond-counting-comparisons-of-density-maps","title":"Beyond Counting: Comparisons of Density Maps for Crowd Analysis Tasks - Counting, Detection, and Tracking","arxiv_id":"1705.10118","date":"2017-05-29","proceeding":null,"authors":["Di Kang","Zheng Ma","Antoni B. Chan"],"abstract":"For crowded scenes, the accuracy of object-based computer vision methods\ndeclines when the images are low-resolution and objects have severe occlusions.\nTaking counting methods for example, almost all the recent state-of-the-art\ncounting methods bypass explicit detection and adopt regression-based methods\nto directly count the objects of interest. Among regression-based methods,\ndensity map estimation, where the number of objects inside a subregion is the\nintegral of the density map over that subregion, is especially promising\nbecause it preserves spatial information, which makes it useful for both\ncounting and localization (detection and tracking). With the power of deep\nconvolutional neural networks (CNNs) the counting performance has improved\nsteadily. The goal of this paper is to evaluate density maps generated by\ndensity estimation methods on a variety of crowd analysis tasks, including\ncounting, detection, and tracking. Most existing CNN methods produce density\nmaps with resolution that is smaller than the original images, due to the\ndownsample strides in the convolution/pooling operations. To produce an\noriginal-resolution density map, we also evaluate a classical CNN that uses a\nsliding window regressor to predict the density for every pixel in the image.\nWe also consider a fully convolutional (FCNN) adaptation, with skip connections\nfrom lower convolutional layers to compensate for loss in spatial information\nduring upsampling. In our experiments, we found that the lower-resolution\ndensity maps sometimes have better counting performance. In contrast, the\noriginal-resolution density maps improved localization tasks, such as detection\nand tracking, compared to bilinear upsampling the lower-resolution density\nmaps. Finally, we also propose several metrics for measuring the quality of a\ndensity map, and relate them to experiment results on counting and\nlocalization.","url_abs":"http://arxiv.org/abs/1705.10118v2","url_pdf":"http://arxiv.org/pdf/1705.10118v2.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":"beyond-counting-comparisons-of-density-maps","repo_url":"https://github.com/krutikabapat/Crowd_Counting","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}}],"tasks":[{"task_slug":"density-estimation","task_name":"Density Estimation"},{"task_slug":"regression-1","task_name":"regression"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1705.10118","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}