{"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/recurrent-scale-approximation-for-object","title":"Recurrent Scale Approximation for Object Detection in CNN","arxiv_id":"1707.09531","date":"2017-07-29","proceeding":"ICCV 2017 10","authors":["Yu Liu","Hongyang Li","Junjie Yan","Fangyin Wei","Xiaogang Wang","Xiaoou Tang"],"abstract":"Since convolutional neural network (CNN) lacks an inherent mechanism to\nhandle large scale variations, we always need to compute feature maps multiple\ntimes for multi-scale object detection, which has the bottleneck of\ncomputational cost in practice. To address this, we devise a recurrent scale\napproximation (RSA) to compute feature map once only, and only through this map\ncan we approximate the rest maps on other levels. At the core of RSA is the\nrecursive rolling out mechanism: given an initial map at a particular scale, it\ngenerates the prediction at a smaller scale that is half the size of input. To\nfurther increase efficiency and accuracy, we (a): design a scale-forecast\nnetwork to globally predict potential scales in the image since there is no\nneed to compute maps on all levels of the pyramid. (b): propose a landmark\nretracing network (LRN) to trace back locations of the regressed landmarks and\ngenerate a confidence score for each landmark; LRN can effectively alleviate\nfalse positives caused by the accumulated error in RSA. The whole system can be\ntrained end-to-end in a unified CNN framework. Experiments demonstrate that our\nproposed algorithm is superior against state-of-the-art methods on face\ndetection benchmarks and achieves comparable results for generic proposal\ngeneration. The source code of RSA is available at\ngithub.com/sciencefans/RSA-for-object-detection.","url_abs":"http://arxiv.org/abs/1707.09531v2","url_pdf":"http://arxiv.org/pdf/1707.09531v2.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":"recurrent-scale-approximation-for-object","repo_url":"https://github.com/sciencefans/RSA-for-object-detection","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":null}],"tasks":[{"task_slug":"face-detection","task_name":"Face Detection"},{"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":[{"leaderboard":"/sota/face-detection-on-annotated-faces-in-the-wild","task":"Face Detection","dataset":"Annotated Faces in the Wild","model":"LRN + RSA","rank_in_archive_order":3,"of":7,"metrics":{"AP":"0.9917"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1707.09531","atlas_url":"https://app.syntology.ai/?focus=1707.09531","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}