{"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/learning-attraction-field-representation-for","title":"Learning Attraction Field Representation for Robust Line Segment Detection","arxiv_id":"1812.02122","date":"2018-12-05","proceeding":"CVPR 2019 6","authors":["Nan Xue","Song Bai","Fu-Dong Wang","Gui-Song Xia","Tianfu Wu","Liangpei Zhang"],"abstract":"This paper presents a region-partition based attraction field dual\nrepresentation for line segment maps, and thus poses the problem of line\nsegment detection (LSD) as the region coloring problem. The latter is then\naddressed by learning deep convolutional neural networks (ConvNets) for\naccuracy, robustness and efficiency. For a 2D line segment map, our dual\nrepresentation consists of three components: (i) A region-partition map in\nwhich every pixel is assigned to one and only one line segment; (ii) An\nattraction field map in which every pixel in a partition region is encoded by\nits 2D projection vector w.r.t. the associated line segment; and (iii) A\nsqueeze module which squashes the attraction field to a line segment map that\nalmost perfectly recovers the input one. By leveraging the duality, we learn\nConvNets to compute the attraction field maps for raw in-put images, followed\nby the squeeze module for LSD, in an end-to-end manner. Our method rigorously\naddresses several challenges in LSD such as local ambiguity and class\nimbalance. Our method also harnesses the best practices developed in ConvNets\nbased semantic segmentation methods such as the encoder-decoder architecture\nand the a-trous convolution. In experiments, our method is tested on the\nWireFrame dataset and the YorkUrban dataset with state-of-the-art performance\nobtained. Especially, we advance the performance by 4.5 percents on the\nWireFrame dataset. Our method is also fast with 6.6~10.4 FPS, outperforming\nmost of existing line segment detectors.","url_abs":"http://arxiv.org/abs/1812.02122v2","url_pdf":"http://arxiv.org/pdf/1812.02122v2.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":"learning-attraction-field-representation-for","repo_url":"https://github.com/cherubicXN/afm_cvpr2019","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"line-segment-detection","task_name":"Line Segment Detection"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1812.02122","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}