{"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/spatial-as-deep-spatial-cnn-for-traffic-scene","title":"Spatial As Deep: Spatial CNN for Traffic Scene Understanding","arxiv_id":"1712.06080","date":"2017-12-17","proceeding":null,"authors":["Xingang Pan","Jianping Shi","Ping Luo","Xiaogang Wang","Xiaoou Tang"],"abstract":"Convolutional neural networks (CNNs) are usually built by stacking\nconvolutional operations layer-by-layer. Although CNN has shown strong\ncapability to extract semantics from raw pixels, its capacity to capture\nspatial relationships of pixels across rows and columns of an image is not\nfully explored. These relationships are important to learn semantic objects\nwith strong shape priors but weak appearance coherences, such as traffic lanes,\nwhich are often occluded or not even painted on the road surface as shown in\nFig. 1 (a). In this paper, we propose Spatial CNN (SCNN), which generalizes\ntraditional deep layer-by-layer convolutions to slice-byslice convolutions\nwithin feature maps, thus enabling message passings between pixels across rows\nand columns in a layer. Such SCNN is particular suitable for long continuous\nshape structure or large objects, with strong spatial relationship but less\nappearance clues, such as traffic lanes, poles, and wall. We apply SCNN on a\nnewly released very challenging traffic lane detection dataset and Cityscapse\ndataset. The results show that SCNN could learn the spatial relationship for\nstructure output and significantly improves the performance. We show that SCNN\noutperforms the recurrent neural network (RNN) based ReNet and MRF+CNN (MRFNet)\nin the lane detection dataset by 8.7% and 4.6% respectively. Moreover, our SCNN\nwon the 1st place on the TuSimple Benchmark Lane Detection Challenge, with an\naccuracy of 96.53%.","url_abs":"http://arxiv.org/abs/1712.06080v1","url_pdf":"http://arxiv.org/pdf/1712.06080v1.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":"spatial-as-deep-spatial-cnn-for-traffic-scene","repo_url":"https://github.com/XingangPan/SCNN","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"torch","reach":{"status":"unanswered"}},{"paper_slug":"spatial-as-deep-spatial-cnn-for-traffic-scene","repo_url":"https://github.com/GuangyanZhang/SCNN-Deeplabv3-bisenet-icnet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"paddle","reach":{"status":"ok"}},{"paper_slug":"spatial-as-deep-spatial-cnn-for-traffic-scene","repo_url":"https://github.com/cardwing/Codes-for-Lane-Detection","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"spatial-as-deep-spatial-cnn-for-traffic-scene","repo_url":"https://github.com/harryhan618/SCNN_Pytorch","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"spatial-as-deep-spatial-cnn-for-traffic-scene","repo_url":"https://github.com/harryhan618/scnn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"torch","reach":{"status":"unanswered"}},{"paper_slug":"spatial-as-deep-spatial-cnn-for-traffic-scene","repo_url":"https://github.com/wind754203900/SCNN-TF2","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}},{"paper_slug":"spatial-as-deep-spatial-cnn-for-traffic-scene","repo_url":"https://github.com/Turoad/lanedet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":null},{"paper_slug":"spatial-as-deep-spatial-cnn-for-traffic-scene","repo_url":"https://github.com/zkyntu/UnLanedet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":null},{"paper_slug":"spatial-as-deep-spatial-cnn-for-traffic-scene","repo_url":"https://github.com/zkyseu/PPlanedet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"paddle","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"lane-detection","task_name":"Lane Detection"},{"task_slug":"scene-understanding","task_name":"Scene Understanding"}],"methods":[],"datasets_introduced":[{"slug":"culane","name":"CULane","full_name":""}],"methods_introduced":[],"results":[{"leaderboard":"/sota/lane-detection-on-culane","task":"Lane Detection","dataset":"CULane","model":"SCNN","rank_in_archive_order":58,"of":63,"metrics":{"F1 score":"71.6"},"uses_additional_data":true}],"syntology":{"syntology_url":"https://syntology.ai/paper/1712.06080","atlas_url":"https://app.syntology.ai/?focus=1712.06080","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}