{"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/floors-are-flat-leveraging-semantics-for-real","title":"Floors are Flat: Leveraging Semantics for Real-Time Surface Normal Prediction","arxiv_id":"1906.06792","date":"2019-06-16","proceeding":null,"authors":["Steven Hickson","Karthik Raveendran","Alireza Fathi","Kevin Murphy","Irfan Essa"],"abstract":"We propose 4 insights that help to significantly improve the performance of deep learning models that predict surface normals and semantic labels from a single RGB image. These insights are: (1) denoise the \"ground truth\" surface normals in the training set to ensure consistency with the semantic labels; (2) concurrently train on a mix of real and synthetic data, instead of pretraining on synthetic and finetuning on real; (3) jointly predict normals and semantics using a shared model, but only backpropagate errors on pixels that have valid training labels; (4) slim down the model and use grayscale instead of color inputs. Despite the simplicity of these steps, we demonstrate consistently improved results on several datasets, using a model that runs at 12 fps on a standard mobile phone.","url_abs":"https://arxiv.org/abs/1906.06792v1","url_pdf":"https://arxiv.org/pdf/1906.06792v1.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":"floors-are-flat-leveraging-semantics-for-real","repo_url":"https://github.com/StevenHickson/CreateNormals","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null}],"tasks":[{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"},{"task_slug":"surface-normals-estimation","task_name":"Surface Normals Estimation"},{"task_slug":null,"task_name":"valid"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/semantic-segmentation-on-scannetv2","task":"Semantic Segmentation","dataset":"ScanNetV2","model":"Floors are Flat","rank_in_archive_order":12,"of":12,"metrics":{"Pixel Accuracy":"65.6"},"uses_additional_data":false},{"leaderboard":"/sota/surface-normals-estimation-on-nyu-depth-v2-1","task":"Surface Normals Estimation","dataset":"NYU Depth v2","model":"Floors are Flat","rank_in_archive_order":6,"of":6,"metrics":{"% < 11.25":"59.5","% < 22.5":"72.2","% < 30":"77.3","Mean Angle Error":"19.7","RMSE":"19.3"},"uses_additional_data":false},{"leaderboard":"/sota/surface-normals-estimation-on-scannetv2","task":"Surface Normals Estimation","dataset":"ScanNetV2","model":"Floors are Flat","rank_in_archive_order":3,"of":3,"metrics":{"% < 11.25":"50.9","% < 22.5":"65.2","% < 30":"70","Mean Angle Error":"28"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1906.06792","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}