{"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/sparsity-invariant-cnns","title":"Sparsity Invariant CNNs","arxiv_id":"1708.06500","date":"2017-08-22","proceeding":null,"authors":["Jonas Uhrig","Nick Schneider","Lukas Schneider","Uwe Franke","Thomas Brox","Andreas Geiger"],"abstract":"In this paper, we consider convolutional neural networks operating on sparse\ninputs with an application to depth upsampling from sparse laser scan data.\nFirst, we show that traditional convolutional networks perform poorly when\napplied to sparse data even when the location of missing data is provided to\nthe network. To overcome this problem, we propose a simple yet effective sparse\nconvolution layer which explicitly considers the location of missing data\nduring the convolution operation. We demonstrate the benefits of the proposed\nnetwork architecture in synthetic and real experiments with respect to various\nbaseline approaches. Compared to dense baselines, the proposed sparse\nconvolution network generalizes well to novel datasets and is invariant to the\nlevel of sparsity in the data. For our evaluation, we derive a novel dataset\nfrom the KITTI benchmark, comprising 93k depth annotated RGB images. Our\ndataset allows for training and evaluating depth upsampling and depth\nprediction techniques in challenging real-world settings and will be made\navailable upon publication.","url_abs":"http://arxiv.org/abs/1708.06500v2","url_pdf":"http://arxiv.org/pdf/1708.06500v2.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":"sparsity-invariant-cnns","repo_url":"https://github.com/PeterTor/sparse_convolution","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"depth-completion","task_name":"Depth Completion"},{"task_slug":"depth-estimation","task_name":"Depth Estimation"},{"task_slug":"depth-prediction","task_name":"Depth Prediction"}],"methods":[{"method_slug":"convolution","method_name":"Convolution"}],"datasets_introduced":[{"slug":"kitti-depth","name":"KITTI-Depth","full_name":null}],"methods_introduced":[],"results":[{"leaderboard":"/sota/depth-completion-on-kitti-depth-completion","task":"Depth Completion","dataset":"KITTI Depth Completion","model":"SparseConvs","rank_in_archive_order":16,"of":16,"metrics":{"MAE":"481","RMSE":"1601","Runtime [ms]":"10"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1708.06500","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}