{"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/what-makes-good-synthetic-training-data-for","title":"What Makes Good Synthetic Training Data for Learning Disparity and Optical Flow Estimation?","arxiv_id":"1801.06397","date":"2018-01-19","proceeding":null,"authors":["Nikolaus Mayer","Eddy Ilg","Philipp Fischer","Caner Hazirbas","Daniel Cremers","Alexey Dosovitskiy","Thomas Brox"],"abstract":"The finding that very large networks can be trained efficiently and reliably\nhas led to a paradigm shift in computer vision from engineered solutions to\nlearning formulations. As a result, the research challenge shifts from devising\nalgorithms to creating suitable and abundant training data for supervised\nlearning. How to efficiently create such training data? The dominant data\nacquisition method in visual recognition is based on web data and manual\nannotation. Yet, for many computer vision problems, such as stereo or optical\nflow estimation, this approach is not feasible because humans cannot manually\nenter a pixel-accurate flow field. In this paper, we promote the use of\nsynthetically generated data for the purpose of training deep networks on such\ntasks.We suggest multiple ways to generate such data and evaluate the influence\nof dataset properties on the performance and generalization properties of the\nresulting networks. We also demonstrate the benefit of learning schedules that\nuse different types of data at selected stages of the training process.","url_abs":"http://arxiv.org/abs/1801.06397v3","url_pdf":"http://arxiv.org/pdf/1801.06397v3.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":"what-makes-good-synthetic-training-data-for","repo_url":"https://github.com/lmb-freiburg/optical-flow-2d-data-generation","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"optical-flow-estimation","task_name":"Optical Flow Estimation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1801.06397","atlas_url":"https://app.syntology.ai/?focus=1801.06397","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}