{"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/models-matter-so-does-training-an-empirical","title":"Models Matter, So Does Training: An Empirical Study of CNNs for Optical Flow Estimation","arxiv_id":"1809.05571","date":"2018-09-14","proceeding":null,"authors":["Deqing Sun","Xiaodong Yang","Ming-Yu Liu","Jan Kautz"],"abstract":"We investigate two crucial and closely related aspects of CNNs for optical\nflow estimation: models and training. First, we design a compact but effective\nCNN model, called PWC-Net, according to simple and well-established principles:\npyramidal processing, warping, and cost volume processing. PWC-Net is 17 times\nsmaller in size, 2 times faster in inference, and 11\\% more accurate on Sintel\nfinal than the recent FlowNet2 model. It is the winning entry in the optical\nflow competition of the robust vision challenge. Next, we experimentally\nanalyze the sources of our performance gains. In particular, we use the same\ntraining procedure of PWC-Net to retrain FlowNetC, a sub-network of FlowNet2.\nThe retrained FlowNetC is 56\\% more accurate on Sintel final than the\npreviously trained one and even 5\\% more accurate than the FlowNet2 model. We\nfurther improve the training procedure and increase the accuracy of PWC-Net on\nSintel by 10\\% and on KITTI 2012 and 2015 by 20\\%. Our newly trained model\nparameters and training protocols will be available on\nhttps://github.com/NVlabs/PWC-Net","url_abs":"http://arxiv.org/abs/1809.05571v1","url_pdf":"http://arxiv.org/pdf/1809.05571v1.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":"models-matter-so-does-training-an-empirical","repo_url":"https://github.com/NVlabs/PWC-Net","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"NOASSERTION"}},{"paper_slug":"models-matter-so-does-training-an-empirical","repo_url":"https://github.com/fpsandnoob/pwc_net","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"mindspore","reach":{"status":"ok"}}],"tasks":[{"task_slug":"optical-flow-estimation","task_name":"Optical Flow Estimation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/optical-flow-estimation-on-kitti-2012","task":"Optical Flow Estimation","dataset":"KITTI 2012","model":"PWC-Net + ft - axXiv","rank_in_archive_order":8,"of":12,"metrics":{"Average End-Point Error":"1.5"},"uses_additional_data":false}],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1809.05571","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}