{"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/dynamic-filter-networks","title":"Dynamic Filter Networks","arxiv_id":"1605.09673","date":"2016-05-31","proceeding":"NeurIPS 2016 12","authors":["Bert De Brabandere","Xu Jia","Tinne Tuytelaars","Luc van Gool"],"abstract":"In a traditional convolutional layer, the learned filters stay fixed after\ntraining. In contrast, we introduce a new framework, the Dynamic Filter\nNetwork, where filters are generated dynamically conditioned on an input. We\nshow that this architecture is a powerful one, with increased flexibility\nthanks to its adaptive nature, yet without an excessive increase in the number\nof model parameters. A wide variety of filtering operations can be learned this\nway, including local spatial transformations, but also others like selective\n(de)blurring or adaptive feature extraction. Moreover, multiple such layers can\nbe combined, e.g. in a recurrent architecture. We demonstrate the effectiveness\nof the dynamic filter network on the tasks of video and stereo prediction, and\nreach state-of-the-art performance on the moving MNIST dataset with a much\nsmaller model. By visualizing the learned filters, we illustrate that the\nnetwork has picked up flow information by only looking at unlabelled training\ndata. This suggests that the network can be used to pretrain networks for\nvarious supervised tasks in an unsupervised way, like optical flow and depth\nestimation.","url_abs":"http://arxiv.org/abs/1605.09673v2","url_pdf":"http://arxiv.org/pdf/1605.09673v2.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":"dynamic-filter-networks","repo_url":"https://github.com/dbbert/dfn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}}],"tasks":[{"task_slug":"depth-estimation","task_name":"Depth Estimation"},{"task_slug":"optical-flow-estimation","task_name":"Optical Flow Estimation"},{"task_slug":"video-prediction","task_name":"Video Prediction"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/video-prediction-on-kth","task":"Video Prediction","dataset":"KTH","model":"DFN","rank_in_archive_order":26,"of":31,"metrics":{"Cond":"10","PSNR":"27.26","Pred":"20","SSIM":"0.794"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1605.09673","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}