{"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/learning-on-the-edge-explicit-boundary","title":"Learning on the Edge: Explicit Boundary Handling in CNNs","arxiv_id":"1805.03106","date":"2018-05-08","proceeding":null,"authors":["Carlo Innamorati","Tobias Ritschel","Tim Weyrich","Niloy J. Mitra"],"abstract":"Convolutional neural networks (CNNs) handle the case where filters extend\nbeyond the image boundary using several heuristics, such as zero, repeat or\nmean padding. These schemes are applied in an ad-hoc fashion and, being weakly\nrelated to the image content and oblivious of the target task, result in low\noutput quality at the boundary. In this paper, we propose a simple and\neffective improvement that learns the boundary handling itself. At\ntraining-time, the network is provided with a separate set of explicit boundary\nfilters. At testing-time, we use these filters which have learned to\nextrapolate features at the boundary in an optimal way for the specific task.\nOur extensive evaluation, over a wide range of architectural changes\n(variations of layers, feature channels, or both), shows how the explicit\nfilters result in improved boundary handling. Consequently, we demonstrate an\nimprovement of 5% to 20% across the board of typical CNN applications\n(colorization, de-Bayering, optical flow, and disparity estimation).","url_abs":"http://arxiv.org/abs/1805.03106v1","url_pdf":"http://arxiv.org/pdf/1805.03106v1.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":"learning-on-the-edge-explicit-boundary","repo_url":"https://github.com/stfc-sciml/differentialconv2d","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"colorization","task_name":"Colorization"},{"task_slug":"disparity-estimation","task_name":"Disparity Estimation"},{"task_slug":"optical-flow-estimation","task_name":"Optical Flow Estimation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}