{"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/adm-for-grid-crf-loss-in-cnn-segmentation","title":"Beyond Gradient Descent for Regularized Segmentation Losses","arxiv_id":"1809.02322","date":"2018-09-07","proceeding":"CVPR 2019 6","authors":["Dmitrii Marin","Meng Tang","Ismail Ben Ayed","Yuri Boykov"],"abstract":"The simplicity of gradient descent (GD) made it the default method for\ntraining ever-deeper and complex neural networks. Both loss functions and\narchitectures are often explicitly tuned to be amenable to this basic local\noptimization. In the context of weakly-supervised CNN segmentation, we\ndemonstrate a well-motivated loss function where an alternative optimizer (ADM)\nachieves the state-of-the-art while GD performs poorly. Interestingly, GD\nobtains its best result for a \"smoother\" tuning of the loss function. The\nresults are consistent across different network architectures. Our loss is\nmotivated by well-understood MRF/CRF regularization models in \"shallow\"\nsegmentation and their known global solvers. Our work suggests that network\ndesign/training should pay more attention to optimization methods.","url_abs":"http://arxiv.org/abs/1809.02322v2","url_pdf":"http://arxiv.org/pdf/1809.02322v2.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":"adm-for-grid-crf-loss-in-cnn-segmentation","repo_url":"https://github.com/dmitrii-marin/adm-seg","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"segmentation","task_name":"Segmentation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}