{"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/unsupervised-total-variation-loss-for-semi","title":"Unsupervised Total Variation Loss for Semi-supervised Deep Learning of Semantic Segmentation","arxiv_id":"1605.01368","date":"2016-05-04","proceeding":null,"authors":["Mehran Javanmardi","Mehdi Sajjadi","Ting Liu","Tolga Tasdizen"],"abstract":"We introduce a novel unsupervised loss function for learning semantic\nsegmentation with deep convolutional neural nets (ConvNet) when densely labeled\ntraining images are not available. More specifically, the proposed loss\nfunction penalizes the L1-norm of the gradient of the label probability vector\nimage , i.e. total variation, produced by the ConvNet. This can be seen as a\nregularization term that promotes piecewise smoothness of the label probability\nvector image produced by the ConvNet during learning. The unsupervised loss\nfunction is combined with a supervised loss in a semi-supervised setting to\nlearn ConvNets that can achieve high semantic segmentation accuracy even when\nonly a tiny percentage of the pixels in the training images are labeled. We\ndemonstrate significant improvements over the purely supervised setting in the\nWeizmann horse, Stanford background and Sift Flow datasets. Furthermore, we\nshow that using the proposed piecewise smoothness constraint in the learning\nphase significantly outperforms post-processing results from a purely\nsupervised approach with Markov Random Fields (MRF). Finally, we note that the\nframework we introduce is general and can be used to learn to label other types\nof structures such as curvilinear structures by modifying the unsupervised loss\nfunction accordingly.","url_abs":"http://arxiv.org/abs/1605.01368v3","url_pdf":"http://arxiv.org/pdf/1605.01368v3.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":"unsupervised-total-variation-loss-for-semi","repo_url":"https://github.com/HiLab-git/WSL4MIS","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"unsupervised-total-variation-loss-for-semi","repo_url":"https://github.com/luoxd1996/wsl4mis","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}