{"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/decoupled-learning-for-conditional","title":"Decoupled Learning for Conditional Adversarial Networks","arxiv_id":"1801.06790","date":"2018-01-21","proceeding":null,"authors":["Zhifei Zhang","Yang song","Hairong Qi"],"abstract":"Incorporating encoding-decoding nets with adversarial nets has been widely\nadopted in image generation tasks. We observe that the state-of-the-art\nachievements were obtained by carefully balancing the reconstruction loss and\nadversarial loss, and such balance shifts with different network structures,\ndatasets, and training strategies. Empirical studies have demonstrated that an\ninappropriate weight between the two losses may cause instability, and it is\ntricky to search for the optimal setting, especially when lacking prior\nknowledge on the data and network.\n  This paper gives the first attempt to relax the need of manual balancing by\nproposing the concept of \\textit{decoupled learning}, where a novel network\nstructure is designed that explicitly disentangles the backpropagation paths of\nthe two losses.\n  Experimental results demonstrate the effectiveness, robustness, and\ngenerality of the proposed method. The other contribution of the paper is the\ndesign of a new evaluation metric to measure the image quality of generative\nmodels. We propose the so-called \\textit{normalized relative discriminative\nscore} (NRDS), which introduces the idea of relative comparison, rather than\nproviding absolute estimates like existing metrics.","url_abs":"http://arxiv.org/abs/1801.06790v1","url_pdf":"http://arxiv.org/pdf/1801.06790v1.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":"decoupled-learning-for-conditional","repo_url":"https://github.com/ZZUTK/Decoupled-Learning-Conditional-GAN","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"image-generation","task_name":"Image Generation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1801.06790","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}