{"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/global-versus-localized-generative","title":"Global versus Localized Generative Adversarial Nets","arxiv_id":"1711.06020","date":"2017-11-16","proceeding":"CVPR 2018 6","authors":["Guo-Jun Qi","Liheng Zhang","Hao Hu","Marzieh Edraki","Jingdong Wang","Xian-Sheng Hua"],"abstract":"In this paper, we present a novel localized Generative Adversarial Net (GAN)\nto learn on the manifold of real data. Compared with the classic GAN that {\\em\nglobally} parameterizes a manifold, the Localized GAN (LGAN) uses local\ncoordinate charts to parameterize distinct local geometry of how data points\ncan transform at different locations on the manifold. Specifically, around each\npoint there exists a {\\em local} generator that can produce data following\ndiverse patterns of transformations on the manifold. The locality nature of\nLGAN enables local generators to adapt to and directly access the local\ngeometry without need to invert the generator in a global GAN. Furthermore, it\ncan prevent the manifold from being locally collapsed to a dimensionally\ndeficient tangent subspace by imposing an orthonormality prior between\ntangents. This provides a geometric approach to alleviating mode collapse at\nleast locally on the manifold by imposing independence between data\ntransformations in different tangent directions. We will also demonstrate the\nLGAN can be applied to train a robust classifier that prefers locally\nconsistent classification decisions on the manifold, and the resultant\nregularizer is closely related with the Laplace-Beltrami operator. Our\nexperiments show that the proposed LGANs can not only produce diverse image\ntransformations, but also deliver superior classification performances.","url_abs":"http://arxiv.org/abs/1711.06020v2","url_pdf":"http://arxiv.org/pdf/1711.06020v2.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":"global-versus-localized-generative","repo_url":"https://github.com/TwistedW/LGAN_pytorch","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"global-versus-localized-generative","repo_url":"https://github.com/z331565360/Localized-GAN","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"classification","task_name":"General Classification"}],"methods":[{"method_slug":"convolution","method_name":"Convolution"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}