{"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/structured-generative-adversarial-networks","title":"Structured Generative Adversarial Networks","arxiv_id":"1711.00889","date":"2017-11-02","proceeding":"NeurIPS 2017 12","authors":["Zhijie Deng","Hao Zhang","Xiaodan Liang","Luona Yang","Shizhen Xu","Jun Zhu","Eric P. Xing"],"abstract":"We study the problem of conditional generative modeling based on designated\nsemantics or structures. Existing models that build conditional generators\neither require massive labeled instances as supervision or are unable to\naccurately control the semantics of generated samples. We propose structured\ngenerative adversarial networks (SGANs) for semi-supervised conditional\ngenerative modeling. SGAN assumes the data x is generated conditioned on two\nindependent latent variables: y that encodes the designated semantics, and z\nthat contains other factors of variation. To ensure disentangled semantics in y\nand z, SGAN builds two collaborative games in the hidden space to minimize the\nreconstruction error of y and z, respectively. Training SGAN also involves\nsolving two adversarial games that have their equilibrium concentrating at the\ntrue joint data distributions p(x, z) and p(x, y), avoiding distributing the\nprobability mass diffusely over data space that MLE-based methods may suffer.\nWe assess SGAN by evaluating its trained networks, and its performance on\ndownstream tasks. We show that SGAN delivers a highly controllable generator,\nand disentangled representations; it also establishes start-of-the-art results\nacross multiple datasets when applied for semi-supervised image classification\n(1.27%, 5.73%, 17.26% error rates on MNIST, SVHN and CIFAR-10 using 50, 1000\nand 4000 labels, respectively). Benefiting from the separate modeling of y and\nz, SGAN can generate images with high visual quality and strictly following the\ndesignated semantic, and can be extended to a wide spectrum of applications,\nsuch as style transfer.","url_abs":"http://arxiv.org/abs/1711.00889v1","url_pdf":"http://arxiv.org/pdf/1711.00889v1.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":"structured-generative-adversarial-networks","repo_url":"https://github.com/thudzj/StructuredGAN","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"semi-supervised-image-classification","task_name":"Semi-Supervised Image Classification"},{"task_slug":"style-transfer","task_name":"Style Transfer"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1711.00889","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}