{"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/stacked-generative-adversarial-networks","title":"Stacked Generative Adversarial Networks","arxiv_id":"1612.04357","date":"2016-12-13","proceeding":"CVPR 2017 7","authors":["Xun Huang","Yixuan Li","Omid Poursaeed","John Hopcroft","Serge Belongie"],"abstract":"In this paper, we propose a novel generative model named Stacked Generative\nAdversarial Networks (SGAN), which is trained to invert the hierarchical\nrepresentations of a bottom-up discriminative network. Our model consists of a\ntop-down stack of GANs, each learned to generate lower-level representations\nconditioned on higher-level representations. A representation discriminator is\nintroduced at each feature hierarchy to encourage the representation manifold\nof the generator to align with that of the bottom-up discriminative network,\nleveraging the powerful discriminative representations to guide the generative\nmodel. In addition, we introduce a conditional loss that encourages the use of\nconditional information from the layer above, and a novel entropy loss that\nmaximizes a variational lower bound on the conditional entropy of generator\noutputs. We first train each stack independently, and then train the whole\nmodel end-to-end. Unlike the original GAN that uses a single noise vector to\nrepresent all the variations, our SGAN decomposes variations into multiple\nlevels and gradually resolves uncertainties in the top-down generative process.\nBased on visual inspection, Inception scores and visual Turing test, we\ndemonstrate that SGAN is able to generate images of much higher quality than\nGANs without stacking.","url_abs":"http://arxiv.org/abs/1612.04357v4","url_pdf":"http://arxiv.org/pdf/1612.04357v4.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":"stacked-generative-adversarial-networks","repo_url":"https://github.com/xunhuang1995/SGAN","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}},{"paper_slug":"stacked-generative-adversarial-networks","repo_url":"https://github.com/jinsel/Text-to-Image-Generation","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"conditional-image-generation","task_name":"Conditional Image Generation"}],"methods":[{"method_slug":"convolution","method_name":"Convolution"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/conditional-image-generation-on-cifar-10","task":"Conditional Image Generation","dataset":"CIFAR-10","model":"SGAN","rank_in_archive_order":19,"of":25,"metrics":{"Inception score":"8.59"},"uses_additional_data":false}],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1612.04357","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}