{"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/dist-gan-an-improved-gan-using-distance","title":"Dist-GAN: An Improved GAN using Distance Constraints","arxiv_id":"1803.08887","date":"2018-03-23","proceeding":"ECCV 2018 9","authors":["Ngoc-Trung Tran","Tuan-Anh Bui","Ngai-Man Cheung"],"abstract":"We introduce effective training algorithms for Generative Adversarial\nNetworks (GAN) to alleviate mode collapse and gradient vanishing. In our\nsystem, we constrain the generator by an Autoencoder (AE). We propose a\nformulation to consider the reconstructed samples from AE as \"real\" samples for\nthe discriminator. This couples the convergence of the AE with that of the\ndiscriminator, effectively slowing down the convergence of discriminator and\nreducing gradient vanishing. Importantly, we propose two novel distance\nconstraints to improve the generator. First, we propose a latent-data distance\nconstraint to enforce compatibility between the latent sample distances and the\ncorresponding data sample distances. We use this constraint to explicitly\nprevent the generator from mode collapse. Second, we propose a\ndiscriminator-score distance constraint to align the distribution of the\ngenerated samples with that of the real samples through the discriminator\nscore. We use this constraint to guide the generator to synthesize samples that\nresemble the real ones. Our proposed GAN using these distance constraints,\nnamely Dist-GAN, can achieve better results than state-of-the-art methods\nacross benchmark datasets: synthetic, MNIST, MNIST-1K, CelebA, CIFAR-10 and\nSTL-10 datasets. Our code is published here (https://github.com/tntrung/gan)\nfor research.","url_abs":"http://arxiv.org/abs/1803.08887v3","url_pdf":"http://arxiv.org/pdf/1803.08887v3.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":"dist-gan-an-improved-gan-using-distance","repo_url":"https://github.com/tntrung/gan","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"GPL-3.0"}}],"tasks":[{"task_slug":"image-generation","task_name":"Image Generation"}],"methods":[{"method_slug":"ae","method_name":"AE"},{"method_slug":"convolution","method_name":"Convolution"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/image-generation-on-cifar-10","task":"Image Generation","dataset":"CIFAR-10","model":"Dist-GAN","rank_in_archive_order":50,"of":78,"metrics":{"FID":"17.61"},"uses_additional_data":false},{"leaderboard":"/sota/image-generation-on-stl-10","task":"Image Generation","dataset":"STL-10","model":"Dist-GAN","rank_in_archive_order":24,"of":31,"metrics":{"FID":"36.19"},"uses_additional_data":false}],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1803.08887","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}