{"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/transferring-gans-generating-images-from","title":"Transferring GANs: generating images from limited data","arxiv_id":"1805.01677","date":"2018-05-04","proceeding":"ECCV 2018 9","authors":["Yaxing Wang","Chenshen Wu","Luis Herranz","Joost Van de Weijer","Abel Gonzalez-Garcia","Bogdan Raducanu"],"abstract":"Transferring the knowledge of pretrained networks to new domains by means of\nfinetuning is a widely used practice for applications based on discriminative\nmodels. To the best of our knowledge this practice has not been studied within\nthe context of generative deep networks. Therefore, we study domain adaptation\napplied to image generation with generative adversarial networks. We evaluate\nseveral aspects of domain adaptation, including the impact of target domain\nsize, the relative distance between source and target domain, and the\ninitialization of conditional GANs. Our results show that using knowledge from\npretrained networks can shorten the convergence time and can significantly\nimprove the quality of the generated images, especially when the target data is\nlimited. We show that these conclusions can also be drawn for conditional GANs\neven when the pretrained model was trained without conditioning. Our results\nalso suggest that density may be more important than diversity and a dataset\nwith one or few densely sampled classes may be a better source model than more\ndiverse datasets such as ImageNet or Places.","url_abs":"http://arxiv.org/abs/1805.01677v2","url_pdf":"http://arxiv.org/pdf/1805.01677v2.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":"transferring-gans-generating-images-from","repo_url":"https://github.com/WuChenshen/MeRGAN","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"10-shot-image-generation","task_name":"10-shot image generation"},{"task_slug":"diversity","task_name":"Diversity"},{"task_slug":"domain-adaptation","task_name":"Domain Adaptation"},{"task_slug":"image-generation","task_name":"Image Generation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/10-shot-image-generation-on-babies","task":"10-shot image generation","dataset":"Babies","model":"TGAN","rank_in_archive_order":7,"of":7,"metrics":{"FID":"101.58"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1805.01677","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}