{"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/from-source-to-target-and-back-symmetric-bi","title":"From source to target and back: symmetric bi-directional adaptive GAN","arxiv_id":"1705.08824","date":"2017-05-24","proceeding":"CVPR 2018 6","authors":["Paolo Russo","Fabio Maria Carlucci","Tatiana Tommasi","Barbara Caputo"],"abstract":"The effectiveness of generative adversarial approaches in producing images\naccording to a specific style or visual domain has recently opened new\ndirections to solve the unsupervised domain adaptation problem. It has been\nshown that source labeled images can be modified to mimic target samples making\nit possible to train directly a classifier in the target domain, despite the\noriginal lack of annotated data. Inverse mappings from the target to the source\ndomain have also been evaluated but only passing through adapted feature\nspaces, thus without new image generation. In this paper we propose to better\nexploit the potential of generative adversarial networks for adaptation by\nintroducing a novel symmetric mapping among domains. We jointly optimize\nbi-directional image transformations combining them with target self-labeling.\nMoreover we define a new class consistency loss that aligns the generators in\nthe two directions imposing to conserve the class identity of an image passing\nthrough both domain mappings. A detailed qualitative and quantitative analysis\nof the reconstructed images confirm the power of our approach. By integrating\nthe two domain specific classifiers obtained with our bi-directional network we\nexceed previous state-of-the-art unsupervised adaptation results on four\ndifferent benchmark datasets.","url_abs":"http://arxiv.org/abs/1705.08824v2","url_pdf":"http://arxiv.org/pdf/1705.08824v2.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":[],"tasks":[{"task_slug":"domain-adaptation","task_name":"Domain Adaptation"},{"task_slug":"image-generation","task_name":"Image Generation"},{"task_slug":"unsupervised-domain-adaptation","task_name":"Unsupervised Domain Adaptation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/domain-adaptation-on-svhn-to-mnist","task":"Domain Adaptation","dataset":"SVHN-to-MNIST","model":"SBADA","rank_in_archive_order":14,"of":14,"metrics":{"Accuracy":"76.1"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1705.08824","atlas_url":"https://app.syntology.ai/?focus=1705.08824","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}