{"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/genegan-learning-object-transfiguration-and","title":"GeneGAN: Learning Object Transfiguration and Attribute Subspace from Unpaired Data","arxiv_id":"1705.04932","date":"2017-05-14","proceeding":null,"authors":["Shuchang Zhou","Taihong Xiao","Yi Yang","Dieqiao Feng","Qinyao He","Weiran He"],"abstract":"Object Transfiguration replaces an object in an image with another object\nfrom a second image. For example it can perform tasks like \"putting exactly\nthose eyeglasses from image A on the nose of the person in image B\". Usage of\nexemplar images allows more precise specification of desired modifications and\nimproves the diversity of conditional image generation. However, previous\nmethods that rely on feature space operations, require paired data and/or\nappearance models for training or disentangling objects from background. In\nthis work, we propose a model that can learn object transfiguration from two\nunpaired sets of images: one set containing images that \"have\" that kind of\nobject, and the other set being the opposite, with the mild constraint that the\nobjects be located approximately at the same place. For example, the training\ndata can be one set of reference face images that have eyeglasses, and another\nset of images that have not, both of which spatially aligned by face landmarks.\nDespite the weak 0/1 labels, our model can learn an \"eyeglasses\" subspace that\ncontain multiple representatives of different types of glasses. Consequently,\nwe can perform fine-grained control of generated images, like swapping the\nglasses in two images by swapping the projected components in the \"eyeglasses\"\nsubspace, to create novel images of people wearing eyeglasses.\n  Overall, our deterministic generative model learns disentangled attribute\nsubspaces from weakly labeled data by adversarial training. Experiments on\nCelebA and Multi-PIE datasets validate the effectiveness of the proposed model\non real world data, in generating images with specified eyeglasses, smiling,\nhair styles, and lighting conditions etc. The code is available online.","url_abs":"http://arxiv.org/abs/1705.04932v1","url_pdf":"http://arxiv.org/pdf/1705.04932v1.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":"genegan-learning-object-transfiguration-and","repo_url":"https://github.com/megvii-research/genegan","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"genegan-learning-object-transfiguration-and","repo_url":"https://github.com/Prinsphield/GeneGAN","is_official":0,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"attribute","task_name":"Attribute"},{"task_slug":"conditional-image-generation","task_name":"Conditional Image Generation"},{"task_slug":"image-generation","task_name":"Image Generation"},{"task_slug":"object","task_name":"Object"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1705.04932","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}