{"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/detecting-overfitting-of-deep-generative","title":"Detecting Overfitting of Deep Generative Networks via Latent Recovery","arxiv_id":"1901.03396","date":"2019-01-09","proceeding":"CVPR 2019 6","authors":["Ryan Webster","Julien Rabin","Loic Simon","Frederic Jurie"],"abstract":"State of the art deep generative networks are capable of producing images\nwith such incredible realism that they can be suspected of memorizing training\nimages. It is why it is not uncommon to include visualizations of training set\nnearest neighbors, to suggest generated images are not simply memorized. We\ndemonstrate this is not sufficient and motivates the need to study\nmemorization/overfitting of deep generators with more scrutiny. This paper\naddresses this question by i) showing how simple losses are highly effective at\nreconstructing images for deep generators ii) analyzing the statistics of\nreconstruction errors when reconstructing training and validation images, which\nis the standard way to analyze overfitting in machine learning. Using this\nmethodology, this paper shows that overfitting is not detectable in the pure\nGAN models proposed in the literature, in contrast with those using hybrid\nadversarial losses, which are amongst the most widely applied generative\nmethods. The paper also shows that standard GAN evaluation metrics fail to\ncapture memorization for some deep generators. Finally, the paper also shows\nhow off-the-shelf GAN generators can be successfully applied to face inpainting\nand face super-resolution using the proposed reconstruction method, without\nhybrid adversarial losses.","url_abs":"http://arxiv.org/abs/1901.03396v1","url_pdf":"http://arxiv.org/pdf/1901.03396v1.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":"detecting-overfitting-of-deep-generative","repo_url":"https://github.com/ryanwebster90/gen-overfitting-latent-recovery","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"facial-inpainting","task_name":"Facial Inpainting"},{"task_slug":"memorization","task_name":"Memorization"},{"task_slug":"super-resolution","task_name":"Super-Resolution"}],"methods":[{"method_slug":"convolution","method_name":"Convolution"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1901.03396","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}