{"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/the-devil-of-face-recognition-is-in-the-noise","title":"The Devil of Face Recognition is in the Noise","arxiv_id":"1807.11649","date":"2018-07-31","proceeding":"ECCV 2018 9","authors":["Fei Wang","Liren Chen","Cheng Li","Shiyao Huang","Yanjie Chen","Chen Qian","Chen Change Loy"],"abstract":"The growing scale of face recognition datasets empowers us to train strong\nconvolutional networks for face recognition. While a variety of architectures\nand loss functions have been devised, we still have a limited understanding of\nthe source and consequence of label noise inherent in existing datasets. We\nmake the following contributions: 1) We contribute cleaned subsets of popular\nface databases, i.e., MegaFace and MS-Celeb-1M datasets, and build a new\nlarge-scale noise-controlled IMDb-Face dataset. 2) With the original datasets\nand cleaned subsets, we profile and analyze label noise properties of MegaFace\nand MS-Celeb-1M. We show that a few orders more samples are needed to achieve\nthe same accuracy yielded by a clean subset. 3) We study the association\nbetween different types of noise, i.e., label flips and outliers, with the\naccuracy of face recognition models. 4) We investigate ways to improve data\ncleanliness, including a comprehensive user study on the influence of data\nlabeling strategies to annotation accuracy. The IMDb-Face dataset has been\nreleased on https://github.com/fwang91/IMDb-Face.","url_abs":"http://arxiv.org/abs/1807.11649v1","url_pdf":"http://arxiv.org/pdf/1807.11649v1.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":"the-devil-of-face-recognition-is-in-the-noise","repo_url":"https://github.com/fwang91/IMDb-Face","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}},{"paper_slug":"the-devil-of-face-recognition-is-in-the-noise","repo_url":"https://github.com/wangx404/script_for_downloading_IMDb_face","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}}],"tasks":[{"task_slug":"face-recognition","task_name":"Face Recognition"}],"methods":[],"datasets_introduced":[{"slug":"imdb-face","name":"IMDb-Face","full_name":""}],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1807.11649","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}