{"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/normface-l2-hypersphere-embedding-for-face","title":"NormFace: L2 Hypersphere Embedding for Face Verification","arxiv_id":"1704.06369","date":"2017-04-21","proceeding":null,"authors":["Feng Wang","Xiang Xiang","Jian Cheng","Alan L. Yuille"],"abstract":"Thanks to the recent developments of Convolutional Neural Networks, the\nperformance of face verification methods has increased rapidly. In a typical\nface verification method, feature normalization is a critical step for boosting\nperformance. This motivates us to introduce and study the effect of\nnormalization during training. But we find this is non-trivial, despite\nnormalization being differentiable. We identify and study four issues related\nto normalization through mathematical analysis, which yields understanding and\nhelps with parameter settings. Based on this analysis we propose two strategies\nfor training using normalized features. The first is a modification of softmax\nloss, which optimizes cosine similarity instead of inner-product. The second is\na reformulation of metric learning by introducing an agent vector for each\nclass. We show that both strategies, and small variants, consistently improve\nperformance by between 0.2% to 0.4% on the LFW dataset based on two models.\nThis is significant because the performance of the two models on LFW dataset is\nclose to saturation at over 98%. Codes and models are released on\nhttps://github.com/happynear/NormFace","url_abs":"http://arxiv.org/abs/1704.06369v4","url_pdf":"http://arxiv.org/pdf/1704.06369v4.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":"normface-l2-hypersphere-embedding-for-face","repo_url":"https://github.com/happynear/NormFace","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":{"status":"unanswered"}},{"paper_slug":"normface-l2-hypersphere-embedding-for-face","repo_url":"https://github.com/anax32/dimensionality-reduction","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}},{"paper_slug":"normface-l2-hypersphere-embedding-for-face","repo_url":"https://github.com/doanmanhduy0210/ResearchPaperfacerecognitions","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"face-verification","task_name":"Face Verification"},{"task_slug":"metric-learning","task_name":"Metric Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1704.06369","atlas_url":"https://app.syntology.ai/?focus=1704.06369","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}