{"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/l2-constrained-softmax-loss-for","title":"L2-constrained Softmax Loss for Discriminative Face Verification","arxiv_id":"1703.09507","date":"2017-03-28","proceeding":null,"authors":["Rajeev Ranjan","Carlos D. Castillo","Rama Chellappa"],"abstract":"In recent years, the performance of face verification systems has\nsignificantly improved using deep convolutional neural networks (DCNNs). A\ntypical pipeline for face verification includes training a deep network for\nsubject classification with softmax loss, using the penultimate layer output as\nthe feature descriptor, and generating a cosine similarity score given a pair\nof face images. The softmax loss function does not optimize the features to\nhave higher similarity score for positive pairs and lower similarity score for\nnegative pairs, which leads to a performance gap. In this paper, we add an\nL2-constraint to the feature descriptors which restricts them to lie on a\nhypersphere of a fixed radius. This module can be easily implemented using\nexisting deep learning frameworks. We show that integrating this simple step in\nthe training pipeline significantly boosts the performance of face\nverification. Specifically, we achieve state-of-the-art results on the\nchallenging IJB-A dataset, achieving True Accept Rate of 0.909 at False Accept\nRate 0.0001 on the face verification protocol. Additionally, we achieve\nstate-of-the-art performance on LFW dataset with an accuracy of 99.78%, and\ncompeting performance on YTF dataset with accuracy of 96.08%.","url_abs":"http://arxiv.org/abs/1703.09507v3","url_pdf":"http://arxiv.org/pdf/1703.09507v3.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":"l2-constrained-softmax-loss-for","repo_url":"https://github.com/samisoto/keras_cosine_based_loss","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"CC0-1.0"}}],"tasks":[{"task_slug":"face-verification","task_name":"Face Verification"}],"methods":[{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/face-verification-on-ijb-a","task":"Face Verification","dataset":"IJB-A","model":"L2-constrained softmax loss","rank_in_archive_order":4,"of":17,"metrics":{"TAR @ FAR=0.01":"97%"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1703.09507","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}