{"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/a-good-practice-towards-top-performance-of","title":"A Good Practice Towards Top Performance of Face Recognition: Transferred Deep Feature Fusion","arxiv_id":"1704.00438","date":"2017-04-03","proceeding":null,"authors":["Lin Xiong","Jayashree Karlekar","Jian Zhao","Yi Cheng","Yan Xu","Jiashi Feng","Sugiri Pranata","ShengMei Shen"],"abstract":"Unconstrained face recognition performance evaluations have traditionally\nfocused on Labeled Faces in the Wild (LFW) dataset for imagery and the\nYouTubeFaces (YTF) dataset for videos in the last couple of years. Spectacular\nprogress in this field has resulted in saturation on verification and\nidentification accuracies for those benchmark datasets. In this paper, we\npropose a unified learning framework named Transferred Deep Feature Fusion\n(TDFF) targeting at the new IARPA Janus Benchmark A (IJB-A) face recognition\ndataset released by NIST face challenge. The IJB-A dataset includes real-world\nunconstrained faces from 500 subjects with full pose and illumination\nvariations which are much harder than the LFW and YTF datasets. Inspired by\ntransfer learning, we train two advanced deep convolutional neural networks\n(DCNN) with two different large datasets in source domain, respectively. By\nexploring the complementarity of two distinct DCNNs, deep feature fusion is\nutilized after feature extraction in target domain. Then, template specific\nlinear SVMs is adopted to enhance the discrimination of framework. Finally,\nmultiple matching scores corresponding different templates are merged as the\nfinal results. This simple unified framework exhibits excellent performance on\nIJB-A dataset. Based on the proposed approach, we have submitted our IJB-A\nresults to National Institute of Standards and Technology (NIST) for official\nevaluation. Moreover, by introducing new data and advanced neural architecture,\nour method outperforms the state-of-the-art by a wide margin on IJB-A dataset.","url_abs":"http://arxiv.org/abs/1704.00438v2","url_pdf":"http://arxiv.org/pdf/1704.00438v2.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":"a-good-practice-towards-top-performance-of","repo_url":"https://github.com/bruinxiong/Evaluation_IJBA","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null}],"tasks":[{"task_slug":"face-recognition","task_name":"Face Recognition"},{"task_slug":"transfer-learning","task_name":"Transfer Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}