{"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/pose-selective-max-pooling-for-measuring","title":"Pose-Selective Max Pooling for Measuring Similarity","arxiv_id":"1609.07042","date":"2016-09-22","proceeding":null,"authors":["Xiang Xiang","Trac. D. Tran"],"abstract":"In this paper, we deal with two challenges for measuring the similarity of\nthe subject identities in practical video-based face recognition - the\nvariation of the head pose in uncontrolled environments and the computational\nexpense of processing videos. Since the frame-wise feature mean is unable to\ncharacterize the pose diversity among frames, we define and preserve the\noverall pose diversity and closeness in a video. Then, identity will be the\nonly source of variation across videos since the pose varies even within a\nsingle video. Instead of simply using all the frames, we select those faces\nwhose pose point is closest to the centroid of the K-means cluster containing\nthat pose point. Then, we represent a video as a bag of frame-wise deep face\nfeatures while the number of features has been reduced from hundreds to K.\nSince the video representation can well represent the identity, now we measure\nthe subject similarity between two videos as the max correlation among all\npossible pairs in the two bags of features. On the official 5,000 video-pairs\nof the YouTube Face dataset for face verification, our algorithm achieves a\ncomparable performance with VGG-face that averages over deep features of all\nframes. Other vision tasks can also benefit from the generic idea of employing\ngeometric cues to improve the descriptiveness of deep features.","url_abs":"http://arxiv.org/abs/1609.07042v4","url_pdf":"http://arxiv.org/pdf/1609.07042v4.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":"pose-selective-max-pooling-for-measuring","repo_url":"https://github.com/eglxiang/vgg_face","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null},{"paper_slug":"pose-selective-max-pooling-for-measuring","repo_url":"https://github.com/eglxiang/ytf","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"diversity","task_name":"Diversity"},{"task_slug":"face-recognition","task_name":"Face Recognition"},{"task_slug":"face-verification","task_name":"Face Verification"},{"task_slug":"video-similarity","task_name":"Video Similarity"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}