{"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/disentangling-3d-pose-in-a-dendritic-cnn-for","title":"Disentangling 3D Pose in A Dendritic CNN for Unconstrained 2D Face Alignment","arxiv_id":"1802.06713","date":"2018-02-19","proceeding":"CVPR 2018 6","authors":["Amit Kumar","Rama Chellappa"],"abstract":"Heatmap regression has been used for landmark localization for quite a while\nnow. Most of the methods use a very deep stack of bottleneck modules for\nheatmap classification stage, followed by heatmap regression to extract the\nkeypoints. In this paper, we present a single dendritic CNN, termed as Pose\nConditioned Dendritic Convolution Neural Network (PCD-CNN), where a\nclassification network is followed by a second and modular classification\nnetwork, trained in an end to end fashion to obtain accurate landmark points.\nFollowing a Bayesian formulation, we disentangle the 3D pose of a face image\nexplicitly by conditioning the landmark estimation on pose, making it different\nfrom multi-tasking approaches. Extensive experimentation shows that\nconditioning on pose reduces the localization error by making it agnostic to\nface pose. The proposed model can be extended to yield variable number of\nlandmark points and hence broadening its applicability to other datasets.\nInstead of increasing depth or width of the network, we train the CNN\nefficiently with Mask-Softmax Loss and hard sample mining to achieve upto\n$15\\%$ reduction in error compared to state-of-the-art methods for extreme and\nmedium pose face images from challenging datasets including AFLW, AFW, COFW and\nIBUG.","url_abs":"http://arxiv.org/abs/1802.06713v3","url_pdf":"http://arxiv.org/pdf/1802.06713v3.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":[],"tasks":[{"task_slug":"face-alignment","task_name":"Face Alignment"},{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"regression-1","task_name":"regression"}],"methods":[{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"heatmap","method_name":"Heatmap"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/face-alignment-on-cofw","task":"Face Alignment","dataset":"COFW","model":"PCD-CNNCVPR 18","rank_in_archive_order":22,"of":28,"metrics":{"NME (inter-ocular)":"5.77%"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1802.06713","atlas_url":"https://app.syntology.ai/?focus=1802.06713","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}