{"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/kepler-keypoint-and-pose-estimation-of","title":"KEPLER: Keypoint and Pose Estimation of Unconstrained Faces by Learning Efficient H-CNN Regressors","arxiv_id":"1702.05085","date":"2017-02-16","proceeding":null,"authors":["Amit Kumar","Azadeh Alavi","Rama Chellappa"],"abstract":"Keypoint detection is one of the most important pre-processing steps in tasks\nsuch as face modeling, recognition and verification. In this paper, we present\nan iterative method for Keypoint Estimation and Pose prediction of\nunconstrained faces by Learning Efficient H-CNN Regressors (KEPLER) for\naddressing the face alignment problem. Recent state of the art methods have\nshown improvements in face keypoint detection by employing Convolution Neural\nNetworks (CNNs). Although a simple feed forward neural network can learn the\nmapping between input and output spaces, it cannot learn the inherent\nstructural dependencies. We present a novel architecture called H-CNN\n(Heatmap-CNN) which captures structured global and local features and thus\nfavors accurate keypoint detecion. HCNN is jointly trained on the visibility,\nfiducials and 3D-pose of the face. As the iterations proceed, the error\ndecreases making the gradients small and thus requiring efficient training of\nDCNNs to mitigate this. KEPLER performs global corrections in pose and\nfiducials for the first four iterations followed by local corrections in the\nsubsequent stage. As a by-product, KEPLER also provides 3D pose (pitch, yaw and\nroll) of the face accurately. In this paper, we show that without using any 3D\ninformation, KEPLER outperforms state of the art methods for alignment on\nchallenging datasets such as AFW and AFLW.","url_abs":"http://arxiv.org/abs/1702.05085v1","url_pdf":"http://arxiv.org/pdf/1702.05085v1.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":"head-pose-estimation","task_name":"Head Pose Estimation"},{"task_slug":"keypoint-detection","task_name":"Keypoint Detection"},{"task_slug":"keypoint-estimation","task_name":"Keypoint Estimation"},{"task_slug":"pose-estimation","task_name":"Pose Estimation"},{"task_slug":"pose-prediction","task_name":"Pose Prediction"}],"methods":[{"method_slug":"convolution","method_name":"Convolution"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/head-pose-estimation-on-biwi","task":"Head Pose Estimation","dataset":"BIWI","model":"KEPLER","rank_in_archive_order":20,"of":29,"metrics":{"MAE (trained with other data)":"13.852"},"uses_additional_data":false}],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}