{"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/wing-loss-for-robust-facial-landmark","title":"Wing Loss for Robust Facial Landmark Localisation with Convolutional Neural Networks","arxiv_id":"1711.06753","date":"2017-11-17","proceeding":"CVPR 2018 6","authors":["Zhen-Hua Feng","Josef Kittler","Muhammad Awais","Patrik Huber","Xiao-Jun Wu"],"abstract":"We present a new loss function, namely Wing loss, for robust facial landmark\nlocalisation with Convolutional Neural Networks (CNNs). We first compare and\nanalyse different loss functions including L2, L1 and smooth L1. The analysis\nof these loss functions suggests that, for the training of a CNN-based\nlocalisation model, more attention should be paid to small and medium range\nerrors. To this end, we design a piece-wise loss function. The new loss\namplifies the impact of errors from the interval (-w, w) by switching from L1\nloss to a modified logarithm function.\n  To address the problem of under-representation of samples with large\nout-of-plane head rotations in the training set, we propose a simple but\neffective boosting strategy, referred to as pose-based data balancing. In\nparticular, we deal with the data imbalance problem by duplicating the minority\ntraining samples and perturbing them by injecting random image rotation,\nbounding box translation and other data augmentation approaches. Last, the\nproposed approach is extended to create a two-stage framework for robust facial\nlandmark localisation. The experimental results obtained on AFLW and 300W\ndemonstrate the merits of the Wing loss function, and prove the superiority of\nthe proposed method over the state-of-the-art approaches.","url_abs":"http://arxiv.org/abs/1711.06753v5","url_pdf":"http://arxiv.org/pdf/1711.06753v5.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":"wing-loss-for-robust-facial-landmark","repo_url":"https://github.com/mikgur/MADE_CV_1000_facial_landmarks","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"wing-loss-for-robust-facial-landmark","repo_url":"https://github.com/mktoid/made-thousand-facial-landmarks","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"wing-loss-for-robust-facial-landmark","repo_url":"https://github.com/noelmat/facelandmarks_wingloss","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null},{"paper_slug":"wing-loss-for-robust-facial-landmark","repo_url":"https://github.com/parthapratimbanik/facial-keypoint-detection-udacity-ppb","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"wing-loss-for-robust-facial-landmark","repo_url":"https://github.com/xialuxi/arcface-caffe","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null},{"paper_slug":"wing-loss-for-robust-facial-landmark","repo_url":"https://github.com/open-mmlab/mmpose","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"data-augmentation","task_name":"Data Augmentation"},{"task_slug":"face-alignment","task_name":"Face Alignment"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/face-alignment-on-300w","task":"Face Alignment","dataset":"300W","model":"Wing","rank_in_archive_order":47,"of":48,"metrics":{"NME_inter-pupil (%, Challenge)":"7.18","NME_inter-pupil (%, Common)":"3.27","NME_inter-pupil (%, Full)":"4.04"},"uses_additional_data":false},{"leaderboard":"/sota/face-alignment-on-aflw-19","task":"Face Alignment","dataset":"AFLW-19","model":"Wing","rank_in_archive_order":14,"of":23,"metrics":{"AUC_box@0.07 (%, Full)":"53.5","NME_box (%, Full)":"3.56","NME_diag (%, Full)":"1.65"},"uses_additional_data":false},{"leaderboard":"/sota/face-alignment-on-cofw","task":"Face Alignment","dataset":"COFW","model":"Wing (Feng et al., 2018)","rank_in_archive_order":20,"of":28,"metrics":{"NME (inter-ocular)":"5.07"},"uses_additional_data":false},{"leaderboard":"/sota/face-alignment-on-wflw","task":"Face Alignment","dataset":"WFLW","model":"Wing","rank_in_archive_order":27,"of":36,"metrics":{"AUC@10 (inter-ocular)":"55.4","FR@10 (inter-ocular)":"6.00","NME (inter-ocular)":"5.11"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1711.06753","atlas_url":"https://app.syntology.ai/?focus=1711.06753","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}