{"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/pre-training-strategies-and-datasets-for","title":"Pre-training strategies and datasets for facial representation learning","arxiv_id":"2103.16554","date":"2021-03-30","proceeding":null,"authors":["Adrian Bulat","Shiyang Cheng","Jing Yang","Andrew Garbett","Enrique Sanchez","Georgios Tzimiropoulos"],"abstract":"What is the best way to learn a universal face representation? Recent work on Deep Learning in the area of face analysis has focused on supervised learning for specific tasks of interest (e.g. face recognition, facial landmark localization etc.) but has overlooked the overarching question of how to find a facial representation that can be readily adapted to several facial analysis tasks and datasets. To this end, we make the following 4 contributions: (a) we introduce, for the first time, a comprehensive evaluation benchmark for facial representation learning consisting of 5 important face analysis tasks. (b) We systematically investigate two ways of large-scale representation learning applied to faces: supervised and unsupervised pre-training. Importantly, we focus our evaluations on the case of few-shot facial learning. (c) We investigate important properties of the training datasets including their size and quality (labelled, unlabelled or even uncurated). (d) To draw our conclusions, we conducted a very large number of experiments. Our main two findings are: (1) Unsupervised pre-training on completely in-the-wild, uncurated data provides consistent and, in some cases, significant accuracy improvements for all facial tasks considered. (2) Many existing facial video datasets seem to have a large amount of redundancy. We will release code, and pre-trained models to facilitate future research.","url_abs":"https://arxiv.org/abs/2103.16554v2","url_pdf":"https://arxiv.org/pdf/2103.16554v2.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":"pre-training-strategies-and-datasets-for","repo_url":"https://github.com/1adrianb/unsupervised-face-representation","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":null},{"paper_slug":"pre-training-strategies-and-datasets-for","repo_url":"https://github.com/tomas-gajarsky/facetorch","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"3d-face-reconstruction","task_name":"3D Face Reconstruction"},{"task_slug":"3d-facial-landmark-localization","task_name":"3D Facial Landmark Localization"},{"task_slug":"arousal-estimation","task_name":"Arousal Estimation"},{"task_slug":"emotion-recognition","task_name":"Emotion Recognition"},{"task_slug":"face-alignment","task_name":"Face Alignment"},{"task_slug":"face-recognition","task_name":"Face Recognition"},{"task_slug":"facial-action-unit-detection","task_name":"Facial Action Unit Detection"},{"task_slug":"facial-expression-recognition","task_name":"Facial Expression Recognition (FER)"},{"task_slug":"few-shot-learning","task_name":"Few-Shot Learning"},{"task_slug":"representation-learning","task_name":"Representation Learning"},{"task_slug":"unsupervised-pre-training","task_name":"Unsupervised Pre-training"},{"task_slug":"valence-estimation","task_name":"Valence Estimation"},{"task_slug":null,"task_name":"Valene Estimation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/3d-face-reconstruction-on-aflw2000-3d","task":"3D Face Reconstruction","dataset":"AFLW2000-3D","model":"VGG-F","rank_in_archive_order":2,"of":8,"metrics":{"Mean NME ":"3.42%"},"uses_additional_data":false},{"leaderboard":"/sota/face-alignment-on-aflw-19","task":"Face Alignment","dataset":"AFLW-19","model":"VGG-F","rank_in_archive_order":10,"of":23,"metrics":{"NME_diag (%, Full)":"1.55"},"uses_additional_data":true},{"leaderboard":"/sota/face-alignment-on-cofw","task":"Face Alignment","dataset":"COFW","model":"Ours (VGG-F)","rank_in_archive_order":10,"of":28,"metrics":{"NME (inter-ocular)":"3.32"},"uses_additional_data":false},{"leaderboard":"/sota/face-alignment-on-wfw-extra-data","task":"Face Alignment","dataset":"WFW (Extra Data)","model":"VGG-F","rank_in_archive_order":11,"of":11,"metrics":{"NME (inter-ocular)":"4.57"},"uses_additional_data":true},{"leaderboard":"/sota/facial-expression-recognition-on-bp4d","task":"Facial Expression Recognition (FER)","dataset":"BP4D","model":"Ours (VGG-F)","rank_in_archive_order":2,"of":2,"metrics":{"ICC":"0.719"},"uses_additional_data":false},{"leaderboard":"/sota/facial-expression-recognition-on-disfa","task":"Facial Expression Recognition (FER)","dataset":"DISFA","model":"Ours (VGG-F)","rank_in_archive_order":2,"of":2,"metrics":{"ICC":"0.598"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/2103.16554","atlas_url":"https://app.syntology.ai/?focus=2103.16554","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}