{"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/masked-relation-learning-for-deepfake","title":"Masked Relation Learning for DeepFake Detection","arxiv_id":null,"date":"2023-02-27","proceeding":"2023 2023 2","authors":["Ziming Yang","Jian Liang","Yuting Xu","Xiao-Yu Zhang","Ran He"],"abstract":"Abstract— DeepFake detection aims to differentiate falsified\r\nfaces from real ones. Most approaches formulate it as a binary\r\nclassification problem by solely mining the local artifacts and\r\ninconsistencies of face forgery, which neglect the relation across\r\nlocal regions. Although several recent works explore local relation\r\nlearning for DeepFake detection, they overlook the propagation\r\nof relational information and lead to limited performance gains.\r\nTo address these issues, this paper provides a new perspective\r\nby formulating DeepFake detection as a graph classification\r\nproblem, in which each facial region corresponds to a vertex.\r\nBut relational information with large redundancy hinders the\r\nexpressiveness of graphs. Inspired by the success of masked\r\nmodeling, we propose Masked Relation Learning which decreases\r\nthe redundancy to learn informative relational features. Specifically, a spatiotemporal attention module is exploited to learn the\r\nattention features of multiple facial regions. A relation learning\r\nmodule masks partial correlations between regions to reduce\r\nredundancy and then propagates the relational information\r\nacross regions to capture the irregularity from a global view of\r\nthe graph. We empirically discover that a moderate masking rate\r\n(e.g., 50%) brings the best performance gain. Experiments verify\r\nthe effectiveness of Masked Relation Learning and demonstrate\r\nthat our approach outperforms the state of the art by 2% AUC on\r\nthe cross-dataset DeepFake video detection. Code will be available\r\nat https://github.com/zimyang/MaskRelation.\r\nIndex Terms— Multimedia forensics, DeepFake detection,\r\nmasked learning, relation feature.","url_abs":"https://ieeexplore.ieee.org/abstract/document/10054130","url_pdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=10054130&tag=1","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":"masked-relation-learning-for-deepfake","repo_url":"https://github.com/zimyang/maskrelation","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":null},{"paper_slug":"masked-relation-learning-for-deepfake","repo_url":"https://github.com/2023-MindSpore-4/Code4/tree/main/MaskRelation-main","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":null}],"tasks":[{"task_slug":"binary-classification","task_name":"Binary Classification"},{"task_slug":"deepfake-detection","task_name":"DeepFake Detection"},{"task_slug":"face-swapping","task_name":"Face Swapping"},{"task_slug":"graph-classification","task_name":"Graph Classification"},{"task_slug":null,"task_name":"Relation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}