[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85286-en":3,"doc-seo-85286-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},85286,687197207919,"Theodora","https://ap-avatar.wpscdn.com/avatar/a000253d6f5f7c60be?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779446848396160552",8,"Research & Report","WiFi-JEPA Self-supervised Learning for WiFi-CSI 3D Human Pose Estimation","WiFi Channel State Information (CSI) enables privacy-preserving human pose sensing in camera-denied environments, yet existing WiFi-based pose estimators degrade under environmental shifts and are limited by costly camera-based annotation pipelines. WiFi-JEPA introduces a self-supervised framework that learns CSI-native representations by predicting masked latent embeddings rather than reconstructing raw CSI. Contributions include CSI-specific tokenization with Tx–Rx link masking, a ray-tracing simulation pipeline for unlabeled pre-training data, and state-of-the-art results on Person-in-WiFi-3D, with simulated CSI providing complementary signal.","arXiv :2607 . 11064v1 [ cs .CV] 13 Jul 2026  \nWiFi-JEPA: Self-supervised Learning for WiFi-CSI 3D Human Pose Estimation  \nDoeon Kim 1 ⋆, Jungyoon Lee2⋆, Seongsin Kim3 B, and Seong-heum Kim 1  \n1 Department of Intelligent Semiconductors, Soongsil University, Republic of Korea  \n2 Department of AI Convergence Security, Soongsil University, Republic of Korea  \n3 School of AI Software, Soongsil University, Republic of Korea {ilsin205, [jungyoon}@soongsil.ac.kr](jungyoon}@soongsil.ac.kr), {kss0222, [seongheum}@ssu.ac.kr](seongheum}@ssu.ac.kr)  \nAbstract. WiFi Channel State Information (CSI) enables privacy-preserving human pose sensing in camera-denied environments, but existing WiFi-based pose estimators often fail under environment shifts and rely on costly camera-based annotation pipelines that limit scale. We propose WiFi-JEPA, a self-supervised framework that learns CSI-native representations by predicting masked latent embeddings instead of reconstructing raw CSI signals that may contain hardwarespecific artifacts. WiFi-JEPA makes three contributions: (i) CSI-specific tokenization and link masking tailored to the CSI tensor over channel, time, and link (C, T, L); masking entire Tx–Rx antenna links forces the model to predict one spatial link view from others, capturing cross-link correlations informative of 3D spatial structure. (ii) A ray-tracing CSI simulation pipeline that generates diverse unlabeled CSI from randomized geometric primitives, providing scalable pre-training data without pose annotations. (iii) State-of-the-art results on Person-in-WiFi-3D:  \nWiFi-JEPA outperforms prior WiFi-CSI baselines on both single- and multi-person 3D pose estimation under the same evaluation protocol.  \nWe also show that simulated CSI provides complementary pre-training signal to real CSI, and that four vision-native SSL objectives degrade performance below training from scratch, whereas WiFi-JEPA consistently improves downstream pose estimation.  \nKeywords: WiFi Sensing · 3D Human Pose Estimation · Synthetic Data · Self-supervised learning  \n1 Introduction  \nHuman pose estimation (HPE) is a fundamental component of systems for health monitoring, human–computer interaction, and safety-critical perception. Despite rapid progress in camera-based HPE under line-of-sight conditions [20, 21, 27], vision is often limited in camera-denied scenarios: (i) occlusion by walls or obstacles, (ii) low or no illumination, and (iii) privacy and regulatory constraints that prohibit visual capture in sensitive spaces.  \n⋆ These authors contributed equally to this work.  \n2 D. Kim, J. Lee et al.  \nFig. 1: Overall framework of WiFi-JEPA. Left: Pre-training data — sim-object and real CSI from PiW3D. Center: Generated CSI input and WiFi-JEPA. Right: GT and predicted 3D poses.  \nWiFi channel state information (CSI) provides a compelling alternative. WiFi signals are ubiquitous indoors and can propagate through many common obstructions. Human motion modulates multipath propagation, which is captured as CSI—a factored, complex-valued measurement over subcarriers (frequency), time, and multiple Tx–Rx antenna links. Prior work has shown that WiFi sensing can recover human-centric outputs without visual imagery, including fine-grained person perception and pose-related estimates [19, 23, 28] . Most notably, PiW3D [28] demonstrates the feasibility of multi-person 3D pose estimation with commodity WiFi even under visual occlusion, reporting a ∼90K-frame dataset collected in multiple indoor areas and 3D joint localization errors on the order of ∼ 100 mm.  \nHowever, robust WiFi-based HPE still faces three bottlenecks. First, crossdomain generalization is fragile: performance can degrade when deployment conditions differ from training, such as changes in transceiver placement, surrounding objects and furniture. Second, label scalability is limited: CSI 3D pose datasets typically rely on camera-based annotation pipelines and are collected in a small number of rooms wi","cbCaieO5UYCbl5hB","https://ap.wps.com/l/cbCaieO5UYCbl5hB","pdf",2172110,1,17,"English","en",105,"# Abstract\n# Introduction\n# Related","[{\"question\":\"What problem does WiFi-JEPA address in WiFi-based 3D human pose estimation?\",\"answer\":\"It targets cross-domain fragility, limited scalability due to camera-based labels, and hardware-dependent noise in CSI that cause pose estimators to underperform when conditions differ from training.\"},{\"question\":\"How does WiFi-JEPA learn representations without reconstructing raw CSI?\",\"answer\":\"WiFi-JEPA uses CSI-specific tokenization and predicts masked latent embeddings, avoiding reconstruction of raw signals that may carry hardware-specific artifacts.\"},{\"question\":\"How are training data and supervision scaled in WiFi-JEPA?\",\"answer\":\"It introduces a ray-tracing CSI simulation pipeline that generates diverse unlabeled CSI from randomized geometric primitives, reducing reliance on scarce pose annotations.\"}]",1784202258,43,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"wifi-jepa-self-supervised-learning-for-wifi-csi-3d-human-pose-estimation","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/wifi-jepa-self-supervised-learning-for-wifi-csi-3d-human-pose-estimation/85286/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does WiFi-JEPA address in WiFi-based 3D human pose estimation?","Question",{"text":75,"@type":76},"It targets cross-domain fragility, limited scalability due to camera-based labels, and hardware-dependent noise in CSI that cause pose estimators to underperform when conditions differ from training.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does WiFi-JEPA learn representations without reconstructing raw CSI?",{"text":80,"@type":76},"WiFi-JEPA uses CSI-specific tokenization and predicts masked latent embeddings, avoiding reconstruction of raw signals that may carry hardware-specific artifacts.",{"name":82,"@type":73,"acceptedAnswer":83},"How are training data and supervision scaled in WiFi-JEPA?",{"text":84,"@type":76},"It introduces a ray-tracing CSI simulation pipeline that generates diverse unlabeled CSI from randomized geometric primitives, reducing reliance on scarce pose annotations.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]