[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-272562-105":3,"detail-sidebar-cat-0-en-105":81,"doc-detail-272562-en":130},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":74,"head_meta":76,"extra_data":78,"updated_unix":80},105,"en","self-supervised-re-identification-for-online-joint-multi-object-tracking","Self-supervised re-identification for online joint multi-object tracking","","Multi-object tracking bottlenecks are increasingly driven by association quality rather than detection accuracy, yet association research commonly depends on identity labels that are more costly than detection labels. A self-supervised re-identification module for online joint multi-object tracking (SR-MOT) is proposed to eliminate identity-label requirements. An appearance discriminator judges identities using only detection hypotheses, while negative pairs come from detections within the same frame and positive pairs are formed via box-level augmentations. Dynamic loss re-weighting balances detection and re-ID tasks for stable optimization.",{"@graph":14,"@context":73},[15,34,56],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/self-supervised-re-identification-for-online-joint-multi-object-tracking/272562/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/self-supervised-re-identification-for-online-joint-multi-object-tracking/272562.png","ImageObject",300,407,{"name":42,"@type":43},"Aran","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-11-05","2026-09-15",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",9,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"What problem does the paper address in multi-object tracking research?","Question",{"text":63,"@type":64},"It addresses the shift of the MOT bottleneck from detection performance to association performance, where association typically requires expensive identity labels.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"How does SR-MOT avoid using identity labels for training?",{"text":68,"@type":64},"It constructs negative pairs from detections within the same video frame and builds positive pairs through box-level data augmentations, then trains an appearance discriminator without identity labels.",{"name":70,"@type":61,"acceptedAnswer":71},"What mechanism balances detection and re-identification during optimization?",{"text":72,"@type":64},"The method uses different output features and dynamically adjusts detection and re-ID loss weights based on the information content of the loss distribution to improve balance.","https://schema.org",{"og:url":32,"og:type":75,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":77,"canonical":32},"index,follow",{"doc_id":79,"site_id":7},272562,1789466089,{"code":4,"msg":82,"data":83},"success",[84,88,92,96,101,106,111,115,119,122,126],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":85,"show_sort_weight":86,"slug":87},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":89,"show_sort_weight":90,"slug":91},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":93,"show_sort_weight":94,"slug":95},"Exam",70,"exam",{"id":97,"doc_module":4,"doc_module_name":25,"category_name":98,"show_sort_weight":99,"slug":100},5,"Comic",60,"comic",{"id":102,"doc_module":4,"doc_module_name":25,"category_name":103,"show_sort_weight":104,"slug":105},6,"Technology",50,"technology",{"id":107,"doc_module":4,"doc_module_name":25,"category_name":108,"show_sort_weight":109,"slug":110},7,"Healthcare",40,"healthcare",{"id":112,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":113,"slug":114},8,30,"research-report",{"id":55,"doc_module":4,"doc_module_name":25,"category_name":116,"show_sort_weight":117,"slug":118},"Religion & Spirituality",20,"religion-spirituality",{"id":117,"doc_module":4,"doc_module_name":25,"category_name":120,"show_sort_weight":117,"slug":121},"World Cup","world-cup",{"id":123,"doc_module":4,"doc_module_name":25,"category_name":124,"show_sort_weight":123,"slug":125},10,"Lifestyle","lifestyle",{"id":127,"doc_module":4,"doc_module_name":25,"category_name":128,"show_sort_weight":97,"slug":129},19,"General","general",{"code":4,"msg":82,"data":131},{"doc_id":79,"user_id":132,"nickname":42,"user_avatar":133,"doc_module":4,"category_id":112,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":139,"language":140,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":12,"update_tm":80,"read_time":144,"preview_page_count":4},137455076865,"https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0","Knowledge and Information Systems  \n[https://doi.org/10.1007/s101](https://doi.org/10.1007/s101)15-024-02237-w  \nREGULAR PAPER  \nSelf-supervised re-identification for online joint multi-object tracking  \nShuman Li1 · Longqi Yang2 · Huibin Tan1 · Binglin Wang1 · Wanrong Huang1 · Hengzhu Liu1 · Wenjing Yang1 · Long Lan1  \nReceived: 9 June 2023 / Revised: 30 August 2024 / Accepted: 11 September 2024 © The Author(s) 2024  \nAbstract  \nRecently, the bottleneck of multi-object tracking is shifting from detection performance to association performance. However, research on association algorithms requires a large number of identity labels, which are more expensive than detection labels. To circumvent the need for identity labels, we propose a Self-supervised Re-identification module for online joint Multi-Object Tracking (SR-MOT) . Specifically, we design an appearance discriminator to judge identities based solely on detection hypotheses and then associate the same identity with the final trajectory. To train the discriminator without using identity labels, we construct negative pairs by the detections that appear in the same video frame, as they definitely belong to different identities. Positive pairs are naturally constructed through several useful data augmentation strategies at the box level. In addition, our proposed method balances conflicting detection and re-ID tasks by using different output features and dynamically adjusts detection and re-ID loss weights based on the information content of the loss distribution to promote balance between the two tasks from the feature level and optimization methods.  \nB Wanrong Huang [huangwanrong12@nudt.edu.cn](huangwanrong12@nudt.edu.cn)  \nShuman Li  \n[lishuman13@nudt.edu.cn](lishuman13@nudt.edu.cn)  \nLongqi Yang  \n[yanglongqi_1988@qq.com](yanglongqi_1988@qq.com)  \nHuibin Tan  \n[tanhuibin2815@163.com](tanhuibin2815@163.com)  \nBinglin Wang  \n[wangbinglin14@nudt.edu.cn](wangbinglin14@nudt.edu.cn)  \nHengzhu Liu  \n[hengzhu_liu@263.net](hengzhu_liu@263.net)  \nWenjing Yang  \n[wenjing.yang@nudt.edu.cn](wenjing.yang@nudt.edu.cn)  \nLong Lan  \n[long.lan@nudt.edu.cn](long.lan@nudt.edu.cn)  \n1 College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, China  \n2 Intelligent Game and Decision Lab, Beijing 100073, China  \n1 3  \nIn our evaluation on the MOT Challenge benchmark, we show that our SR-MOT performs comparably to supervised methods and is significantly superior to other unsupervised methods. Our proposed method provides a practical solution for multi-object tracking without the need for identity labels, making it more accessible for real-world applications.  \nKeywords Multi-object tracking · Self-supervised learning · Re-identification · Box-level augmentation · Adaptive weight adjustment mechanism  \n1 Introduction  \nMulti-object tracking (MOT) is an important problem in the field of computer vision, which is widely used in fields such as safety monitoring, autonomous driving, human–machine interface, and medical image processing [1, 2] . The multi-object tracking method needs to accomplish two tasks: first, it needs to detect constantly changing objects dynamically (detection), and second, it needs to maintain the identity consistency of different objects (tracking) . Some works [3, 4] have introduced re-identification (re-ID) features [5, 6] into multi-object tracking, focusing on differences in the fine-grained appearance of objects and achieving better tracking. Therefore, in many multi-object tracking studies, tracking is regarded as a re-ID task. Depending on whether a single network is used for both detection and re-ID, as shown in Fig. 1, existing multi-object tracking (MOT) methods can be divided into two categories: joint detection and tracking (JDT) methods and tracking-by-detection (TBD) methods.  \nTBD methods usually outperform their joint counterparts because they use state-of-theart models [7–9] deliberately designed for detection to pre-calculat","cbCaiaEI9fUySAX9","https://ap.wps.com/l/cbCaiaEI9fUySAX9","pdf",12022026,34,"English","# Abstract\n# Keywords\n# 1 Introduction\n## Multi-object tracking problem and tasks\n## Re-identification in MOT and method categories\n## Limitations of TBD and benefits of JDT\n## Need for annotated data and motivation for unsupervised learning","[{\"question\":\"What problem does the paper address in multi-object tracking research?\",\"answer\":\"It addresses the shift of the MOT bottleneck from detection performance to association performance, where association typically requires expensive identity labels.\"},{\"question\":\"How does SR-MOT avoid using identity labels for training?\",\"answer\":\"It constructs negative pairs from detections within the same video frame and builds positive pairs through box-level data augmentations, then trains an appearance discriminator without identity labels.\"},{\"question\":\"What mechanism balances detection and re-identification during optimization?\",\"answer\":\"The method uses different output features and dynamically adjusts detection and re-ID loss weights based on the information content of the loss distribution to improve balance.\"}]","Self-supervised re-identification for online joint multi-object tracking | PDF",86]