[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84477-en":3,"doc-seo-84477-105":29,"detail-sidebar-cat-0-en-105":90},{"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":4,"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},84477,687197100911,"Himbo","https://ap-avatar.wpscdn.com/avatar/a000239b6f1da00475?x-image-process=image/resize,m_fixed,w_180,h_180&k=1782698725881665579",8,"Research & Report","Satellite-Free Training for Drone-View Geo-Localization","Drone-view geo-localization (DVGL) retrieves geotagged satellite tiles from a reference gallery using UAV observations in GPS-denied environments, but many existing training schemes depend on satellite imagery, limiting deployment when such data are restricted or delayed. Satellite-free training (SFT) addresses this gap by learning cross-view compatible representations from multi-view drone sequences only. The pipeline reconstructs dense 3D scenes from drone views with 3D Gaussian splatting, projects geometry into pseudo-orthophotos via PCA-guided orthographic projection, completes textures using lightweight geometry-guided inpainting, then aggregates DINOv3 patch features using a Fisher vector model trained on drone data and reuses it to encode satellite tiles at test time.","arXiv :2604 .0 158 1v2 [ cs .CV] 3 Apr 2026  \nSatellite-Free Training for Drone-View Geo-Localization  \nTao Liu  \n[liutao23@njust.edu.cn](liutao23@njust.edu.cn)  \nNanjing University of Science and Technology Nanjing, Jiangsu, China  \nKan Ren  \n[k.ren@njust.edu.cn](k.ren@njust.edu.cn)  \nNanjing University of Science and Technology Nanjing, Jiangsu, China  \nYingzhi Zhang  \n[zhang_y@css.risk.tsukuba.ac.jp](zhang_y@css.risk.tsukuba.ac.jp)[ ](zhang_y@css.risk.tsukuba.ac.jp)University of Tsukuba Tsukuba, Ibaraki, Japan  \nXiaoqi Zhao  \n[xiaoqi.zhao@ntu.edu.sg](xiaoqi.zhao@ntu.edu.sg)[ ](xiaoqi.zhao@ntu.edu.sg)Nanyang Technological University Singapore, Singapore  \nSupervised DVGL  \nRequires large-scale paired drone-satellite data High cost and limited scalability  \nUnsupervised/weakly-supervised DVGL  \nDrone Satellite Unsupervised Training  \nNo strict pairing, but requires satellite imagery during training Key assumption: satellite images are accessible in advance  \nReal-world:no satellite imagery during training  \nMilitary Confidentiality Transmission Delay  \nCommercial Paid Data Overload  \nSatellite imagery: inaccessible before mission  \nBefore mission: UAV team prepares without satellite After arrival: satellite images may be released for the target area  \nOur method: Drone-View Geo-Localization without Satellite Training imagery  \nTraining (Drone-only, no satellite)  \nOrthographic UAV view (no satellite) No camera intrinsics/extrinsics required  \nUnifiled feature space via pretrained DINOv3  \nUnified embedding space UAV and satellite features of the same place are clos  \nTesting (satellite gallery available)  \nDrone Query Image  \nOrthorgaphic Image  \nOnce the satellite gallery is provided at test time We can perform effective cross-view image retrieval  \nSimilarity  \nDINOv3  \nFigure 1: Motivation and overview of the proposed satellite-free training (SFT) framework. Unlike existing DVGL methods that rely on satellite imagery during training, our method uses only multi-view drone images. The UAV sequence is reconstructed into a 3D scene and converted into a pseudo-orthophoto, which is then encoded by a frozen DINOv3 backbone. The retrieval space is constructed solely from drone-side features via drone-only GMM learning and Fisher vector aggregation, so satellite tiles are encoded at test time in a space induced entirely by drone data.  \nAbstract  \nDrone-view geo-localization (DVGL) aims to determine the location of drones in GPS-denied environments by retrieving the corresponding geotagged satellite tile from a reference gallery given UAV observations of a location. In many existing formulations, these observations are represented by a single oblique UAV image. In contrast, our satellite-free setting is designed for multi-view UAV sequences, which are used to construct a geometry-normalized  \nConference’17, Washington, DC, USA  \n2026. ACM ISBN 978-x-xxxx-xxxx-x/YYYY/MM [https://doi.org/10.1145/nnnnnnn.nnnnnnn](https://doi.org/10.1145/nnnnnnn.nnnnnnn)  \nUAV-side location representation before cross-view retrieval. Existing approaches rely on satellite imagery during training, either through paired supervision or unsupervised alignment, which limits practical deployment when satellite data are unavailable or restricted. In this paper, we propose a satellite-free training (SFT) framework that converts drone imagery into cross-view compatible representations through three main stages: drone-side 3D scene reconstruction, geometry-based pseudo-orthophoto generation, and satellite-free feature aggregation for retrieval. Specifically, we first reconstruct dense 3D scenes from multi-view drone images using 3D Gaussian splatting and project the reconstructed geometry into  \npseudo-orthophotos via PCA-guided orthographic projection. This rendering stage operates directly on reconstructed scene geometry without requiring camera parameters at rendering time. Next, we refine these orthophotos with lightweight geometry-guided inpainting to obtain","cbCaiu1w0MhjQwhN","https://ap.wps.com/l/cbCaiu1w0MhjQwhN","pdf",8687754,1,14,"English","en",105,"# Introduction\n## Problem: satellite-dependent training vs real-world constraints\n# Proposed Method: Satellite-Free Training (SFT)\n## Drone-side 3D reconstruction\n## Pseudo-orthophoto generation\n## Geometry-guided inpainting and feature aggregation\n# Experiments\n## University-1652 and SUES-200 results","[{\"question\":\"What problem does satellite-free training address in drone-view geo-localization?\",\"answer\":\"Existing DVGL methods typically require satellite imagery during training for supervision or alignment, which is often unavailable due to confidentiality, access delays, transmission limits, or data volume constraints. Satellite-free training targets accurate retrieval when satellite data cannot be used during training.\"},{\"question\":\"How does the method convert multi-view drone data into a cross-view compatible representation?\",\"answer\":\"It reconstructs dense 3D scenes from drone sequences using 3D Gaussian splatting, renders geometry-normalized pseudo-orthophotos via PCA-guided orthographic projection, then uses geometry-guided inpainting to complete textures. DINOv3 features extracted from the generated orthophotos are aggregated for retrieval.\"},{\"question\":\"What is the key idea behind encoding satellite tiles at test time without using satellite data during training?\",\"answer\":\"A Fisher vector aggregation model is learned solely from drone-side data, and the learned aggregation space is used to encode satellite tiles when the satellite gallery becomes available at test time. This enables effective cross-view image retrieval despite the absence of satellite training imagery.\"}]",1784195926,35,{"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":85,"head_meta":87,"extra_data":89,"updated_unix":27},"satellite-free-training-for-drone-view-geo-localization","",{"@graph":35,"@context":84},[36,53,67],{"@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/satellite-free-training-for-drone-view-geo-localization/84477/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":61,"encodingFormat":60,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":4},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"What problem does satellite-free training address in drone-view geo-localization?","Question",{"text":74,"@type":75},"Existing DVGL methods typically require satellite imagery during training for supervision or alignment, which is often unavailable due to confidentiality, access delays, transmission limits, or data volume constraints. Satellite-free training targets accurate retrieval when satellite data cannot be used during training.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How does the method convert multi-view drone data into a cross-view compatible representation?",{"text":79,"@type":75},"It reconstructs dense 3D scenes from drone sequences using 3D Gaussian splatting, renders geometry-normalized pseudo-orthophotos via PCA-guided orthographic projection, then uses geometry-guided inpainting to complete textures. DINOv3 features extracted from the generated orthophotos are aggregated for retrieval.",{"name":81,"@type":72,"acceptedAnswer":82},"What is the key idea behind encoding satellite tiles at test time without using satellite data during training?",{"text":83,"@type":75},"A Fisher vector aggregation model is learned solely from drone-side data, and the learned aggregation space is used to encode satellite tiles when the satellite gallery becomes available at test time. This enables effective cross-view image retrieval despite the absence of satellite training imagery.","https://schema.org",{"og:url":51,"og:type":86,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":88,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":91},[92,96,100,104,109,114,119,122,127,130,134],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":45,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":45,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":45,"category_name":136,"show_sort_weight":105,"slug":137},19,"General","general"]