[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-81768-en":3,"doc-seo-81768-105":30,"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":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},81768,4398048949847,"Eliana","https://ap-avatar.wpscdn.com/avatar/400002536579ef2da7f?_k=1778318612642679267",8,"Research & Report","DroneFINE Domain Aware Parameter Efficient Fine Tuning of Vision Language Detectors for Drone Images","UAV object detection in open and dynamic environments remains challenging because Vision-Language Models face a large domain gap between their pretraining data and aerial imagery. Existing parameter-efficient fine-tuning strategies struggle: natural-scene, foreground-dominant visual priors conflict with UAV data patterns dominated by bird’s-eye views, background clutter, and small objects. DroneFINE introduces a PEFT paradigm with domain-aware modules, including HyperAdapter for data-dependent foreground adaptation and SemanticGate for background suppression using text-conditioned guidance and contrastive learning. Experiments on VisDrone and UAVDT show strong gains over prior PEFT while using far fewer trainable parameters.","arXiv :2607 .00338v 1 [ cs .CV] 1 Jul 2026  \nDroneFINE: Domain-Aware Parameter-Efficient Fine-Tuning of Vision-Language Detectors for  \nDrone Images  \nKe Wu 1 ,2 ,5 ,6, Yanan Zhang3, Yingjie Gao2, Wenhao Li2, Chenyu Zhou2, Xinzhu Ma2, Jiaxin Chen⋆2 ,4 , and Di Huang2  \n1 National College for Excellent Engineers, Beihang University, Beijing, China  \n2 School of Computer Science and Engineering, Beihang University, Beijing, China  \n3 School of Computer Science and Information Engineering, Hefei University of Technology, Hefei, China  \n4 State Key Laboratory of Virtual Reality Technology and Systems, Beihang  \nUniversity, Beijing, China  \n5 Innovation Center for Intelligent System Cognition and Decision-Making, Beijing, China  \n6 Information Science Academy of China Electronics Technology Group Corporation, Beijing, China  \n{20373041,[jiaxinchen}@buaa.edu.cn](jiaxinchen}@buaa.edu.cn)[ ](jiaxinchen}@buaa.edu.cn)Abstract. Object detection for Unmanned Aerial Vehicles (UAVs) working in open and dynamic environments is a highly challenging task. While Vision-Language Models (VLMs) have offered a powerful solution for universal object detection, adapting them to UAV scenarios remains non-trivial due to a substantial domain gap between VLM pre-training data and aerial imagery. The prevailing Parameter-Efficient Fine-Tuning (PEFT) methods prove ineffective in bridging this gap, as VLMs’“naturalscene, foreground-dominant” visual priors misalign with the “bird’s-eyeview, background-dominant, small-object” characteristics of UAV data.  \nTo address this issue, we propose DroneFINE, a novel PEFT paradigm comprising two domain-aware complementary modules tailored for VLMbased drone image detectors. Specifically, a data-dependent, foregroundaware, and multi-path adaptation mechanism named HyperAdapter is designed, which overcomes the static structural constraints of PEFT. In addition, a background suppression algorithm named SemanticGate is developed, which is a text-conditioned guidance strategy that employs background vocabulary to actively guide the model in suppressing responses from irrelevant regions. Extensive experiments on VisDrone and UAVDT demonstrate that DroneFINE significantly outperforms existing PEFT methods and achieves performance comparable to full fine-tuning while substantially reducing the fine-tuned parameters.  \nKeywords: Object Detection · UAV · VLM · PEFT  \n1 Introduction  \nUAVs are playing an increasingly vital role in diverse applications such as intelligent inspection, emergency rescue, and logistics delivery, where object detec-  \n⋆ Corresponding author.  \n2 Wu et al.  \nFig. 1: Analysis of Domain Gap and Adaptation Bottlenecks for UAVs. (a)  \nComparison of image characteristics between common ground-level datasets [24] and aerial datasets [6] . (b) Rank-Energy(∥A∥2F) plot of FFN layers’weights after Full Fine-Tuning (Full FT) . (c) Feature similarity comparison between Full FT, standard LoRA [14](dashed line), and our method (solid line) .  \ntion [3,7,8] serves as a critical perception task to enable these functions [30,37, 39,45,48] . Recently, VLMs [3,21,29] have shown remarkable generalization and strong adaptability in universal object detection. Therefore, applying the broad knowledge of VLMs to drone detectors is expected to strengthen the robust perceptual capabilities of UAVs operating in complex and dynamic environments.  \nHowever, a fundamental domain gap between VLM pre-training data and aerial imagery prevents VLMs from leveraging their inherent strengths when directly applied to UAV scenarios. As shown in Fig. 1(a), the visual priors from foreground-dominant data [10, 21, 36, 38] generalize poorly to UAV imagery, which typically features bird’s-eye-view and small objects in backgrounddominant scenes. While full fine-tuning or retraining [23, 33] might boost performance on UAV data, they often cause overfitting, representation collapse, or  \nhigh computational costs, defeating our original purpose of u","cbCaioTIigjLQs28","https://ap.wps.com/l/cbCaioTIigjLQs28","pdf",12543527,3,1,18,"English","en",105,"# Introduction\n## Domain gap in UAV perception\n## Limitations of standard PEFT\n## Proposed DroneFINE approach\n### HyperAdapter\n### SemanticGate\n# Experimental results","[{\"question\":\"为什么直接将Vision-Language Models用于无人机场景会遇到困难？\",\"answer\":\"VLM预训练数据与航拍图像之间存在显著领域差距，视觉先验（自然场景与前景占主导）与无人机数据的鸟瞰视角、背景占主导以及小目标特征不匹配。\"},{\"question\":\"文中指出的标准PEFT方法主要有哪些不足？\",\"answer\":\"其一，标准PEFT在低秩空间更新，适配能力有限，难以覆盖无人机任务与更深层的高复杂度需求；其二，无人机图像背景主导带来噪声，难以让模型把注意力更集中在前景区域。\"},{\"question\":\"DroneFINE如何在参数高效微调的同时提升无人机检测效果？\",\"answer\":\"DroneFINE包含两个领域感知模块：HyperAdapter通过数据依赖的前景感知多路径适配增强表示能力；SemanticGate利用基于文本的背景抑制与对比学习，引导模型抑制无关区域响应，从而达到接近全量微调的性能并显著减少训练参数。\"}]",1784176016,45,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"dronefine-domain-aware-parameter-efficient-fine-tuning-of-vision-language-detectors-for-drone-images","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":20},"https://docshare.wps.com/document/research-report/",{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/dronefine-domain-aware-parameter-efficient-fine-tuning-of-vision-language-detectors-for-drone-images/81768/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-25","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},"为什么直接将Vision-Language Models用于无人机场景会遇到困难？","Question",{"text":75,"@type":76},"VLM预训练数据与航拍图像之间存在显著领域差距，视觉先验（自然场景与前景占主导）与无人机数据的鸟瞰视角、背景占主导以及小目标特征不匹配。","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"文中指出的标准PEFT方法主要有哪些不足？",{"text":80,"@type":76},"其一，标准PEFT在低秩空间更新，适配能力有限，难以覆盖无人机任务与更深层的高复杂度需求；其二，无人机图像背景主导带来噪声，难以让模型把注意力更集中在前景区域。",{"name":82,"@type":73,"acceptedAnswer":83},"DroneFINE如何在参数高效微调的同时提升无人机检测效果？",{"text":84,"@type":76},"DroneFINE包含两个领域感知模块：HyperAdapter通过数据依赖的前景感知多路径适配增强表示能力；SemanticGate利用基于文本的背景抑制与对比学习，引导模型抑制无关区域响应，从而达到接近全量微调的性能并显著减少训练参数。","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":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]