[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85813-en":3,"doc-seo-85813-105":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},85813,8796095461564,"Liam","https://ap-avatar.wpscdn.com/davatar_155a257f0dc6eb9ab79c44ca47cae57d",8,"Research & Report","CVKD-UDA: Cross-View Knowledge Distillation for 3D Unsupervised Domain Adaptive Segmentation","3D unsupervised domain adaptive (UDA) segmentation addresses the high cost of manual annotations for new-domain point cloud data. Self-training is widely used, yet its performance depends on a reliable warm-up model that produces accurate pseudo labels. Existing warm-up strategies often rely on source supervision or adversarial output alignment, leading to limited generalization and unstable training under large domain gaps. CVKD-UDA revisits voxel size to construct domain-similar representations and applies cross-view knowledge distillation with a Decouple-Adapter and an auxiliary imitation classifier to balance transferability and discriminability, improving benchmark performance.","CVKD-UDA: Cross-View Knowledge Distillation for 3D Unsupervised  \nDomain Adaptive Segmentation  \nZhimin Yuan, Ming Cheng, Member, IEEE, Shangshu Yu, Wen Li, Dunqiang Liu, Xin Huang  \nand Cheng Wang, Senior Member, IEEE  \narXiv :2607 . 10087v1 [ cs .CV] 11 Jul 2026  \nAbstract—3D unsupervised domain adaptive (UDA) segmentation mitigates the high cost of manual annotations of the new domain data. Self-training has emerged as the dominant approach in this area, where its success heavily depends on a well-initialized warm-up model to generate reliable pseudo labels. However, existing methods often depend on source supervision or output-level adversarial alignment to obtain the warm-up model, which suffer from limited generalization and training instability due to the large domain gap between domains. Constructing domain-similar representations is an effective way to bridge this gap. In this work, we propose CVKD-UDA, which revisits voxel size as a core design factor to construct domain-similar representations and leverages cross-view complementary cues to balance transferability and discriminability of the warm-up model. First, we generate two complementary views by varying voxel sizes and introduce a cross-view knowledge distillation (CVKD) to enhance generalization and target perception of the model. Second, to balance transferability and discriminability, we design a lightweight Decouple-Adapter and an auxiliary imitation classifier to decouple cross-view knowledge transfer. Extensive experiments on two benchmarks demonstrate that CVKD-UDA effectively improves the performance of self-training methods and provides a new perspective for 3D UDA segmentation. Our code will be available at GitHub.  \nIndex Terms—Unsupervised domain adaptation, point cloud, cross-view knowledge distillation, semantic segmentation.  \nI. INTRODUCTION  \nSEMANTIC segmentation of 3D LiDAR point clouds  \nis fundamental to achieving accurate spatial perception and comprehensive environmental understanding, facilitating applications in smart cities [1]–[3], 3D map construction [4], autonomous driving [5], [6], etc. With the rapid development of supervised 3D segmentation methods [5]–[10], accurate segmentation results have been achieved. However, these methods require massive annotated data, which is costly and laborious to obtain, especially in complex urban driving scenarios. To alleviate this limitation, unsupervised domain adaptation [11](UDA) offers a promising solution, which transfers the knowledge learned from easily obtained synthetic data with rich annotations to real-world unlabeled data.  \nManuscript received 18 November 2025; revised 28 February 2026; accepted 1 May 2026 . This work was supported in part by the Natural Science Foundation of Henan Province under Grant 262300422567, in part by the National Natural Science Foundation of China under Grant 62502243, in part by the Fundamental Research Funds for the Central Universities under Grant N25XQD053 . (Corresponding author: Ming Cheng.)  \nZ. Yuan and X. Huang are with School of Artificial Intelligence, Nanyang Normal University, Nanyang 473061, China.  \nS. Yu is with the School of Computer Science and Engineering, Northeastern University, Shenyang 110819, China.  \nM. Cheng, W. Li, D. Liu and C. Wang are with Fujian Key Laboratory of Sensing and Computing for Smart Cities, School of Informatics, Xiamen University, Xiamen 361005, China (e-mail: [chm99@xmu.edu.cn](chm99@xmu.edu.cn)).  \nSelf-training (ST), which iteratively refines the model using pseudo labels generated from its own predictions, has become a dominant paradigm for 3D UDA segmentation [12]–[16] . Despite its promising performance, its success heavily depends on a well-initialized warm-up (pre-trained) model to generate highquality pseudo labels. Currently, the warm-up model adopted by most existing ST-based methods [13],[15],[16] is trained on the source domain in a fully supervised manner. However, the significant domain gap betwee","cbCaibXNk7F9GTA8","https://ap.wps.com/l/cbCaibXNk7F9GTA8","pdf",4819388,1,12,"English","en",105,"# Introduction\n## Motivation: 3D LiDAR semantic segmentation without target labels\n## Self-training and the need for a stable warm-up model\n## Transferability vs. discriminability and domain-similar representations\n## Voxel-based representations and voxel size as a design factor","[{\"question\":\"What is the main problem CVKD-UDA targets in 3D UDA segmentation?\",\"answer\":\"CVKD-UDA targets the annotation cost for new-domain 3D data by enabling unsupervised domain adaptive segmentation, where target data has no labels. It focuses on improving the reliability of pseudo labels used during self-training.\"},{\"question\":\"Why do existing self-training warm-up methods often fail under domain shift?\",\"answer\":\"They often depend on source supervision or output-level adversarial alignment to obtain the warm-up model. Due to large domain gaps, this can produce noisy pseudo labels and lead to limited generalization and training instability.\"},{\"question\":\"How does CVKD-UDA construct domain-similar representations and balance transferability and discriminability?\",\"answer\":\"It revisits voxel size to generate two complementary views and employs cross-view knowledge distillation to improve generalization and target perception. To balance transferability and discriminability, it uses a lightweight Decouple-Adapter and an auxiliary imitation classifier to decouple the cross-view knowledge transfer.\"}]",1784206410,30,{"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},"cvkd-uda-cross-view-knowledge-distillation-for-3d-unsupervised-domain-adaptive-segmentation","",{"@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/cvkd-uda-cross-view-knowledge-distillation-for-3d-unsupervised-domain-adaptive-segmentation/85813/",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 is the main problem CVKD-UDA targets in 3D UDA segmentation?","Question",{"text":75,"@type":76},"CVKD-UDA targets the annotation cost for new-domain 3D data by enabling unsupervised domain adaptive segmentation, where target data has no labels. It focuses on improving the reliability of pseudo labels used during self-training.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"Why do existing self-training warm-up methods often fail under domain shift?",{"text":80,"@type":76},"They often depend on source supervision or output-level adversarial alignment to obtain the warm-up model. Due to large domain gaps, this can produce noisy pseudo labels and lead to limited generalization and training instability.",{"name":82,"@type":73,"acceptedAnswer":83},"How does CVKD-UDA construct domain-similar representations and balance transferability and discriminability?",{"text":84,"@type":76},"It revisits voxel size to generate two complementary views and employs cross-view knowledge distillation to improve generalization and target perception. To balance transferability and discriminability, it uses a lightweight Decouple-Adapter and an auxiliary imitation classifier to decouple the cross-view knowledge transfer.","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},[93,97,101,105,110,115,120,122,127,130,134],{"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":28,"slug":121},"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":106,"slug":137},19,"General","general"]