[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85717-en":3,"doc-seo-85717-105":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},85717,4398048950312,"Violet","https://ap-avatar.wpscdn.com/avatar/400002538284de19e3c?_k=1778320343897328908",8,"Research & Report","Adversarially Guided Diffusion for LiDAR Range Image Synthesis","LiDAR semantic segmentation is central to autonomous driving perception, yet adversarial inputs can distort segmentation maps and propagate faults to planning and safety-critical decisions. While adversarial examples are well studied for image and 3D point-cloud tasks, unrestricted attacks on 2D LiDAR range-image projections remain largely unexplored. The method introduces the first diffusion-based unrestricted adversarial attack for 2D range-image segmentation, using segmentation-loss gradients as guidance during sampling to preserve manifold realism while inducing structured errors. Experiments on SemanticKITTI with RangeNet++ and CENet show adjustable degradation, strong transferability, and a controllable effectiveness–realism trade-off versus norm-bounded FGSM and SegPGD baselines.","arXiv :2607 .09787v 1 [ cs .CV] 8 Jul 2026  \nAdversarially Guided Diffusion for LiDAR Range  \nImage Synthesis  \nStavros Bouras 1[0009−0009−2948−1833], Antonios Makris 1[0000−0003−0514−4292], Alexandros Gkillas2[0000−0001−5339−2018], Aris S. Lalos2[0000−0003−0511−9302],  \nand Konstantinos Tserpes 1[0000−0001−5183−1443]  \n1 School of Electrical and Computer Engineering, National Technical University of Athens, Greece {stavros_bouras,antoniosmakris,[tserpes}@mail.ntua.gr](tserpes}@mail.ntua.gr)  \n2 Industrial Systems Institute, Athena Research Center, Patras Science Park, Greece {gillas,[lalos}@athenarc.gr](lalos}@athenarc.gr)  \nAbstract. LiDAR semantic segmentation is a key perception task in autonomous driving, where false predictions can affect downstream planning and safety-critical decision-making. Although adversarial attacks, and specifically adversarial examples, have been widely studied for image classification and 3D point cloud segmentation, unrestricted adversarial examples remain largely unexplored in the space of 2D range images, which are projections of 3D point clouds. The proposed method is, to the best of our knowledge, the first diffusion-based unrestricted adversarial attack against 2D range-image segmentation, using adversarial guidance from a segmentation loss. By applying guidance directly during sampling, the method produces unrestricted adversarial examples that remain close to the learned LiDAR data manifold while inducing structured segmentation errors. Experiments on the SemanticKITTI dataset using RangeNet++ and CENet segmentation networks demonstrate that the attack provides adjustable degradation across guidance strengthsand transfers across segmentation architectures. Compared with normbounded FGSM and SegPGD baselines, the proposed attack offers a distinct effectiveness–realism trade-off, achieving controllable white-box and transfer degradation while maintaining competitive distributionaland visual realism.  \nKeywords: Adversarial diffusion sampling · LiDAR Diffusion Model · Unrestricted adversarial examples.  \n1 Introduction  \nDeep Learning (DL) models are increasingly deployed in real-world, safetycritical perception tasks such as semantic segmentation [14], where reliable scene understanding is required under strict operational constraints [37] . In the automotive domain specifically, segmentation networks must process live data sourced from Light Detection and Ranging (LiDAR) sensors to understand the surrounding scene. LiDAR sensors provide accurate depth information in the form of 3D  \n2 S. Bouras et al.  \npoint clouds, making them particularly valuable for autonomous driving perception. Semantic segmentation can then be applied to these LiDAR points, orto LiDAR-derived representations, to assign semantic labels to relevant scene elements and support accurate, robust, and real-time scene understanding in complex driving environments [10] .  \nLiDAR semantic segmentation can be performed directly on the 3D point cloud [19] or on their 2D range image projection space [22] . This projection is widely adopted in the literature, offering a computationally efficient alternative to direct 3D point-cloud processing and enabling more scalable operations in 2D space for embedded automotive systems. However, given their reliance on complex DL architectures, LiDAR semantic segmentation networks inherit the susceptibility of conventional deep neural networks to adversarial examples [30, 12] . These adversarial inputs are deliberately crafted perturbations designed to mislead the model while remaining imperceptible or structurally inconspicuous. For LiDAR semantic segmentation, such attacks may take the form of point perturbation, point injection, point removal, or physically realizable object-based manipulations [38] . These attacks can corrupt the predicted segmentation map by producing large, spatially coherent misclassification regions. As a result, erroneous semantic outputs may propagate to downstre","cbCainh3rbaW96PD","https://ap.wps.com/l/cbCainh3rbaW96PD","pdf",2891496,1,16,"English","en",105,"# Abstract\n# Introduction\n## Motivation: adversarial vulnerability in LiDAR segmentation\n## Diffusion models for manifold-preserving adversarial generation\n## Paper contributions and organization","[{\"question\":\"What problem does the paper address in LiDAR semantic segmentation?\",\"answer\":\"It targets the susceptibility of LiDAR segmentation to adversarial inputs, where false predictions can mislead downstream autonomous-driving planning and safety-critical decisions.\"},{\"question\":\"How does the proposed method generate adversarial range images?\",\"answer\":\"It uses a latent diffusion model guided during sampling by gradients from a dense per-pixel segmentation loss, steering generation to remain close to the learned LiDAR data manifold while producing structured segmentation errors.\"},{\"question\":\"How effective is the attack and what trade-off does it achieve?\",\"answer\":\"Experiments on SemanticKITTI with RangeNet++ and CENet show adjustable degradation across guidance strengths and transfer across segmentation architectures, providing a controllable effectiveness–realism trade-off compared with norm-bounded FGSM and SegPGD.\"}]",1784205763,40,{"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},"adversarially-guided-diffusion-for-lidar-range-image-synthesis","",{"@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/adversarially-guided-diffusion-for-lidar-range-image-synthesis/85717/",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 the paper address in LiDAR semantic segmentation?","Question",{"text":74,"@type":75},"It targets the susceptibility of LiDAR segmentation to adversarial inputs, where false predictions can mislead downstream autonomous-driving planning and safety-critical decisions.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How does the proposed method generate adversarial range images?",{"text":79,"@type":75},"It uses a latent diffusion model guided during sampling by gradients from a dense per-pixel segmentation loss, steering generation to remain close to the learned LiDAR data manifold while producing structured segmentation errors.",{"name":81,"@type":72,"acceptedAnswer":82},"How effective is the attack and what trade-off does it achieve?",{"text":83,"@type":75},"Experiments on SemanticKITTI with RangeNet++ and CENet show adjustable degradation across guidance strengths and transfer across segmentation architectures, providing a controllable effectiveness–realism trade-off compared with norm-bounded FGSM and SegPGD.","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},[92,96,100,104,109,114,118,121,126,129,133],{"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":28,"slug":117},7,"Healthcare","healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":119,"slug":120},30,"research-report",{"id":122,"doc_module":4,"doc_module_name":45,"category_name":123,"show_sort_weight":124,"slug":125},9,"Religion & Spirituality",20,"religion-spirituality",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":127,"show_sort_weight":124,"slug":128},"World Cup","world-cup",{"id":130,"doc_module":4,"doc_module_name":45,"category_name":131,"show_sort_weight":130,"slug":132},10,"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":45,"category_name":135,"show_sort_weight":105,"slug":136},19,"General","general"]