[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-517706-105":3,"detail-sidebar-cat-0-en-105":81,"doc-detail-517706-en":131},{"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","lattice-democratize-high-fidelity-3d-generation-at-scale","LATTICE - Democratize High-Fidelity 3D Generation at Scale","","LATTICE is a new framework for high-fidelity 3D asset generation that addresses the quality–scalability gap between 3D and 2D generative models. It targets the core difficulty of predicting both spatial structure and detailed geometry from scratch under the heavy compute cost of existing 3D representations. The paper introduces VoxSet, a semi-structured latent encoding anchored to a coarse voxel grid, enabling position-aware generation and token-level test-time scaling. A two-stage pipeline generates a sparse voxelized geometry anchor and then refines geometry via a rectified flow transformer, supporting arbitrary resolution decoding and efficient training and inference.",{"@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/lattice-democratize-high-fidelity-3d-generation-at-scale/517706/",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/lattice-democratize-high-fidelity-3d-generation-at-scale/517706.png","ImageObject",300,407,{"name":42,"@type":43},"Aldword","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-11-15","2026-10-07",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",12,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"What problem does LATTICE address in 3D generation?","Question",{"text":63,"@type":64},"LATTICE targets the gap in quality and scalability between 3D and 2D generative models, which is driven by the difficulty of representing and generating 3D spatial structure and detailed geometry efficiently.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"What is VoxSet and why is it important?",{"text":68,"@type":64},"VoxSet is a semi-structured representation that compresses 3D assets into latent vectors anchored to a coarse voxel grid. It introduces explicit structure in the latent space to make generation position-aware and support token-level test-time scaling.",{"name":70,"@type":61,"acceptedAnswer":71},"How does LATTICE generate 3D assets in its two-stage pipeline?",{"text":72,"@type":64},"The method first generates a sparse voxelized geometry anchor, then produces detailed geometry using a rectified flow transformer, enabling arbitrary resolution decoding and flexible inference.","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},517706,1791432603,{"code":4,"msg":82,"data":83},"success",[84,88,92,96,101,106,111,115,120,123,127],{"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":116,"doc_module":4,"doc_module_name":25,"category_name":117,"show_sort_weight":118,"slug":119},9,"Religion & Spirituality",20,"religion-spirituality",{"id":118,"doc_module":4,"doc_module_name":25,"category_name":121,"show_sort_weight":118,"slug":122},"World Cup","world-cup",{"id":124,"doc_module":4,"doc_module_name":25,"category_name":125,"show_sort_weight":124,"slug":126},10,"Lifestyle","lifestyle",{"id":128,"doc_module":4,"doc_module_name":25,"category_name":129,"show_sort_weight":97,"slug":130},19,"General","general",{"code":4,"msg":82,"data":132},{"doc_id":79,"user_id":133,"nickname":42,"user_avatar":134,"doc_module":4,"category_id":112,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":140,"language":141,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":142,"faqs":143,"seo_title":144,"seo_description":12,"update_tm":145,"read_time":146,"preview_page_count":33,"preview_urls":147},2336478940917,"https://ap-avatar.wpscdn.com/davatar_6f874abed73319feea01a86fa6f0fab8","This CVPR paper is the Open Access version, provided by the Computer Vision Foundation.  \nExcept for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore.  \nLATTICE: Democratize High-Fidelity 3D Generation at Scale  \nZeqiang Lai 1 ,2⋆ , Yunfei Zhao2⋆ , Zibo Zhao2 , Haolin Liu2 Qingxiang Lin2 , Jingwei Huang2 , Chunchao Guo2‡ , Xiangyu Yue 1‡ 1MMLab, CUHK 2Tencent Hunyuan  \n[https://lattice3d.github.io](https://lattice3d.github.io)  \nFigure 1 . High quality 3D assets generated by LATTICE from a single image.  \nAbstract  \nWe present LATTICE, a new framework for high-fidelity 3D asset generation that bridges the quality and scalability gap between 3D and 2D generative models. While 2D image synthesis benefits from fixed spatial grids and wellestablished transformer architectures, 3D generation remains fundamentally more challenging due to the need to predict both spatial structure and detailed geometric surfaces from scratch. These challenges are exacerbated by the computational complexity of existing 3D representations and the lack of structured and scalable 3D asset encoding schemes. To address this, we propose VoxSet, a semi-structured representation that compresses 3D assets  \n⋆ Equal contribution.† Corresponding authors.  \ninto a compact set of latent vectors anchored to a coarse voxel grid, enabling efficient and position-aware generation. VoxSet retains the simplicity and compression advantages of prior VecSet methods while introducing explicit structure into the latent space, allowing positional embeddings to guide generation and enabling strong token-level test-time scaling. Built upon this representation, LATTICE adopts a two-stage pipeline: first generating a sparse voxelized geometry anchor, then producing detailed geometry using a recitified flow transformer. Our method is simple at its core, but supports arbitrary resolution decoding, low-cost training, and flexible inference schemes, achieving state-of-the-art performance on various aspects, and offering a significant step toward scalable, high-quality 3D asset creation.  \n1. Introduction  \nCreating high-quality 3D assets is central to modern content pipelines across visual effects, gaming, virtual reality, and industrial design. Yet, manual creation remains laborintensive and demands expert skills. Thus, automating 3Dasset generation has become a key challenge at the intersection of vision, graphics, and machine learning.  \nDespite the impressive progress demonstrated by recent advances in 3D generation [46, 55, 57], the question of how to represent 3D assets remains the “dark cloud”over scalable 3D generation — a fundamental problem that continues to hinder progress in fidelity, efficiency, and generalization. This challenge is deeply intertwined not only with classical 3D representations —such as meshes, point clouds, Signed Distance Functions (SDFs), radiance fields [28], and 3D Gaussian Splattings [14] — but also with VAE [6] representations adopted by latent diffusion models [35], a key paradigm underpinning the recent advances in 3D generation, as well as in image and video synthesis [16, 17] . Even with latent diffusion–based compression, the underlying complexity of 3D structure and its cubic growth in memory and computation remain major obstacles, underscoring the need for compact representations. Asa result, compression, reconstruction, and generation have become even more crucial in 3D learning, standing as longstanding themes throughout prior research.  \nThe pursuit of such an ideal 3D representation has thus led to what we call representation-centric research, primarily revolving around compression and reconstruction by two leading paradigms: Sparse Voxel and VecSet. Sparse Voxelbased methods [34, 46] aim for efficiency by restricting computation to the active voxels near the object surface. However, as reported in Trellis [46], even with the inherent sparsity of 3D data, t","cbCaibIaJlYmcoTQ","https://ap.wps.com/l/cbCaibIaJlYmcoTQ","pdf",17501394,11,"English","# Abstract\n# 1. Introduction","[{\"question\":\"What problem does LATTICE address in 3D generation?\",\"answer\":\"LATTICE targets the gap in quality and scalability between 3D and 2D generative models, which is driven by the difficulty of representing and generating 3D spatial structure and detailed geometry efficiently.\"},{\"question\":\"What is VoxSet and why is it important?\",\"answer\":\"VoxSet is a semi-structured representation that compresses 3D assets into latent vectors anchored to a coarse voxel grid. It introduces explicit structure in the latent space to make generation position-aware and support token-level test-time scaling.\"},{\"question\":\"How does LATTICE generate 3D assets in its two-stage pipeline?\",\"answer\":\"The method first generates a sparse voxelized geometry anchor, then produces detailed geometry using a rectified flow transformer, enabling arbitrary resolution decoding and flexible inference.\"}]","LATTICE - Democratize High-Fidelity 3D Generation at Scale | PDF",1791343903,28,[148,152,156,160],{"page_no":22,"url":149,"width":150,"height":151},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/301a4b05207427781246909b54f0925f.webp",983,1272,{"page_no":26,"url":153,"width":154,"height":155},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/1491fa22c41c92d3b878661a11a8b355.webp",982,1270,{"page_no":30,"url":157,"width":158,"height":159},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/c2187df0b8e489106783e22f3a4bd6c4.webp",981,1269,{"page_no":33,"url":161,"width":162,"height":163},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/7a510f1f55397902f32614179ae42bb8.webp",980,1268]