[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84420-en":3,"doc-seo-84420-105":29,"detail-sidebar-cat-0-en-105":83},{"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},84420,1099513958607,"Jiven","https://ap-avatar.wpscdn.com/avatar/100002390cf8733938c?x-image-process=image/resize,m_fixed,w_180,h_180&k=1778829742770036399",8,"Research & Report","SparseLGS: Sparse View Language Embedded Gaussian Splatting","SparseLGS addresses 3D scene understanding with pose-free, sparse-view input images by enabling open-vocabulary semantic field reconstruction without the dense multiview requirement of prior Gaussian Splatting plus language-embedding approaches. The method uses a learning-based dense stereo model for pose estimation under sparse, unposed views, then applies a three-step region matching strategy to resolve multi-view semantic inconsistency. Low-dimensional feature bijections avoid excessive CLIP learning and storage costs, while a reconstruction loss improves Gaussian positions and shapes. Experiments show comparable quality with only 3–4 views versus dense-input SOTA, plus about 5× faster computation.","SparseLGS: Sparse View Language Embedded  \nGaussian Splatting  \nJun Hu, Zhang Chen†, Zhong Li, Yi Xu, and Juyong Zhang†  \narXiv :2412 .02245v3 [ cs .CV] 12 Jul 2026  \nAbstract—Recently, several studies have combined Gaussian Splatting to obtain scene representations with language embeddings for open-vocabulary 3D scene understanding. While these methods perform well, they essentially require very dense multiview inputs, limiting their applicability in real-world scenarios. In this work, we propose SparseLGS to address the challenge of 3D scene understanding with pose-free and sparse view input images. Our method leverages a learning-based dense stereo model to handle pose-free and sparse inputs, and a three-step region matching approach to address the multi-view semantic inconsistency problem, which is especially important for sparse inputs. Different from directly learning high-dimensional CLIP features, we extract low-dimensional information and build bijections to avoid excessive learning and storage costs. We introduce areconstruction loss during semantic training to improve Gaussian positions and shapes. To the best of our knowledge, we are the first to address the 3D semantic field problem with sparse posefree inputs. Experimental results show that SparseLGS achieves comparable quality when reconstructing semantic fields with fewer inputs (3-4 views) compared to previous SOTA methods with dense input. Besides, when using the same sparse input, SparseLGS leads significantly in quality and heavily improves the computation speed (5× speedup). Project page: [https:](https:)//[ustc3dv.github.io/SparseLGS](ustc3dv.github.io/SparseLGS)  \nIndex Terms—Sparse, Semantic Field, Reconstruction, Semantic Alignment.  \nI. INTRODUCTION  \n3 D language field modeling is an important research prob  \nlem in computer vision, offering extensive application prospects in fields such as autonomous driving, robotic manipulation [1], [2], and VR/AR. To obtain and enhance the quality of a 3D language field, high-precision 3D reconstruction is often necessary. Following the advent of NeRF [3], numerous works focusing on 3D semantic fields have emerged [4], [5] . Initially, these semantic fields were more akin to rendering mask segmentation for each view, heavily reliant on semantic annotations of the data and lacking the capability for open language queries. To address these shortcomings, LERF [6] distills the required features from the language-image model CLIP and integrates them into NeRF. However, the bottlenecks of slow training and volumetric rendering in NeRF, as well as the quality limitations due to CLIP features being imagealigned rather than region or pixel-aligned, remain unresolved.  \nThe recently proposed explicit 3D reconstruction method, 3D Gaussian Splatting [8], offers fast training and real-time  \nJ. Hu, and J. Zhang are with the School of Mathematical Science, University of Science and Technology of China. Z. Chen is with the Meta, Z. Li is with the Apple Inc. and Y. Xu is with the Alpha Labs at Goertek.  \n†Corresponding author. Email: [lansburyc@gmail.com](lansburyc@gmail.com) , [juyong@ustc.edu.cn](juyong@ustc.edu.cn).  \nFig. 1. We present the semantic renderings from sparse, pose-free inputs using our method and LangSplat [7] . Our method outperforms LangSplat in both multi-view consistency and rendering quality, producing more accurate and visually coherent results.  \nrendering, effectively addressing the speed issues associated with previous NeRF-based methods. Additionally, by using SAM [9] for mask segmentation and integrating semantic models such as CLIP [10],[11] or DINOv2 [12], it tackles the quality issues caused by unclear semantic boundaries. These methods [7], [13] optimize the semantics of Gaussians by downscaling the original CLIP features through techniques such as autoencoding and quantization with MLP. However, after obtaining the downscaled semantic features, they need to reconstruct the raw CLIP features. This","cbCaiuWOGN8jwoQX","https://ap.wps.com/l/cbCaiuWOGN8jwoQX","pdf",7829857,1,14,"English","en",105,"# Introduction\n## Prior semantic field methods and limitations\n## Motivation for sparse, pose-free inputs\n# Method overview\n## Pose estimation and initial point cloud generation\n## Semantic extraction and multi-view alignment","[{\"question\":\"Why is multi-view semantic inconsistency more harmful with sparse inputs, and how is it addressed?\",\"answer\":\"With only 3–4 views, incorrect semantic matches can distort the reconstruction. SparseLGS introduces a three-step region matching and semantic alignment strategy (including pixel matching and region fusion) to improve cross-view consistency and rendering quality.\"}]",1784195523,35,{"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":78,"head_meta":80,"extra_data":82,"updated_unix":27},"sparselgs-sparse-view-language-embedded-gaussian-splatting","",{"@graph":35,"@context":77},[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/sparselgs-sparse-view-language-embedded-gaussian-splatting/84420/",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],{"name":72,"@type":73,"acceptedAnswer":74},"Why is multi-view semantic inconsistency more harmful with sparse inputs, and how is it addressed?","Question",{"text":75,"@type":76},"With only 3–4 views, incorrect semantic matches can distort the reconstruction. SparseLGS introduces a three-step region matching and semantic alignment strategy (including pixel matching and region fusion) to improve cross-view consistency and rendering quality.","Answer","https://schema.org",{"og:url":51,"og:type":79,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":81,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":84},[85,89,93,97,102,107,112,115,120,123,127],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":86,"show_sort_weight":87,"slug":88},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":90,"show_sort_weight":91,"slug":92},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Exam",70,"exam",{"id":98,"doc_module":4,"doc_module_name":45,"category_name":99,"show_sort_weight":100,"slug":101},5,"Comic",60,"comic",{"id":103,"doc_module":4,"doc_module_name":45,"category_name":104,"show_sort_weight":105,"slug":106},6,"Technology",50,"technology",{"id":108,"doc_module":4,"doc_module_name":45,"category_name":109,"show_sort_weight":110,"slug":111},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":113,"slug":114},30,"research-report",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},9,"Religion & Spirituality",20,"religion-spirituality",{"id":118,"doc_module":4,"doc_module_name":45,"category_name":121,"show_sort_weight":118,"slug":122},"World Cup","world-cup",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":124,"slug":126},10,"Lifestyle","lifestyle",{"id":128,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":98,"slug":130},19,"General","general"]