[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82974-en":3,"doc-seo-82974-105":29,"detail-sidebar-cat-0-en-105":89},{"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":11},82974,687197207639,"Asher","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",8,"Research & Report","Clustered Codebook Quantization for 2D Gaussian-based Image Compression","Gaussian-based image representations model image content with compact 2D Gaussian primitives while preserving visual fidelity, but storing many floating-point parameters per primitive harms rate–distortion efficiency at higher quality targets. CGVQ proposes clustered codebook quantization to partition Gaussian parameters into homogeneous groups before quantization. K-Means clustering drives per-cluster, localized codebooks and independent encoding/decoding. Experiments show a 20% reduction in bits per pixel versus the GI baseline while keeping comparable visual quality.","Clustered Codebook Quantization for 2D Gaussian-based Image  \nCompression  \nRunze Cheng∗ University College London London, UK  \nYicheng Zhan  \nUniversity College London London, UK  \nJosef Spjut  \nNVIDIA Durham, USA  \nKaan Akşit  \nUniversity College London London, UK  \narXiv :2607 .05667v 1 [ cs .CV] 6 Jul 2026  \nFigure 1: Overview of our Cluster-Guided Vector Quantization (CGVQ) pipeline for 2D Gaussian-based Image Compression.  \nGiven a ground truth image 􀀞 , a set of 2D Gaussian primitives G = {􀀜 􀀸} is first fitted to the image. K-Means clustering partitions G into 􀀣 groups G􀀣 based on appearance and anisotropy similarity. Per-cluster codebooks are then trained and used to encode each group independently. At decoding, each cluster is reconstructed via multiple codebooks lookup and all clusters are composed for rendering, producing the reconstructed output 􀀞′ with Bit-Per-Pixel (bpp) at 1.92 and the baseline GaussianImage (GI) [Zhang et al. 2025a] output atbpp=2.40 under similar reconstruction fidelity (PSNR = 31.0􀀭􀀳􀀗) . Source image © Animal Faces.  \nAbstract  \nGaussian-based image representations effectively model image content using compact parametric primitives while preserving high visual fidelity [Zhang et al. 2025a], yet storing a large number of floating-point parameters per primitive degrades rate-distortion efficiency at higher fidelity targets. To improve the rate-distortion performance in Gaussian representation, we present our CGVQ, a Gaussian primitive based image compression method. Our key idea is to partition Gaussian parameters further into homogeneous groups prior to quantization, enabling higher compression efficiency and accurate parameter reconstruction. In practice, our extensive experiments show that CGVQ decreases the bpp by 20%↓ with respect to our baseline GI [Zhang et al. 2025a], while maintaining on-par visual quality.  \nCCS Concepts  \n• Computing methodologies → Image compression; Rendering.  \nKeywords  \nImage Compression, 2D Gaussian Splatting, K-Means, Vector Quantizer  \nACM Reference Format:  \nRunze Cheng, Yicheng Zhan, Josef Spjut, and Kaan Akşit. 2026. Clustered Codebook Quantization for 2D Gaussian-based Image Compression. In  \n∗ Primary contributor to this work  \nThis work is licensed under a Creative Commons Attribution 4 .0 International License. SIGGRAPH Posters ’26, Los Angeles, CA, USA  \n© 2026 Copyright held by the owner/author(s) .  \nACM ISBN 979-8-4007-2548-7/2026/07  \n[https://doi.org/10.1145/3799825.3818700](https://doi.org/10.1145/3799825.3818700)  \nSpecial Interest Group on Computer Graphics and Interactive Techniques Conference Posters (SIGGRAPH Posters ’26), July 19–23, 2026, Los Angeles, CA, USA. ACM, New York, NY, USA, 3 pages. [https://doi.org/10.1145/3799825](https://doi.org/10.1145/3799825) . 3818700  \n1 Introduction  \nImage compression is essential for real-time rendering & VR/AR [Walton et al. 2021] and high-density media storage [Wang et al. 2010], demanding realistic visual fidelity alongside computational efficiency. While traditional codecs like JPEG and PNG provide robust baselines, deep learning architectures [Walton et al. 2021] have been proposed to further improve the boundaries of rate-distortion performance. More Recently, have been posed as a promising alternative for compact image representation [Zhang et al. 2025a,b]. By modeling local color and structural details with anisotropic Gaussian primitives, Gaussian-based image representation enables flexible, resolution-independent rendering and ultra-fast, GPU-friendly inference [Huang et al. 2024; Mescheder et al. 2026; Zhang et al. 2025a] . However, this flexibility incurs a storage cost. Each 2D Gaussian primitive consists of multiple unconstrained floatingpoint parameters (e.g., positions, covariance, and color). Accurately capturing high-frequency textures requires dense splat allocations, bloating the parameter count and degrading compression efficiency compared to SOTA VAE-based codecs [van den Oord et","cbCaibWEqOa4fjBm","https://ap.wps.com/l/cbCaibWEqOa4fjBm","pdf",2094976,4,1,3,"English","en",105,"# Abstract\n# Introduction\n# Method","[{\"question\":\"What problem does CGVQ address in Gaussian-based image compression?\",\"answer\":\"CGVQ targets the rate–distortion inefficiency caused by quantizing many unconstrained floating-point parameters with a capacity-limited global codebook, which can introduce errors and artifacts.\"},{\"question\":\"How does CGVQ improve quantization efficiency?\",\"answer\":\"CGVQ uses K-Means to cluster 2D Gaussian primitives into homogeneous groups, then trains localized, per-cluster codebooks so quantization matches narrower parameter distributions.\"},{\"question\":\"What compression gains does CGVQ report compared with the GI baseline?\",\"answer\":\"The experiments report about a 20% lower bits-per-pixel (bpp) than GI while maintaining on-par visual quality, with similar reconstruction fidelity.\"}]",1784184409,{"code":4,"msg":30,"data":31},"ok",{"site_id":25,"language":24,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":84,"head_meta":86,"extra_data":88,"updated_unix":28},"clustered-codebook-quantization-for-2d-gaussian-based-image-compression","",{"@graph":35,"@context":83},[36,51,66],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,49],{"item":40,"name":41,"@type":42,"position":21},"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":22},"https://docshare.wps.com/document/research-report/",{"item":50,"name":13,"@type":42,"position":20},"https://docshare.wps.com/document/clustered-codebook-quantization-for-2d-gaussian-based-image-compression/82974/",{"url":50,"name":13,"@type":52,"author":53,"headline":13,"publisher":55,"fileFormat":58,"inLanguage":24,"description":14,"dateModified":59,"datePublished":60,"encodingFormat":58,"isAccessibleForFree":61,"interactionStatistic":62},"DigitalDocument",{"name":9,"@type":54},"Person",{"url":40,"name":56,"@type":57},"DocShare","Organization","application/pdf","2026-07-23","2026-07-16",true,{"@type":63,"interactionType":64,"userInteractionCount":20},"InteractionCounter",{"@type":65},"ViewAction",{"@type":67,"mainEntity":68},"FAQPage",[69,75,79],{"name":70,"@type":71,"acceptedAnswer":72},"What problem does CGVQ address in Gaussian-based image compression?","Question",{"text":73,"@type":74},"CGVQ targets the rate–distortion inefficiency caused by quantizing many unconstrained floating-point parameters with a capacity-limited global codebook, which can introduce errors and artifacts.","Answer",{"name":76,"@type":71,"acceptedAnswer":77},"How does CGVQ improve quantization efficiency?",{"text":78,"@type":74},"CGVQ uses K-Means to cluster 2D Gaussian primitives into homogeneous groups, then trains localized, per-cluster codebooks so quantization matches narrower parameter distributions.",{"name":80,"@type":71,"acceptedAnswer":81},"What compression gains does CGVQ report compared with the GI baseline?",{"text":82,"@type":74},"The experiments report about a 20% lower bits-per-pixel (bpp) than GI while maintaining on-par visual quality, with similar reconstruction fidelity.","https://schema.org",{"og:url":50,"og:type":85,"og:title":13,"og:site_name":56,"og:description":14},"article",{"robots":87,"canonical":50},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":90},[91,95,99,103,108,113,118,121,126,129,133],{"id":21,"doc_module":4,"doc_module_name":45,"category_name":92,"show_sort_weight":93,"slug":94},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":96,"show_sort_weight":97,"slug":98},"Literature",80,"literature",{"id":20,"doc_module":4,"doc_module_name":45,"category_name":100,"show_sort_weight":101,"slug":102},"Exam",70,"exam",{"id":104,"doc_module":4,"doc_module_name":45,"category_name":105,"show_sort_weight":106,"slug":107},5,"Comic",60,"comic",{"id":109,"doc_module":4,"doc_module_name":45,"category_name":110,"show_sort_weight":111,"slug":112},6,"Technology",50,"technology",{"id":114,"doc_module":4,"doc_module_name":45,"category_name":115,"show_sort_weight":116,"slug":117},7,"Healthcare",40,"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":104,"slug":136},19,"General","general"]