[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82503-en":3,"doc-seo-82503-105":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},82503,687197100911,"Himbo","https://ap-avatar.wpscdn.com/avatar/a000239b6f1da00475?x-image-process=image/resize,m_fixed,w_180,h_180&k=1782698725881665579",8,"Research & Report","Vitality-Aware Compression for Efficient Image-to-Shape Diffusion Transformers","This paper presents a geometry-aware compression framework for image-to-shape Diffusion Transformers (DiTs) that reduces model size while preserving geometric fidelity. It targets the difficulty of transferring diffusion compression methods from image generation to 3D shape synthesis, where structural collapse and topology distortion can occur. Using per-layer ablation with Earth Mover’s Distance, the approach identifies non-uniform layer importance and builds a vitality-guided pipeline with structured pruning, adaptive quantization, and targeted fine-tuning. Results reach up to 66% model-size reduction across state-of-the-art models with comparable synthesis quality.","arXiv :2607 .00382v 1 [ cs .CV] 1 Jul 2026  \nVitality-Aware Compression for Efficient Image-to-Shape Diffusion Transformers  \nJaeah Lee 1 ⋆, Hyunjin Kim 1 ⋆, Jaewoong Cho 1, and Gihyun Kwon2†  \n1 KRAFTON AI, Republic of Korea  \n2 Amazon, Australia  \nFig. 1: In this paper, we introduce a vitality-guided Diffusion Transformer (DiT) compression pipeline for image-to-3D shape generation. Our approach reduces model size while preserving synthesis quality.  \nAbstract. We propose the first compression approach for image-toshape Diffusion Transformers (DiTs) that substantially reduces model size while preserving geometric fidelity. Despite remarkable progress in 3D shape generation, large DiT-based models remain computationally prohibitive in resource-constrained settings. Furthermore, it is difficult to directly transfer existing diffusion model compression strategies developed for different domains to 3D generation, and prior 3D efficiency approaches focus primarily on inference speed rather than backbone compression. To address this limitation, we build a geometry-aware compression framework tailored to image-to-shape DiTs. Guided by the observation that 3DDiT layers exhibit non-uniform importance for geometry synthesis, we introduce a vitality-guided framework integrating structured pruning, adaptive quantization, and targeted fine-tuning. Our method achieves up to 66% model-size reduction across state-of-the-art image-to-3D models while maintaining synthesis fidelity comparable to full-sized counterparts.  \nThis highlights the potential of our framework as a plug-and-play solution for efficient 3D shape generation across diverse models.  \nKeywords: Model Compression · 3D Geometry Generation · Image-to- 3D Synthesis · Diffusion Transformers · Efficient Generative Models  \n⋆ Authors contributed equally to this work.† Worked done at KRAFTON AI.  \n2 J. Lee et al.  \nFig. 2: Domain Gap of DiT Compression on Step1X-3D. Diffusion model compression strategies designed for image generation [ 12 , 13] do not directly transfer to image-to-shape synthesis, as preserving visual quality does not guarantee geometric fidelity. This highlights the need for an alternative approach to shape generation.  \n1 Introduction  \nThe rapid expansion of 3D content across AR/VR, gaming, and embodied AI has intensified the demand for scalable and automated 3D shape generation. Recent image-to-3D synthesis has progressed from GAN-based priors [14, 52 , 54] and Large Reconstruction Models [16, 41 , 55] to 3D-native Diffusion Transformer (DiT) models [46, 56] and flow-matching frameworks [28, 47 , 57], achieving geometrically consistent meshes from a single image. However, the DiT backbones in these pipelines often exceed 2.5 GB in parameter size alone, limiting their use in real-time and resource-constrained environments.  \nWhile diffusion model compression has been actively studied for image [6, 12 , 13 , 26 , 51] and video [34, 50] generation, where redundancy mainly arises from spatial or temporal correlations, these strategies do not address the structural demands of 3D shape synthesis. As shown in Fig. 2, directly applying diffusion compression methods originally designed for image generation to shape generation results in severe geometric degradation, including structural collapse, distorted topology, and loss of fine details. This discrepancy arises from the differences between 2D and 3D generation, as 3D models must maintain globally consistent geometry across viewpoints, and even small perturbations in the denoising process can propagate into structural artifacts [15] . Meanwhile, existing 3D efficiency approaches [24, 41] mainly focus on inference acceleration rather than backbone compression, providing limited benefit for memory-constrained applications.  \nRecent studies on DiT-based text-to-image [1] and text-to-video [23] synthesis have revealed that only a subset of layers significantly influence output quality. However, these studies analyze per-la","cbCait4AGnTrfnA0","https://ap.wps.com/l/cbCait4AGnTrfnA0","pdf",26277925,1,41,"English","en",105,"# Abstract\n# Introduction\n## Problem and motivation\n## Motivation from domain gap and 2D vs 3D differences\n## Layer importance analysis\n# Vitality-Aware Compression for Image-to-Shape DiT","[{\"question\":\"Why can image-generation diffusion compression strategies fail for image-to-shape DiTs?\",\"answer\":\"Because preserving 2D visual quality does not guarantee geometric fidelity in 3D. Directly applying those methods can cause structural collapse, distorted topology, and loss of fine details.\"},{\"question\":\"How does the paper measure which DiT layers are important for 3D synthesis?\",\"answer\":\"It performs per-layer ablation and uses Earth Mover’s Distance (EMD) on the generated point clouds to compute each layer’s contribution, revealing non-uniform vitality across layers.\"},{\"question\":\"What compression pipeline does the vitality-guided framework use?\",\"answer\":\"It prunes low-vitality layers with separate thresholds for double- and single-block layers, applies adaptive quantization with higher precision for geometry-critical layers, and performs targeted fine-tuning by updating only the least-vital retained layers.\"}]",1784180975,103,{"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},"vitality-aware-compression-for-efficient-image-to-shape-diffusion-transformers","",{"@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/vitality-aware-compression-for-efficient-image-to-shape-diffusion-transformers/82503/",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},"Why can image-generation diffusion compression strategies fail for image-to-shape DiTs?","Question",{"text":75,"@type":76},"Because preserving 2D visual quality does not guarantee geometric fidelity in 3D. Directly applying those methods can cause structural collapse, distorted topology, and loss of fine details.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the paper measure which DiT layers are important for 3D synthesis?",{"text":80,"@type":76},"It performs per-layer ablation and uses Earth Mover’s Distance (EMD) on the generated point clouds to compute each layer’s contribution, revealing non-uniform vitality across layers.",{"name":82,"@type":73,"acceptedAnswer":83},"What compression pipeline does the vitality-guided framework use?",{"text":84,"@type":76},"It prunes low-vitality layers with separate thresholds for double- and single-block layers, applies adaptive quantization with higher precision for geometry-critical layers, and performs targeted fine-tuning by updating only the least-vital retained layers.","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},[93,97,101,105,110,115,120,123,128,131,135],{"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":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]