[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-291290-105":59,"doc-detail-291290-en":131},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":124,"head_meta":126,"extra_data":128,"updated_unix":130},105,"en","distribution-aligned-diffusion-for-human-mesh-recovery-iccv-2023-paper","Distribution-Aligned Diffusion for Human Mesh Recovery - ICCV 2023 paper","","Recovering a 3D human mesh from a single RGB image is impeded by depth ambiguity and self-occlusion, which introduces substantial uncertainty. The paper studies diffusion models as a progressive denoising mechanism and presents Human Mesh Diffusion (HMDiff) to formulate mesh recovery as a reverse diffusion process. It further introduces a Distribution Alignment Technique (DAT) that injects input-specific distribution information as prior knowledge. The approach reaches state-of-the-art results on three widely used datasets.",{"@graph":69,"@context":123},[70,84,106],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/distribution-aligned-diffusion-for-human-mesh-recovery-iccv-2023-paper/291290/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/distribution-aligned-diffusion-for-human-mesh-recovery-iccv-2023-paper/291290.png","ImageObject",300,407,{"name":92,"@type":93},"Jasmine","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-27","2026-09-17",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",11,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"Why is monocular 3D human mesh recovery difficult from a single RGB image?","Question",{"text":113,"@type":114},"Depth ambiguity and self-occlusion make the predicted 3D mesh vertices highly uncertain.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"What is Human Mesh Diffusion (HMDiff)?",{"text":118,"@type":114},"HMDiff frames mesh recovery as a reverse diffusion process that progressively denoises noisy and uncertain inputs into a high-quality mesh distribution.",{"name":120,"@type":111,"acceptedAnswer":121},"What does the Distribution Alignment Technique (DAT) do?",{"text":122,"@type":114},"DAT injects input-specific distribution information into the diffusion process, providing useful prior knowledge to simplify mesh recovery.","https://schema.org",{"og:url":83,"og:type":125,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":127,"canonical":83},"index,follow",{"doc_id":129,"site_id":62},291290,1789644302,{"code":4,"msg":5,"data":132},{"doc_id":129,"user_id":133,"nickname":92,"user_avatar":134,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":105,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":140,"language":141,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":142,"faqs":143,"seo_title":144,"seo_description":67,"update_tm":130,"read_time":41,"preview_page_count":4},2336478487870,"https://ap-avatar.wpscdn.com/davatar_085a072bc5b1113ac321206ff7593b45","This ICCV 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.  \nDistribution-Aligned Diffusion for Human Mesh Recovery  \nLin Geng Foo 1 Jia Gong 1 Hossein Rahmani2 Jun Liu 1†  \n1 Singapore University of Technology and Design 2Lancaster University  \n{lingeng   foo,jia [gong](gong}@mymail.sutd.edu.sg)[}](gong}@mymail.sutd.edu.sg)[@mymail.sutd.edu.sg](gong}@mymail.sutd.edu.sg) , [h.rahmani@lancaster.ac.uk](h.rahmani@lancaster.ac.uk) , jun   [liu@sutd.edu.sg](liu@sutd.edu.sg)  \nAbstract  \nRecovering a 3D human mesh from a single RGB image is a challenging task due to depth ambiguity and selfocclusion, resulting in a high degree of uncertainty. Meanwhile, diffusion models have recently seen much success in generating high-quality outputs by progressively denoising noisy inputs. Inspired by their capability, we explore a diffusion-based approach for human mesh recovery, and propose a Human Mesh Diffusion (HMDiff) framework which frames mesh recovery as a reverse diffusion process. We also propose a Distribution Alignment Technique (DAT) that injects input-specific distribution information into the diffusion process, and provides useful prior knowledge to simplify the mesh recovery task. Our method achieves state-of-the-art performance on three widely used datasets. Project page: [https://gongjia0208.github.io/HMDiff/](https://gongjia0208.github.io/HMDiff/).  \n1. Introduction  \nMonocular 3D human mesh recovery, where the 3D mesh vertex locations of a human are predicted from a single RGB image, is an important task with applications across virtual reality [17], sports motion analysis [1], and healthcare [54] . The field has received a lot of attention in recent years [6, 7, 32, 33, 8, 22], which has led to significant progress, but monocular 3D human mesh estimation still remains very challenging. Human body shapes are not only complex and contain many fine details, but also inherently exhibit depth ambiguity (when recovering 3D information from single 2D images) and self-occlusion (where body parts can be occluded by other body parts) [25, 6, 53] . In particular, the depth ambiguity and self-occlusion in this task often bring much uncertainty to the recovery of 3D mesh vertices, and places a huge burden on the model to handle this inherent uncertainty [30, 39, 8, 22] .  \nAt the same time, denoising diffusion probabilistic models (diffusion models) [16, 49] have recently seen much success in generative tasks, such as image [36], video [48] and  \n† Corresponding author  \ntext [31] generation, where they have been capable of producing highly realistic and good-quality samples. Specifically, diffusion models [16, 49] progressively “denoise” a noisy input – which is uncertain – into a high-quality output from the desired data distribution (e.g., natural images), through estimating the gradients ofthe data distribution [52](also known as the score function) . This progressive denoising helps break down the large gap between distributions (i.e., from a highly uncertain and noisy distribution toa desired target distribution), into smaller intermediate steps [52], which assists the model in converging towards generating the target data distribution smoothly. This gives diffusion models a strong ability to recover high-quality outputs from uncertain and noisy input data.  \nFor the monocular 3D mesh recovery task, we also seek to recover a high-quality mesh prediction from uncertain and noisy input data, and so we leverage diffusion models to effectively tackle this task. To this end, we propose a novel diffusion-based framework for monocular 3D human mesh recovery, called Human Mesh Diffusion (HMDiff), where we frame the mesh recovery task as a reverse diffusion process which recovers a high-quality mesh by progressively denoising noisy and uncertain inputs.  \nIntuitively, ","cbCaifB1TwqHtD5K","https://ap.wps.com/l/cbCaifB1TwqHtD5K","pdf",2833089,12,"English","# Abstract\n# Introduction","[{\"question\":\"Why is monocular 3D human mesh recovery difficult from a single RGB image?\",\"answer\":\"Depth ambiguity and self-occlusion make the predicted 3D mesh vertices highly uncertain.\"},{\"question\":\"What is Human Mesh Diffusion (HMDiff)?\",\"answer\":\"HMDiff frames mesh recovery as a reverse diffusion process that progressively denoises noisy and uncertain inputs into a high-quality mesh distribution.\"},{\"question\":\"What does the Distribution Alignment Technique (DAT) do?\",\"answer\":\"DAT injects input-specific distribution information into the diffusion process, providing useful prior knowledge to simplify mesh recovery.\"}]","Distribution-Aligned Diffusion for Human Mesh Recovery - ICCV 2023 paper | PDF"]