[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82939-en":3,"doc-seo-82939-105":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},82939,8796095461610,"Oliver","https://ap-avatar.wpscdn.com/davatar_276721f389ce27ea32af1340a28f341c",8,"Research & Report","WildSplat: Feedforward Gaussian Splatting from Unposed In-the-Wild Images","WildSplat presents a feedforward 3D Gaussian Splatting framework for in-the-wild images with varying appearances and unposed inputs, enabling photorealistic novel-view synthesis without exhaustive per-scene optimization. The method conditions reconstruction on a designated reference image and uses a dual-branch design to decouple geometry from appearance. A geometry branch predicts invariant 3D Gaussians, while an appearance encoder injects reference-conditioned features through cross-attention. Multi-reference training stabilizes learning under photometric inconsistencies.","WildSplat: Feedforward Gaussian Splatting from Unposed In-the-Wild Images  \nXiyu Zhang 1 ,2 ,∗ ,‡, Jingyu Zhuang2 ,∗ , Hongjia Zhai 1 , Zizheng Yan2 , Jinwei Chen2 , Guofeng Zhang 1 ,†, and Qingnan Fan2 ,†  \n1 State Key Lab of CAD&CG, Zhejiang University, China  \n[ cs .CV] 6 Jul 2026  \n\n| Context |  |  |\n| --- | --- | --- |\n|  |  | \u003Cbr>Images\u003Cbr> |\n|  |  |  |\n|  |  Reference \u003Cbr>\u003Cbr> |  |\n\n2 vivo BlueImage Lab, China  \n* Equal contribution. †Corresponding author.‡Work done during an internship at vivo BlueImage Lab.  \n2 X. Zhang et al.  \n1 Introduction  \nNovel view synthesis from multiple images is a longstanding problem in computer vision and graphics, forming the foundation for many downstream applications, such as virtual reality, augmented reality, and robotics. Recently, Neural Radiance Fields (NeRF) [21] and 3D Gaussian Splatting (3DGS) [12] have achieved photorealistic rendering by representing scenes with neural networks or 3D Gaussians. However, these methods assume static scene geometry, materials, and lighting—an assumption usually violated for in-the-wild photo collections. Such images, typically sourced from the internet, capture the same scene under varying conditions, exhibiting significant appearance differences in lighting, weather, and camera exposure.  \nTo handle such complex scenarios, the community has historically relied on optimization-based approaches. Pioneering methods like NeRF-W [20] and subsequent appearance-aware 3DGS techniques [8, 13, 36, 41] address this by optimizing per-image appearance embeddings to decouple the shared scene structure from these photometric variations. While these methods achieve impressive appearance interpolation, they strictly require accurately precomputed camera poses, and their performance heavily depends on dense image collections and lengthy per-scene optimization. When only sparse views are available, optimizing per-image embeddings becomes severely under-constrained, leading to inaccurate geometry and a complete lack of reconstruction efficiency.  \nMeanwhile, some methods leverage large-scale multi-view datasets to learn generalizable priors and predict 3D representations in a single forward pass [11, 32, 33, 37] . Without requiring exhaustive per-scene optimization, these feedforward methods enable efficient and photorealistic novel-view synthesis. However, extending such models to in-the-wild photo collections remains highly challenging. When the appearance of the input images varies significantly, existing feedforward frameworks struggle to disentangle underlying 3D geometry from illumination changes, resulting in noticeable appearance inconsistencies in the rendered results.  \nTo this end, we introduce WildSplat, the first feedforward 3D Gaussian Splatting framework designed for in-the-wild scenes with varying appearances. By conditioning the reconstruction on a designated reference image, WildSplat enables photorealistic novel-view synthesis without exhaustive per-scene optimization. Specifically, we propose a dual-branch architecture to explicitly decouple geometry from appearance. This design allows us to reconstruct 3D Gaussians by separating appearance-invariant geometry from target-conditioned appearance. The geometry branch is responsible for predicting this invariant geometry, alongside robust content features that encode the structural information of the scene. To control the rendering appearance, an appearance encoder extracts features from the reference image and injects the appearance information into the content features via a cross-attention-based appearance injector. Furthermore, to stabilize the training process, we introduce a multi-reference training paradigm. Within a single training batch, our framework constructs an appearance-invariant geometry but simultaneously renders it into multiple views under distinct appearance conditions. This strategy enables the model to effectively accommodate photometric  \nWildSplat 3  \nvariations, ensuring ac","cbCaiaNROAEGsuND","https://ap.wps.com/l/cbCaiaNROAEGsuND","pdf",28980824,1,22,"English","en",105,"# Introduction\n## Problem Setting: In-the-Wild Novel View Synthesis\n## Existing Approaches and Limitations\n## WildSplat Overview and Contributions\n# Related Work\n## Novel View Synthesis","[{\"question\":\"What problem does WildSplat address?\",\"answer\":\"WildSplat targets novel-view synthesis from in-the-wild photo collections where camera poses are unposed/sparse and appearance varies due to lighting, weather, and exposure.\"},{\"question\":\"How does WildSplat avoid exhaustive per-scene optimization?\",\"answer\":\"It introduces a feedforward 3D Gaussian Splatting pipeline conditioned on a reference image, using a single forward pass rather than lengthy optimization per scene.\"},{\"question\":\"What is the key architectural idea behind WildSplat?\",\"answer\":\"WildSplat uses a dual-branch architecture that explicitly decouples geometry from appearance, injecting target-conditioned appearance features into content features via cross-attention.\"}]",1784184159,55,{"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},"wildsplat-feedforward-gaussian-splatting-from-unposed-in-the-wild-images","",{"@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/wildsplat-feedforward-gaussian-splatting-from-unposed-in-the-wild-images/82939/",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},"What problem does WildSplat address?","Question",{"text":75,"@type":76},"WildSplat targets novel-view synthesis from in-the-wild photo collections where camera poses are unposed/sparse and appearance varies due to lighting, weather, and exposure.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does WildSplat avoid exhaustive per-scene optimization?",{"text":80,"@type":76},"It introduces a feedforward 3D Gaussian Splatting pipeline conditioned on a reference image, using a single forward pass rather than lengthy optimization per scene.",{"name":82,"@type":73,"acceptedAnswer":83},"What is the key architectural idea behind WildSplat?",{"text":84,"@type":76},"WildSplat uses a dual-branch architecture that explicitly decouples geometry from appearance, injecting target-conditioned appearance features into content features via cross-attention.","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"]