Adaptive Control for 3D Gaussian Splatting: A Systematic Regularization Framework

Adaptive Control for 3D Gaussian Splatting: A Systematic Regularization Framework | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Adaptive Control for 3D Gaussian Splatting: A Systematic Regularization Framework Wenxuan Xiong, Fusheng Wang, Wenbin Liu, Xing Li, Zhidong Zhu, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8145791/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract Regularization in 3D Gaussian Splatting (3D-GS) is often piecemeal, applying uniform penalties that fail to resolve the interdependent trade-offs between detail, smoothness, and stability. This paper moves beyond such ad-hoc solutions by introducing a systematic, context-aware regularization framework for 3D Half-Gaussian Splatting (3D-HGS). Our method acts as an adaptive control system, featuring three synergistic techniques that respond to local scene properties and temporal dynamics. First, we introduce an adaptive opacity consistency loss that uses a dynamic, view-dependent geometric proxy to suppress appearance artifacts on smooth surfaces while preserving sharp boundaries. Second, a selective normal smoothness loss leverages a high-performance CUDA KNN search to enforce geometric coherence exclusively within object interiors, critically protecting edge and corner details from over-smoothing. Finally, a novel EMA-based normal anchoring mechanism provides temporal stability, safeguarding learned geometry against parameter drift during the volatile densification and pruning stages. Our integrated framework establishes a new state-of-the-art. Applied to the strong 3D-HGS baseline, it yields remarkable average PSNR gains, including an exceptional +7.63dB on the challenging Deep Blending dataset. These modular yet synergistic techniques offer a new, principled paradigm for robust and high-fidelity primitive-based rendering. Our source code is available at https://github.com/Archaic-Atom/Adaptive-GS . 3D Gaussian Splatting Neural Rendering Geometric Regularization Point-Based Rendering Novel View Synthesis Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 09 Jan, 2026 Reviews received at journal 03 Jan, 2026 Reviews received at journal 27 Dec, 2025 Reviewers agreed at journal 16 Dec, 2025 Reviewers agreed at journal 10 Dec, 2025 Reviewers invited by journal 09 Dec, 2025 Editor assigned by journal 19 Nov, 2025 Submission checks completed at journal 18 Nov, 2025 First submitted to journal 18 Nov, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8145791","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":558236083,"identity":"986a675f-bcea-4875-a6a9-5442e8d218df","order_by":0,"name":"Wenxuan Xiong","email":"","orcid":"","institution":"Nanchang Hangkong University","correspondingAuthor":false,"prefix":"","firstName":"Wenxuan","middleName":"","lastName":"Xiong","suffix":""},{"id":558236084,"identity":"0e0d3c3f-4534-45dc-81c4-534b8be9ecc8","order_by":1,"name":"Fusheng Wang","email":"","orcid":"","institution":"Nanchang Hangkong University","correspondingAuthor":false,"prefix":"","firstName":"Fusheng","middleName":"","lastName":"Wang","suffix":""},{"id":558236085,"identity":"e24b0823-58e7-4d93-b936-7021d0f01e6d","order_by":2,"name":"Wenbin Liu","email":"","orcid":"","institution":"Nanchang Hangkong University","correspondingAuthor":false,"prefix":"","firstName":"Wenbin","middleName":"","lastName":"Liu","suffix":""},{"id":558236086,"identity":"49601e2c-58ef-43ae-8dc5-22e9ebd1061e","order_by":3,"name":"Xing Li","email":"","orcid":"","institution":"Nanchang Hangkong University","correspondingAuthor":false,"prefix":"","firstName":"Xing","middleName":"","lastName":"Li","suffix":""},{"id":558236087,"identity":"2ed1c060-b84d-4e26-8856-24854fbbebbd","order_by":4,"name":"Zhidong Zhu","email":"","orcid":"","institution":"Beihang University","correspondingAuthor":false,"prefix":"","firstName":"Zhidong","middleName":"","lastName":"Zhu","suffix":""},{"id":558236088,"identity":"f118a7d0-ee12-4a04-bc4b-cf15945f4b23","order_by":5,"name":"Bangshu Xiong","email":"","orcid":"","institution":"Nanchang Hangkong University","correspondingAuthor":false,"prefix":"","firstName":"Bangshu","middleName":"","lastName":"Xiong","suffix":""},{"id":558236089,"identity":"edbf579e-6d13-4105-ac66-c18cdb0c50d3","order_by":6,"name":"Zhibo Rao","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzUlEQVRIie2QMQ6CQBBFt4JmEtoxIXiFJUal4DBLNoFGEo9gJYXrXbiBRJKtsJdgQUXFASgoBFoNrJ3FvmqKefnzhxCN5l9hne+Yp3GinqISiHAD2aSgaoxxD8SkEAVln+QSa8giYYpt2h2RWMmFzSp2EXLKvFcsQIalGA7D4pHOKkgObs2giW/I5RMGhWK8oFgtzZiRR4DBueyVFBxTjJzBkFKppWDD6fBkd+xS2RRhuYvF5arr/DWYYle2ve9YyXVe+QB+W9doNBrNV95sHUQQfERdswAAAABJRU5ErkJggg==","orcid":"","institution":"Nanchang Hangkong University","correspondingAuthor":true,"prefix":"","firstName":"Zhibo","middleName":"","lastName":"Rao","suffix":""}],"badges":[],"createdAt":"2025-11-18 13:08:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8145791/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8145791/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":98232087,"identity":"4f94a0cf-87cd-40c2-a21a-8604ec1ec4b1","added_by":"auto","created_at":"2025-12-15 13:47:27","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":52626530,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8145791/v1/b6281d1c209446786d468b02.pdf"},{"id":98232086,"identity":"f430f317-df5e-45ab-b865-e0bd1f937315","added_by":"auto","created_at":"2025-12-15 13:47:26","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":8859,"visible":true,"origin":"","legend":"","description":"","filename":"daa95af1c1bb4545b082babfcf10055d.json","url":"https://assets-eu.researchsquare.com/files/rs-8145791/v1/d3bd836a60d3e485d386d94f.json"},{"id":98434302,"identity":"d3c5c7fe-cdaf-41c1-8d5e-c3c62cb26d93","added_by":"auto","created_at":"2025-12-17 16:51:50","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3869176,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8145791/v1_covered_f8e508ed-3ee8-4a94-8fb5-279b473db654.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Adaptive Control for 3D Gaussian Splatting: A Systematic Regularization Framework","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"the-visual-computer","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"tvcj","sideBox":"Learn more about [The Visual Computer](http://link.springer.com/journal/371)","snPcode":"371","submissionUrl":"https://submission.nature.com/new-submission/371/3","title":"The Visual Computer","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"3D Gaussian Splatting, Neural Rendering, Geometric Regularization, Point-Based Rendering, Novel View Synthesis","lastPublishedDoi":"10.21203/rs.3.rs-8145791/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8145791/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Regularization in 3D Gaussian Splatting (3D-GS) is often piecemeal, applying uniform penalties that fail to resolve the interdependent trade-offs between detail, smoothness, and stability. This paper moves beyond such ad-hoc solutions by introducing a systematic, context-aware regularization framework for 3D Half-Gaussian Splatting (3D-HGS). Our method acts as an adaptive control system, featuring three synergistic techniques that respond to local scene properties and temporal dynamics. First, we introduce an adaptive opacity consistency loss that uses a dynamic, view-dependent geometric proxy to suppress appearance artifacts on smooth surfaces while preserving sharp boundaries. Second, a selective normal smoothness loss leverages a high-performance CUDA KNN search to enforce geometric coherence exclusively within object interiors, critically protecting edge and corner details from over-smoothing. Finally, a novel EMA-based normal anchoring mechanism provides temporal stability, safeguarding learned geometry against parameter drift during the volatile densification and pruning stages. Our integrated framework establishes a new state-of-the-art. Applied to the strong 3D-HGS baseline, it yields remarkable average PSNR gains, including an exceptional +7.63dB on the challenging Deep Blending dataset. These modular yet synergistic techniques offer a new, principled paradigm for robust and high-fidelity primitive-based rendering. Our source code is available at https://github.com/Archaic-Atom/Adaptive-GS.","manuscriptTitle":"Adaptive Control for 3D Gaussian Splatting: A Systematic Regularization Framework","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-15 13:47:21","doi":"10.21203/rs.3.rs-8145791/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-01-09T06:01:24+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-01-03T10:52:44+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-12-28T01:12:15+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"117656344848644268826140325445170875010","date":"2025-12-17T00:41:03+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"156846004021078028761095292316125072543","date":"2025-12-10T12:22:23+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-12-10T00:13:45+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-11-19T22:00:34+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-11-19T03:22:21+00:00","index":"","fulltext":""},{"type":"submitted","content":"The Visual Computer","date":"2025-11-18T12:55:44+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"the-visual-computer","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"tvcj","sideBox":"Learn more about [The Visual Computer](http://link.springer.com/journal/371)","snPcode":"371","submissionUrl":"https://submission.nature.com/new-submission/371/3","title":"The Visual Computer","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"d6ce69f3-6061-4329-bdbe-953f5425cba9","owner":[],"postedDate":"December 15th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-05T21:38:59+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-15 13:47:21","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8145791","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8145791","identity":"rs-8145791","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}