Efficient RGB-D Semantic Segmentation via Nested Decoder Topology and Enhanced Supervision

Efficient RGB-D Semantic Segmentation via Nested Decoder Topology and Enhanced Supervision | 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 Efficient RGB-D Semantic Segmentation via Nested Decoder Topology and Enhanced Supervision Hang Jian, Tianxiang Xie, Yuanfu Ku, Keyan Shen, Ruhuan Du This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9229815/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract RGB-D semantic segmentation benefits from complementary appearance and geometry cues, yet efficient models still suffer from insufficient feature alignment during decoding and limited supervision on newly formed high-resolution paths. Instead of redesigning the already mature encoder-side fusion of SGACNet, this study follows an attributable design principle: preserve the encoding and context aggregation paths, and intervene only in decoder topology and supervision. Specifically, the method adopts a nested decoding topology for progressive feature reconstruction and imposes enhanced supervision on the added high-resolution nodes while retaining the original multi-scale side outputs. Because the encoder backbone and the context aggregation path are preserved, the effect of these decoder-side changes can be interpreted more directly. Experiments on NYUv2 show that, at the best validation checkpoint of each model, SGACNet + + DS improves pixel accuracy, harmonic intersection over union, mean intersection over union, and mean accuracy over SGACNet by 2.04, 0.64, 2.67, and 2.62 percentage points, respectively. The ablation results indicate that the revised decoder topology mainly improves feature alignment during reconstruction, whereas the enhanced supervision strategy mainly improves optimization and stabilizes high-resolution training. These results suggest that, for efficient RGB-D baselines with mature encoder-side fusion, decoder topology and supervision density remain important determinants of final segmentation quality. RGB-D semantic segmentation SGACNet nested decoder topology enhanced supervision progressive feature reconstruction Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 27 Apr, 2026 Reviews received at journal 25 Apr, 2026 Reviewers agreed at journal 22 Apr, 2026 Reviewers invited by journal 22 Apr, 2026 Editor assigned by journal 28 Mar, 2026 Submission checks completed at journal 26 Mar, 2026 First submitted to journal 26 Mar, 2026 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-9229815","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":630035923,"identity":"7d984d8e-5d1f-4b43-bccb-b9909b9c892d","order_by":0,"name":"Hang Jian","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA10lEQVRIiWNgGAWjYBACfmbmAwc+/rHh4WdvPkCcFsn2tsSDMxvS5CR7jiUQp8XgzBnjw7wNh40NbuQYEOmyGwkGB3h3pCVuuJHz8cYbBjs53QYCOhhnJCQckDxjkzjzzNvNlnMYko3NDhDQwiyRcOCAAVtaYt/x3G3SPAwHErcR0sImkdhwIIHtMJDMeUacFh6ewwwHDrYdNhY4kcNGnBYJ9jaGgw1nwIFsbDnHgAi/2B/m//z5TwU4Kh/eeFNhJ0dQC6qVPMRGDZIWUnWMglEwCkbBiAAANcdNXzMxWHYAAAAASUVORK5CYII=","orcid":"","institution":"Southwest Petroleum University","correspondingAuthor":true,"prefix":"","firstName":"Hang","middleName":"","lastName":"Jian","suffix":""},{"id":630035924,"identity":"c0179461-1124-46f8-981a-299c53b63b5c","order_by":1,"name":"Tianxiang Xie","email":"","orcid":"","institution":"Southwest Petroleum University","correspondingAuthor":false,"prefix":"","firstName":"Tianxiang","middleName":"","lastName":"Xie","suffix":""},{"id":630035925,"identity":"0c1d6251-4865-432d-a3da-32937774d1c8","order_by":2,"name":"Yuanfu Ku","email":"","orcid":"","institution":"Southwest Petroleum University","correspondingAuthor":false,"prefix":"","firstName":"Yuanfu","middleName":"","lastName":"Ku","suffix":""},{"id":630035926,"identity":"d4074999-2124-414b-b774-3c947ae3e2c1","order_by":3,"name":"Keyan Shen","email":"","orcid":"","institution":"Southwest Petroleum University","correspondingAuthor":false,"prefix":"","firstName":"Keyan","middleName":"","lastName":"Shen","suffix":""},{"id":630035927,"identity":"334cb4c7-a4c8-42d5-94e2-54559f89648e","order_by":4,"name":"Ruhuan Du","email":"","orcid":"","institution":"Southwest Petroleum University","correspondingAuthor":false,"prefix":"","firstName":"Ruhuan","middleName":"","lastName":"Du","suffix":""}],"badges":[],"createdAt":"2026-03-26 06:09:32","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9229815/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9229815/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108491488,"identity":"3fe9cb25-8580-4e99-9dd6-dee404998c9e","added_by":"auto","created_at":"2026-05-05 09:54:09","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":418733,"visible":true,"origin":"","legend":"","description":"","filename":"ManuscriptMVA.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9229815/v1_covered_b83417b1-36e3-4d3d-9bf5-31cabaf65c49.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Efficient RGB-D Semantic Segmentation via Nested Decoder Topology and Enhanced Supervision","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"machine-vision-and-applications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"mvap","sideBox":"Learn more about [Machine Vision and Applications](https://www.springer.com/journal/138)","snPcode":"138","submissionUrl":"https://submission.springernature.com/new-submission/138/3","title":"Machine Vision and Applications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"RGB-D semantic segmentation, SGACNet, nested decoder topology, enhanced supervision, progressive feature reconstruction","lastPublishedDoi":"10.21203/rs.3.rs-9229815/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9229815/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eRGB-D semantic segmentation benefits from complementary appearance and geometry cues, yet efficient models still suffer from insufficient feature alignment during decoding and limited supervision on newly formed high-resolution paths. Instead of redesigning the already mature encoder-side fusion of SGACNet, this study follows an attributable design principle: preserve the encoding and context aggregation paths, and intervene only in decoder topology and supervision. Specifically, the method adopts a nested decoding topology for progressive feature reconstruction and imposes enhanced supervision on the added high-resolution nodes while retaining the original multi-scale side outputs. Because the encoder backbone and the context aggregation path are preserved, the effect of these decoder-side changes can be interpreted more directly. Experiments on NYUv2 show that, at the best validation checkpoint of each model, SGACNet\u0026thinsp;+\u0026thinsp;+\u0026thinsp;DS improves pixel accuracy, harmonic intersection over union, mean intersection over union, and mean accuracy over SGACNet by 2.04, 0.64, 2.67, and 2.62 percentage points, respectively. The ablation results indicate that the revised decoder topology mainly improves feature alignment during reconstruction, whereas the enhanced supervision strategy mainly improves optimization and stabilizes high-resolution training. These results suggest that, for efficient RGB-D baselines with mature encoder-side fusion, decoder topology and supervision density remain important determinants of final segmentation quality.\u003c/p\u003e","manuscriptTitle":"Efficient RGB-D Semantic Segmentation via Nested Decoder Topology and Enhanced Supervision","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-01 04:43:24","doi":"10.21203/rs.3.rs-9229815/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"161386029037356121643209605522839612749","date":"2026-04-27T06:02:33+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-25T12:21:06+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"231913494159878522499875811455332803342","date":"2026-04-22T06:00:26+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-22T05:52:42+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-28T18:13:27+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-26T13:37:55+00:00","index":"","fulltext":""},{"type":"submitted","content":"Machine Vision and Applications","date":"2026-03-26T06:01:14+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"machine-vision-and-applications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"mvap","sideBox":"Learn more about [Machine Vision and Applications](https://www.springer.com/journal/138)","snPcode":"138","submissionUrl":"https://submission.springernature.com/new-submission/138/3","title":"Machine Vision and Applications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"ef264e0a-4fc6-4005-a8a6-bab01aa341f6","owner":[],"postedDate":"May 1st, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-01T04:43:24+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-01 04:43:24","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9229815","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9229815","identity":"rs-9229815","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}