Efficiency-Aware Crowd Counting: A Framework for Adaptive Model Selection and Deployment

Efficiency-Aware Crowd Counting: A Framework for Adaptive Model Selection and Deployment | 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 Article Efficiency-Aware Crowd Counting: A Framework for Adaptive Model Selection and Deployment Preetam Giridhar Nadoni, Prital Rajkumar Nyamagoud, Rohini R, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9400486/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 Existing crowd counting systems deploy a single fixed model regardless of scene density or target hardware, causing lightweight models to catastrophically underestimate extremely dense crowds by 900–1,900 people (47–73% relative error). We present AdaptiveCount, a three-stage framework that classifies each scene into one of four density levels and routes it to the most appropriate model from a portfolio of four architectures: MCNN, EdgeCrowdNet, EfficientCSRNet, and CSRNet. A lightweight ResNet-18-based density classifier achieves 94.3% accuracy with fewer than 10 ms overhead. We further introduce the Density-Adaptive Loss (DAL), a pixel-level weighting function inspired by information-theoretic principles, which assigns greater penalty to errors in high-density regions and improves overall MAE by 12.3% relative to standard MSE on heterogeneous scenes. Three results define AdaptiveCount’s practical contribution: (i) a Pareto-frontier analysis across twelve 2019–2026 methods on three benchmarks (ShanghaiTech, UCF-QNRF, NWPU-Crowd) confirms AdaptiveCount is the only system achieving MAE ≤ 200 with CPU FPS ≥ 2.0 and memory ≤ 300 MB simultaneously; (ii) five-platform hardware benchmarking (cloud GPU to Raspberry Pi 4 and smartphones) shows CSRNet fails with an out-of-memory error while AdaptiveCount sustains 4.2 FPS within 287 MB; and (iii) a campus deployment case study on a live camera confirms generalisation to continuous real-world operation (4.1 FPS average, 16.8% relative error). Statistical significance is confirmed (p < 0.05, five independent runs). Code and models are openly available at https://github.com/preetamgn09/AdaptiveCount . Physical sciences/Engineering Physical sciences/Mathematics and computing Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 12 May, 2026 Reviewers agreed at journal 07 May, 2026 Reviewers invited by journal 07 May, 2026 Editor assigned by journal 07 May, 2026 Editor invited by journal 06 May, 2026 Submission checks completed at journal 28 Apr, 2026 First submitted to journal 28 Apr, 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-9400486","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":641545663,"identity":"7e6c418f-5982-4259-a387-d4786365b57d","order_by":0,"name":"Preetam Giridhar Nadoni","email":"","orcid":"","institution":"B.M.S. College of Engineering","correspondingAuthor":false,"prefix":"","firstName":"Preetam","middleName":"Giridhar","lastName":"Nadoni","suffix":""},{"id":641545664,"identity":"c82c0cf5-2308-4e08-bf3d-b4cb0209baf0","order_by":1,"name":"Prital Rajkumar Nyamagoud","email":"","orcid":"","institution":"B.M.S. College of Engineering","correspondingAuthor":false,"prefix":"","firstName":"Prital","middleName":"Rajkumar","lastName":"Nyamagoud","suffix":""},{"id":641545665,"identity":"9b236809-f91f-4557-8d29-15c3315262fe","order_by":2,"name":"Rohini R","email":"","orcid":"","institution":"B.M.S. College of Engineering","correspondingAuthor":false,"prefix":"","firstName":"Rohini","middleName":"","lastName":"R","suffix":""},{"id":641545666,"identity":"c8a812fe-d0e3-441b-8fd0-da77990dd0d4","order_by":3,"name":"Nivedana J","email":"","orcid":"","institution":"B.M.S. College of Engineering","correspondingAuthor":false,"prefix":"","firstName":"Nivedana","middleName":"","lastName":"J","suffix":""},{"id":641545667,"identity":"25539585-97f1-41c6-9883-a27ebcf33ef1","order_by":4,"name":"Rashmi K.B","email":"","orcid":"","institution":"B.M.S. College of Engineering","correspondingAuthor":false,"prefix":"","firstName":"Rashmi","middleName":"","lastName":"K.B","suffix":""},{"id":641545668,"identity":"106aac8f-4ac8-4efe-90b1-b1e50d84dc93","order_by":5,"name":"Shobana T.S","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABCUlEQVRIiWNgGAWjYDACCQY2KIuxgYGh4gBcgoeQFgmIljOkaQFZ1HYAp0I4kJ/dfOzBzx12dfINzM0fPs67Ey3ffvgAw48aBhlzHFoM7hxLN+w9kyxhcICxTXLmtme5G86kJTD2HGPgsWzAoUUix0yCt41ZwgDoKmbebYdzN0jwGDDwNjDwGOBwpPyMHDPJv231EvINjM2f/845nDt/Bv8Hxr94tDDcyDGT5m07LMFwgLFBmrHhcG7DDR4GZny2GNxISzeWbTsuueEw0C89x8B+MTgsc0wCj8OSjz1821bNL9/e/vjDj5o7ufPbDz98+KbGxh6nw+CAGYl9ABZRo2AUjIJRMArIAwBqzFx8unWYkwAAAABJRU5ErkJggg==","orcid":"","institution":"B.M.S. College of Engineering","correspondingAuthor":true,"prefix":"","firstName":"Shobana","middleName":"","lastName":"T.S","suffix":""}],"badges":[],"createdAt":"2026-04-13 07:54:03","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9400486/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9400486/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":109759679,"identity":"01e9c8a0-c3d6-406b-ba3a-c5579edac08e","added_by":"auto","created_at":"2026-05-22 07:27:31","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1628146,"visible":true,"origin":"","legend":"","description":"","filename":"EfficiencyAwareCrowdCounting..pdf","url":"https://assets-eu.researchsquare.com/files/rs-9400486/v1_covered_60612b32-7587-44af-808a-1a026b8a26be.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Efficiency-Aware Crowd Counting: A Framework for Adaptive Model Selection and Deployment","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":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-9400486/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9400486/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Existing crowd counting systems deploy a single fixed model regardless of scene density or target hardware, causing lightweight models to catastrophically underestimate extremely dense crowds by 900–1,900 people (47–73% relative error). We present AdaptiveCount, a three-stage framework that classifies each scene into one of four density levels and routes it to the most appropriate model from a portfolio of four architectures: MCNN, EdgeCrowdNet, EfficientCSRNet, and CSRNet. A lightweight ResNet-18-based density classifier achieves 94.3% accuracy with fewer than 10 ms overhead. We further introduce the Density-Adaptive Loss (DAL), a pixel-level weighting function inspired by information-theoretic principles, which assigns greater penalty to errors in high-density regions and improves overall MAE by 12.3% relative to standard MSE on heterogeneous scenes. Three results define AdaptiveCount’s practical contribution: (i) a Pareto-frontier analysis across twelve 2019–2026 methods on three benchmarks (ShanghaiTech, UCF-QNRF, NWPU-Crowd) confirms AdaptiveCount is the only system achieving MAE ≤ 200 with CPU FPS ≥ 2.0 and memory ≤ 300 MB simultaneously; (ii) five-platform hardware benchmarking (cloud GPU to Raspberry Pi 4 and smartphones) shows CSRNet fails with an out-of-memory error while AdaptiveCount sustains 4.2 FPS within 287 MB; and (iii) a campus deployment case study on a live camera confirms generalisation to continuous real-world operation (4.1 FPS average, 16.8% relative error). Statistical significance is confirmed (p \u003c 0.05, five independent runs). Code and models are openly available at https://github.com/preetamgn09/AdaptiveCount.","manuscriptTitle":"Efficiency-Aware Crowd Counting: A Framework for Adaptive Model Selection and Deployment","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-18 12:43:01","doi":"10.21203/rs.3.rs-9400486/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-05-12T20:10:06+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"235612845007739470302863073012205130448","date":"2026-05-07T14:07:46+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-05-07T13:58:04+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-05-07T13:57:07+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-05-06T09:21:44+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-28T10:36:48+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2026-04-28T09:27:54+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"97293b2d-8003-49f0-b2f0-bb6c41f526a4","owner":[],"postedDate":"May 18th, 2026","published":true,"recentEditorialEvents":[{"type":"editorInvitedReview","content":"","date":"2026-05-12T20:10:06+00:00","index":33,"fulltext":""},{"type":"reviewerAgreed","content":"235612845007739470302863073012205130448","date":"2026-05-07T14:07:46+00:00","index":32,"fulltext":""},{"type":"reviewersInvited","content":"6","date":"2026-05-07T13:58:04+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-05-07T13:57:07+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-05-06T09:21:44+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":68240208,"name":"Physical sciences/Engineering"},{"id":68240209,"name":"Physical sciences/Mathematics and computing"}],"tags":[],"updatedAt":"2026-05-18T12:43:01+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-18 12:43:01","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9400486","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9400486","identity":"rs-9400486","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}