{"paper":{"title":"MobileAgeNet: Lightweight Facial Age Estimation for Mobile Deployment","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"MobileAgeNet shows a compact network can estimate facial age accurately enough for mobile phones while keeping inference fast after format conversion.","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Arun Kumar, Aswathy Baiju, Dmitry Ignatov, Radu Timofte","submitted_at":"2026-04-18T14:37:23Z","abstract_excerpt":"Mobile deployment of facial age estimation requires models that balance predictive accuracy with low latency and compact size. In this work, we present MobileAgeNet, a lightweight age-regression framework that achieves an MAE of 4.65 years on the UTKFace held-out test set while maintaining efficient on-device inference with an average latency of 14.4 ms measured using the AI Benchmark application. The model is built on a pretrained MobileNetV3-Large backbone combined with a compact regression head, enabling real-time prediction on mobile devices. The training and evaluation pipeline is integra"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"MobileAgeNet achieves an MAE of 4.65 years on the UTKFace held-out test set while maintaining efficient on-device inference with an average latency of 14.4 ms measured using the AI Benchmark application.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that the bounded age regression and two-stage fine-tuning strategy improve generalization and training stability on the specific dataset without introducing bias or limiting applicability to other face datasets or real-world conditions.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MobileAgeNet uses a MobileNetV3-Large backbone with a regression head to achieve 4.65 years mean absolute error in age estimation and 14.4 ms on-device latency with 3.23 million parameters.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"MobileAgeNet shows a compact network can estimate facial age accurately enough for mobile phones while keeping inference fast after format conversion.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"66ff11fba4616447a91bc60d71b91a0761f5bee675640a74f7f711d898c6a969"},"source":{"id":"2604.17007","kind":"arxiv","version":1},"verdict":{"id":"513309c1-924e-477a-8267-5942c07ac500","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T07:28:48.894708Z","strongest_claim":"MobileAgeNet achieves an MAE of 4.65 years on the UTKFace held-out test set while maintaining efficient on-device inference with an average latency of 14.4 ms measured using the AI Benchmark application.","one_line_summary":"MobileAgeNet uses a MobileNetV3-Large backbone with a regression head to achieve 4.65 years mean absolute error in age estimation and 14.4 ms on-device latency with 3.23 million parameters.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that the bounded age regression and two-stage fine-tuning strategy improve generalization and training stability on the specific dataset without introducing bias or limiting applicability to other face datasets or real-world conditions.","pith_extraction_headline":"MobileAgeNet shows a compact network can estimate facial age accurately enough for mobile phones while keeping inference fast after format conversion."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.17007/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":26,"sample":[{"doi":"","year":2026,"title":"https://susanqq.github.io/UTKFace/","work_id":"09531a4e-305e-49dd-8832-c82203013c4e","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Age estimation via face images: a survey.EURASIP Journal on Image and Video Processing, 2018(1):42, 2018","work_id":"773ca9e4-e4af-44a6-8307-90dcea5e6192","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"DAA: A delta age adain operation for age estimation via binary code transformer","work_id":"f86af4d2-a6fe-4e79-bb6f-75906029505e","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Using ranking-CNN for age estimation","work_id":"5fb051b5-852f-4c00-9786-383b11b4c1cc","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2009,"title":"Imagenet: A large-scale hierarchical image database","work_id":"70a1aeee-71a7-428c-9b45-ceea336e8722","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":26,"snapshot_sha256":"7a8924b52d048af2104672840a7fcabe93c7dcf217f5eb4fbf6c0e77ed8d7b1d","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}