{"paper":{"title":"DPM++: Dynamic Masked Metric Learning for Occluded Person Re-identification","license":"http://creativecommons.org/licenses/by/4.0/","headline":"DPM++ learns an input-adaptive masked metric to select reliable identity subspaces for matching occluded persons.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Lei Tan, Liujuan Cao, Pincong Zou, Pingyang Dai, Yingshi Luan","submitted_at":"2026-05-07T17:47:23Z","abstract_excerpt":"Although person re-identification has made impressive progress, occlusion caused by obstacles remains an unsettled issue in real applications. The difficulty lies in the mismatch between incomplete occluded samples and holistic identity representations. Severe occlusion removes discriminative body cues and introduces interference from background clutter and occluders, making global metric learning unreliable. Existing methods mainly rely on extra pre-trained models to estimate visible parts for alignment or construct occluded samples via data augmentation, but still lack a unified framework th"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"DPM++ learns an input-adaptive masked metric that dynamically selects reliable identity subspaces for each occluded instance, enabling matching to emphasize visibility-consistent evidence while suppressing unreliable components.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the CLIP-derived semantic priors transfer cleanly into the classifier-prototype space and that saliency-guided patch transfer produces occluded samples whose distribution matches real-world occlusion patterns closely enough to improve generalization.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DPM++ learns input-adaptive masked metrics in a CLIP-supervised classifier-prototype space and uses saliency-guided patch transfer to synthesize realistic occluded training samples, outperforming prior methods on occluded and holistic re-ID benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"DPM++ learns an input-adaptive masked metric to select reliable identity subspaces for matching occluded persons.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"5c9845d8a35538e781345a017afc106fba35bec2499b16d68733021a9103826c"},"source":{"id":"2605.06637","kind":"arxiv","version":2},"verdict":{"id":"81ecded6-2a22-4c3a-acb1-af86416a1fee","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-08T12:21:02.458798Z","strongest_claim":"DPM++ learns an input-adaptive masked metric that dynamically selects reliable identity subspaces for each occluded instance, enabling matching to emphasize visibility-consistent evidence while suppressing unreliable components.","one_line_summary":"DPM++ learns input-adaptive masked metrics in a CLIP-supervised classifier-prototype space and uses saliency-guided patch transfer to synthesize realistic occluded training samples, outperforming prior methods on occluded and holistic re-ID benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the CLIP-derived semantic priors transfer cleanly into the classifier-prototype space and that saliency-guided patch transfer produces occluded samples whose distribution matches real-world occlusion patterns closely enough to improve generalization.","pith_extraction_headline":"DPM++ learns an input-adaptive masked metric to select reliable identity subspaces for matching occluded persons."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.06637/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T12:02:03.912970Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-20T07:37:14.475959Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T18:01:19.228721Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T12:32:15.354195Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"dd510343aea6baaacef29f5a1391410e6d4b96a04c9b66b3255a76316d07c2b3"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","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"}