{"paper":{"title":"Segment Anything with Robust Uncertainty-Accuracy Correlation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"RUAC adds a lightweight uncertainty head to SAM and trains it with joint style-deformation attacks plus alignment to keep uncertainty tied to pixel errors under domain shifts.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Hongyou Zhou, Ling Shao, Marc Toussaint, Zihan Ye","submitted_at":"2026-05-11T14:04:02Z","abstract_excerpt":"Despite strong zero-shot performance, SAM is unreliable under domain shift due to Mask-level Confidence Confusion (MCC), where a single IoU-based mask score fails to reflect pixel-wise reliability near boundaries. Motivated by the contrast between texture-biased shortcuts in neural networks and shape-centric processing in human vision, we model out-of-domain variation as appearance shifts and non-rigid deformations that jointly stress calibration. We propose Segment Anything with Robust Uncertainty-Accuracy Correlation (RUAC) for robust pixel-wise uncertainty estimation under appearance and de"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across 23 zero-shot domains, RUAC improves segmentation quality and yields more faithful uncertainty with stronger uncertainty-accuracy correlation.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That out-of-domain variation can be adequately modeled as appearance shifts and non-rigid deformations, and that the collaborative style-deformation attack plus Uncertainty-Accuracy Alignment will produce uncertainty estimates that remain faithful under real-world domain shifts beyond the 23 tested domains.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"RUAC adds an uncertainty head to SAM, trains it via collaborative style-deformation attacks and Uncertainty-Accuracy Alignment, and reports improved segmentation quality plus stronger uncertainty-accuracy correlation across 23 zero-shot domains.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"RUAC adds a lightweight uncertainty head to SAM and trains it with joint style-deformation attacks plus alignment to keep uncertainty tied to pixel errors under domain shifts.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"67d5a0db3800c50a2e4d342d93abb48fc853117a312b93fd8b603a54a46b84c1"},"source":{"id":"2605.10603","kind":"arxiv","version":2},"verdict":{"id":"5f03a62a-1f79-4367-94de-e7cc7ef09c91","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T05:22:38.892348Z","strongest_claim":"Across 23 zero-shot domains, RUAC improves segmentation quality and yields more faithful uncertainty with stronger uncertainty-accuracy correlation.","one_line_summary":"RUAC adds an uncertainty head to SAM, trains it via collaborative style-deformation attacks and Uncertainty-Accuracy Alignment, and reports improved segmentation quality plus stronger uncertainty-accuracy correlation across 23 zero-shot domains.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That out-of-domain variation can be adequately modeled as appearance shifts and non-rigid deformations, and that the collaborative style-deformation attack plus Uncertainty-Accuracy Alignment will produce uncertainty estimates that remain faithful under real-world domain shifts beyond the 23 tested domains.","pith_extraction_headline":"RUAC adds a lightweight uncertainty head to SAM and trains it with joint style-deformation attacks plus alignment to keep uncertainty tied to pixel errors under domain shifts."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.10603/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-19T14:40:49.638185Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T11:01:17.557179Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T09:09:01.967018Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"1c1aed7f2f5ce685ed2e5899194dd4b7e11bee110849973b5611badca8ce7255"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"8cefc2972f83c7e9ee526ce1122bd20c6b660546a4791ee7e4724ea52e4071bf"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}