{"paper":{"title":"CREST: Curvature-Regulated Event-Centric Sampling for Efficient Long-Video Understanding","license":"http://creativecommons.org/licenses/by/4.0/","headline":"CATS selects long-video frames by tracking curvature in query-relevance scores over time.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Abdul Mohaimen Al Radi, Ismat Rahman, Md Mosaddek Khan, Md. Tanvir Alam, Mehrajul Abadin Miraj, Shariful Islam Rayhan, Yu Tian","submitted_at":"2026-05-09T23:47:46Z","abstract_excerpt":"Selecting informative frames from long videos is a combinatorial problem that existing methods address either through efficient heuristics without explicit modeling of query-conditioned temporal structure, or through multi stage retrieval pipelines with substantial preprocessing cost. We propose \\textbf{CREST}, a training-free frame selection method grounded in the temporal geometry of query--frame relevance. CREST is based on the observation that relevance over time exhibits structured local variation: sharp curvature around salient events and flatter regions in redundant segments. By using l"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Under a fixed backbone and frame budget, CATS consistently outperforms prior lightweight approaches such as AKS on LongVideoBench and VideoMME. CATS retains approximately 93-95% of MIRA's performance while requiring only 3-4% of its preprocessing cost.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That curvature of query-frame relevance scores reliably identifies salient events and surrounding context across diverse video domains without systematic omission of critical information.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"CATS uses temporal curvature of query-frame relevance to select informative frames, achieving 93-95% of heavy multi-stage accuracy at 3-4% of the preprocessing cost on long-video benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"CATS selects long-video frames by tracking curvature in query-relevance scores over time.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"e19d20be28c3efd511d6805af14f701fd424f3fb3ff0a6039bf3c0ce696666a6"},"source":{"id":"2605.09223","kind":"arxiv","version":2},"verdict":{"id":"5ca6f1b6-8978-4034-80c7-aa316c050218","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T03:02:55.452855Z","strongest_claim":"Under a fixed backbone and frame budget, CATS consistently outperforms prior lightweight approaches such as AKS on LongVideoBench and VideoMME. CATS retains approximately 93-95% of MIRA's performance while requiring only 3-4% of its preprocessing cost.","one_line_summary":"CATS uses temporal curvature of query-frame relevance to select informative frames, achieving 93-95% of heavy multi-stage accuracy at 3-4% of the preprocessing cost on long-video benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That curvature of query-frame relevance scores reliably identifies salient events and surrounding context across diverse video domains without systematic omission of critical information.","pith_extraction_headline":"CATS selects long-video frames by tracking curvature in query-relevance scores over time."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.09223/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T08:02:09.568722Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T20:35:05.307347Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T13:31:18.149097Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T10:27:51.775941Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"d678533d0540d0dfb1fd734efc217246561df29928c3513fd601e3085b44d08a"},"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"}