{"paper":{"title":"CArtBench: Evaluating Vision-Language Models on Chinese Art Understanding, Interpretation, and Authenticity","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Vision-language models post high overall scores on Chinese art questions yet drop sharply on evidence linking, expert-style appreciation, and authenticity discrimination.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hidetaka Kamigaito, Hongyao Li, Taro Watanabe, Xuan Zhou, Xuefeng Wei, Yusuke Sakai, Zhi Qu, Zhixuan Wang","submitted_at":"2026-04-13T15:44:02Z","abstract_excerpt":"We introduce CARTBENCH, a museum-grounded benchmark for evaluating vision-language models (VLMs) on Chinese artworks beyond short-form recognition and QA. CARTBENCH comprises four subtasks: CURATORQA for evidence-grounded recognition and reasoning, CATALOGCAPTION for structured four-section expert-style appreciation, REINTERPRET for defensible reinterpretation with expert ratings, and CONNOISSEURPAIRS for diagnostic authenticity discrimination under visually similar confounds. CARTBENCH is built by aligning image-bearing Palace Museum objects from Wikidata with authoritative catalog pages, spa"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across nine representative VLMs, we find that high overall CURATORQA accuracy can mask sharp drops on hard evidence linking and style-to-period inference; long-form appreciation remains far from expert references; and authenticity-oriented diagnostic discrimination stays near chance, underscoring the difficulty of connoisseur-level reasoning for current models.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That alignment of Wikidata objects with authoritative Palace Museum catalog pages and expert ratings provides reliable, unbiased ground truth for defensible reinterpretation and authenticity discrimination tasks.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"CArtBench shows VLMs achieve high scores on easy recognition but drop sharply on evidence linking, style inference, long-form appreciation, and authenticity discrimination for Chinese art.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Vision-language models post high overall scores on Chinese art questions yet drop sharply on evidence linking, expert-style appreciation, and authenticity discrimination.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"31818fdd737ccbab93bf4d159525226aeb1e00682e446a19039096f978d987cb"},"source":{"id":"2604.11632","kind":"arxiv","version":2},"verdict":{"id":"dc9f2b65-79cc-4dd2-b49d-160fbc85f504","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T15:17:24.530599Z","strongest_claim":"Across nine representative VLMs, we find that high overall CURATORQA accuracy can mask sharp drops on hard evidence linking and style-to-period inference; long-form appreciation remains far from expert references; and authenticity-oriented diagnostic discrimination stays near chance, underscoring the difficulty of connoisseur-level reasoning for current models.","one_line_summary":"CArtBench shows VLMs achieve high scores on easy recognition but drop sharply on evidence linking, style inference, long-form appreciation, and authenticity discrimination for Chinese art.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That alignment of Wikidata objects with authoritative Palace Museum catalog pages and expert ratings provides reliable, unbiased ground truth for defensible reinterpretation and authenticity discrimination tasks.","pith_extraction_headline":"Vision-language models post high overall scores on Chinese art questions yet drop sharply on evidence linking, expert-style appreciation, and authenticity discrimination."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.11632/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"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"}