{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:MCOFSJK5ZSVABNUPQFQ7K64TV7","short_pith_number":"pith:MCOFSJK5","schema_version":"1.0","canonical_sha256":"609c59255dccaa00b68f8161f57b93afc593e36bb69987ec99971e7aef4172d2","source":{"kind":"arxiv","id":"2312.14232","version":3},"attestation_state":"computed","paper":{"title":"Parrot Captions Teach CLIP to Spot Text","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Alex Jinpeng Wang, Bin Wang, Conghui He, Mike Zheng Shou, Weijia Li, Yiqi Lin","submitted_at":"2023-12-21T18:46:46Z","abstract_excerpt":"Despite CLIP being the foundation model in numerous vision-language applications, the CLIP suffers from a severe text spotting bias. Such bias causes CLIP models to `Parrot' the visual text embedded within images while disregarding the authentic visual semantics. We uncover that in the most popular image-text dataset LAION-2B, the captions also densely parrot (spell) the text embedded in images. Our analysis shows that around 50% of images are embedded with visual text content, and around 30% of captions words are in these embedded visual content. Based on such observation, we thoroughly inspe"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2312.14232","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2023-12-21T18:46:46Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"8e758d1d56fa2cf4269fe3cb21a1ad23ca378081f98e979d51f83511801bf0b9","abstract_canon_sha256":"9c38f6aa2e1dacc0109e6657031eac42acf686802c57739cbef9ae2f95761f75"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:40:05.028418Z","signature_b64":"7+rdJ8eiVxqWFiSRNZHX4/InjbLB94TE3jd5vZ5hIVEYBnKC7jFg1G/zs6KUTcPTP+8Yc6L5wQpkhOumMA73Cw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"609c59255dccaa00b68f8161f57b93afc593e36bb69987ec99971e7aef4172d2","last_reissued_at":"2026-07-05T07:40:05.027933Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:40:05.027933Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Parrot Captions Teach CLIP to Spot Text","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Alex Jinpeng Wang, Bin Wang, Conghui He, Mike Zheng Shou, Weijia Li, Yiqi Lin","submitted_at":"2023-12-21T18:46:46Z","abstract_excerpt":"Despite CLIP being the foundation model in numerous vision-language applications, the CLIP suffers from a severe text spotting bias. Such bias causes CLIP models to `Parrot' the visual text embedded within images while disregarding the authentic visual semantics. We uncover that in the most popular image-text dataset LAION-2B, the captions also densely parrot (spell) the text embedded in images. Our analysis shows that around 50% of images are embedded with visual text content, and around 30% of captions words are in these embedded visual content. Based on such observation, we thoroughly inspe"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2312.14232","kind":"arxiv","version":3},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2312.14232/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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2312.14232","created_at":"2026-07-05T07:40:05.027990+00:00"},{"alias_kind":"arxiv_version","alias_value":"2312.14232v3","created_at":"2026-07-05T07:40:05.027990+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2312.14232","created_at":"2026-07-05T07:40:05.027990+00:00"},{"alias_kind":"pith_short_12","alias_value":"MCOFSJK5ZSVA","created_at":"2026-07-05T07:40:05.027990+00:00"},{"alias_kind":"pith_short_16","alias_value":"MCOFSJK5ZSVABNUP","created_at":"2026-07-05T07:40:05.027990+00:00"},{"alias_kind":"pith_short_8","alias_value":"MCOFSJK5","created_at":"2026-07-05T07:40:05.027990+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.17375","citing_title":"When Text Hijacks Vision: Benchmarking and Mitigating Text Overlay-Induced Hallucination in Vision Language Models","ref_index":32,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MCOFSJK5ZSVABNUPQFQ7K64TV7","json":"https://pith.science/pith/MCOFSJK5ZSVABNUPQFQ7K64TV7.json","graph_json":"https://pith.science/api/pith-number/MCOFSJK5ZSVABNUPQFQ7K64TV7/graph.json","events_json":"https://pith.science/api/pith-number/MCOFSJK5ZSVABNUPQFQ7K64TV7/events.json","paper":"https://pith.science/paper/MCOFSJK5"},"agent_actions":{"view_html":"https://pith.science/pith/MCOFSJK5ZSVABNUPQFQ7K64TV7","download_json":"https://pith.science/pith/MCOFSJK5ZSVABNUPQFQ7K64TV7.json","view_paper":"https://pith.science/paper/MCOFSJK5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2312.14232&json=true","fetch_graph":"https://pith.science/api/pith-number/MCOFSJK5ZSVABNUPQFQ7K64TV7/graph.json","fetch_events":"https://pith.science/api/pith-number/MCOFSJK5ZSVABNUPQFQ7K64TV7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MCOFSJK5ZSVABNUPQFQ7K64TV7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MCOFSJK5ZSVABNUPQFQ7K64TV7/action/storage_attestation","attest_author":"https://pith.science/pith/MCOFSJK5ZSVABNUPQFQ7K64TV7/action/author_attestation","sign_citation":"https://pith.science/pith/MCOFSJK5ZSVABNUPQFQ7K64TV7/action/citation_signature","submit_replication":"https://pith.science/pith/MCOFSJK5ZSVABNUPQFQ7K64TV7/action/replication_record"}},"created_at":"2026-07-05T07:40:05.027990+00:00","updated_at":"2026-07-05T07:40:05.027990+00:00"}