{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:4FS3OBEF5KLZ7MWH4UUCF2I7RE","short_pith_number":"pith:4FS3OBEF","schema_version":"1.0","canonical_sha256":"e165b70485ea979fb2c7e52822e91f8938c94250d5be6e485222183fe667c338","source":{"kind":"arxiv","id":"2406.08478","version":2},"attestation_state":"computed","paper":{"title":"What If We Recaption Billions of Web Images with LLaMA-3?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CV","authors_text":"Bingchen Zhao, Cihang Xie, Haoqin Tu, Huangjie Zheng, Jieru Mei, Junfei Xiao, Mude Hui, Qing Liu, Sucheng Ren, Xianhang Li, Yuyin Zhou, Zeyu Wang","submitted_at":"2024-06-12T17:59:07Z","abstract_excerpt":"Web-crawled image-text pairs are inherently noisy. Prior studies demonstrate that semantically aligning and enriching textual descriptions of these pairs can significantly enhance model training across various vision-language tasks, particularly text-to-image generation. However, large-scale investigations in this area remain predominantly closed-source. Our paper aims to bridge this community effort, leveraging the powerful and \\textit{open-sourced} LLaMA-3, a GPT-4 level LLM. Our recaptioning pipeline is simple: first, we fine-tune a LLaMA-3-8B powered LLaVA-1.5 and then employ it to recapti"},"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":"2406.08478","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2024-06-12T17:59:07Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"74f64b00c764c0fd6bc67d18f8b1429ebb9e781dac57f0fd6dc49113890a9709","abstract_canon_sha256":"a5a1164597c656ba7d614c773e2a8513dd5a9c0a49f4da52b918cde7ed1a7fcd"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:33:40.901907Z","signature_b64":"66bFpF1c6W8skz3/k2616RIynmkuhrBIED1BldLKYaeu2TOFYJitpM9P+Ibs3D1Mep7Xqg3kc1YiLIurZiFHCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e165b70485ea979fb2c7e52822e91f8938c94250d5be6e485222183fe667c338","last_reissued_at":"2026-07-05T08:33:40.901468Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:33:40.901468Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"What If We Recaption Billions of Web Images with LLaMA-3?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CV","authors_text":"Bingchen Zhao, Cihang Xie, Haoqin Tu, Huangjie Zheng, Jieru Mei, Junfei Xiao, Mude Hui, Qing Liu, Sucheng Ren, Xianhang Li, Yuyin Zhou, Zeyu Wang","submitted_at":"2024-06-12T17:59:07Z","abstract_excerpt":"Web-crawled image-text pairs are inherently noisy. Prior studies demonstrate that semantically aligning and enriching textual descriptions of these pairs can significantly enhance model training across various vision-language tasks, particularly text-to-image generation. However, large-scale investigations in this area remain predominantly closed-source. Our paper aims to bridge this community effort, leveraging the powerful and \\textit{open-sourced} LLaMA-3, a GPT-4 level LLM. Our recaptioning pipeline is simple: first, we fine-tune a LLaMA-3-8B powered LLaVA-1.5 and then employ it to recapti"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.08478","kind":"arxiv","version":2},"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/2406.08478/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":"2406.08478","created_at":"2026-07-05T08:33:40.901522+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.08478v2","created_at":"2026-07-05T08:33:40.901522+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.08478","created_at":"2026-07-05T08:33:40.901522+00:00"},{"alias_kind":"pith_short_12","alias_value":"4FS3OBEF5KLZ","created_at":"2026-07-05T08:33:40.901522+00:00"},{"alias_kind":"pith_short_16","alias_value":"4FS3OBEF5KLZ7MWH","created_at":"2026-07-05T08:33:40.901522+00:00"},{"alias_kind":"pith_short_8","alias_value":"4FS3OBEF","created_at":"2026-07-05T08:33:40.901522+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26794","citing_title":"ReasonCLIP-58M: Visually Grounded Commonsense Reasoning Supervision for CLIP","ref_index":53,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28551","citing_title":"DataComp-VLM: Improved Open Datasets for Vision-Language Models","ref_index":163,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28551","citing_title":"DataComp-VLM: Improved Open Datasets for Vision-Language Models","ref_index":163,"is_internal_anchor":false},{"citing_arxiv_id":"2506.18871","citing_title":"OmniGen2: Towards Instruction-Aligned Multimodal Generation","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2512.22437","citing_title":"EmoCtrl: Controllable Emotional Image Content Generation","ref_index":16,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4FS3OBEF5KLZ7MWH4UUCF2I7RE","json":"https://pith.science/pith/4FS3OBEF5KLZ7MWH4UUCF2I7RE.json","graph_json":"https://pith.science/api/pith-number/4FS3OBEF5KLZ7MWH4UUCF2I7RE/graph.json","events_json":"https://pith.science/api/pith-number/4FS3OBEF5KLZ7MWH4UUCF2I7RE/events.json","paper":"https://pith.science/paper/4FS3OBEF"},"agent_actions":{"view_html":"https://pith.science/pith/4FS3OBEF5KLZ7MWH4UUCF2I7RE","download_json":"https://pith.science/pith/4FS3OBEF5KLZ7MWH4UUCF2I7RE.json","view_paper":"https://pith.science/paper/4FS3OBEF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.08478&json=true","fetch_graph":"https://pith.science/api/pith-number/4FS3OBEF5KLZ7MWH4UUCF2I7RE/graph.json","fetch_events":"https://pith.science/api/pith-number/4FS3OBEF5KLZ7MWH4UUCF2I7RE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4FS3OBEF5KLZ7MWH4UUCF2I7RE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4FS3OBEF5KLZ7MWH4UUCF2I7RE/action/storage_attestation","attest_author":"https://pith.science/pith/4FS3OBEF5KLZ7MWH4UUCF2I7RE/action/author_attestation","sign_citation":"https://pith.science/pith/4FS3OBEF5KLZ7MWH4UUCF2I7RE/action/citation_signature","submit_replication":"https://pith.science/pith/4FS3OBEF5KLZ7MWH4UUCF2I7RE/action/replication_record"}},"created_at":"2026-07-05T08:33:40.901522+00:00","updated_at":"2026-07-05T08:33:40.901522+00:00"}