{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2023:GNLFWKJ6EAKKF2KU4DKSIFJGP7","short_pith_number":"pith:GNLFWKJ6","canonical_record":{"source":{"id":"2310.00754","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-10-01T18:10:53Z","cross_cats_sorted":["cs.CL","cs.CV"],"title_canon_sha256":"710315a327cf2970c1ef42a0ea943f384d1b99c568a91f16ec64db2926c4ad7e","abstract_canon_sha256":"43b0362344d3c7cb05687a67c322cdeed04d966c35d846ac64a635b188facd47"},"schema_version":"1.0"},"canonical_sha256":"33565b293e2014a2e954e0d52415267fc9491b4f96356c11127a5610940790a1","source":{"kind":"arxiv","id":"2310.00754","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2310.00754","created_at":"2026-05-17T23:38:12Z"},{"alias_kind":"arxiv_version","alias_value":"2310.00754v2","created_at":"2026-05-17T23:38:12Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.00754","created_at":"2026-05-17T23:38:12Z"},{"alias_kind":"pith_short_12","alias_value":"GNLFWKJ6EAKK","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_16","alias_value":"GNLFWKJ6EAKKF2KU","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_8","alias_value":"GNLFWKJ6","created_at":"2026-05-18T12:33:33Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2023:GNLFWKJ6EAKKF2KU4DKSIFJGP7","target":"record","payload":{"canonical_record":{"source":{"id":"2310.00754","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-10-01T18:10:53Z","cross_cats_sorted":["cs.CL","cs.CV"],"title_canon_sha256":"710315a327cf2970c1ef42a0ea943f384d1b99c568a91f16ec64db2926c4ad7e","abstract_canon_sha256":"43b0362344d3c7cb05687a67c322cdeed04d966c35d846ac64a635b188facd47"},"schema_version":"1.0"},"canonical_sha256":"33565b293e2014a2e954e0d52415267fc9491b4f96356c11127a5610940790a1","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:12.768111Z","signature_b64":"jLU4NaKEOqvI2MXE+/bEMwYRKe6HM72Npd+omkxQTn6LzRKqJM2fwAuCzXKmddc+LdjwrMb1Ti+Mxpw/zdAMDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"33565b293e2014a2e954e0d52415267fc9491b4f96356c11127a5610940790a1","last_reissued_at":"2026-05-17T23:38:12.767088Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:12.767088Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2310.00754","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:12Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"fEIAvrqUpAGSXsL3ktmT3VdGKmZGitAPKAo2kw0CTAFYHOC/+uYaSLg8+uYW218WF3KQJUHGHEVjTX7PkurKDQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-03T22:13:25.377909Z"},"content_sha256":"a235d7a84730b0b505c2bdb41a097b7bf448a357f383633e52352caca6f8f514","schema_version":"1.0","event_id":"sha256:a235d7a84730b0b505c2bdb41a097b7bf448a357f383633e52352caca6f8f514"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2023:GNLFWKJ6EAKKF2KU4DKSIFJGP7","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Analyzing and Mitigating Object Hallucination in Large Vision-Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A post-hoc algorithm called LURE reduces object hallucinations in large vision-language models by reconstructing descriptions based on co-occurrence, uncertainty, and text position.","cross_cats":["cs.CL","cs.CV"],"primary_cat":"cs.LG","authors_text":"Chelsea Finn, Chenhang Cui, Huaxiu Yao, Jaehong Yoon, Linjun Zhang, Mohit Bansal, Yiyang Zhou, Zhun Deng","submitted_at":"2023-10-01T18:10:53Z","abstract_excerpt":"Large vision-language models (LVLMs) have shown remarkable abilities in understanding visual information with human languages. However, LVLMs still suffer from object hallucination, which is the problem of generating descriptions that include objects that do not actually exist in the images. This can negatively impact many vision-language tasks, such as visual summarization and reasoning. To address this issue, we propose a simple yet powerful algorithm, LVLM Hallucination Revisor (LURE), to post-hoc rectify object hallucination in LVLMs by reconstructing less hallucinatory descriptions. LURE "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We evaluate LURE on six open-source LVLMs, achieving a 23% improvement in general object hallucination evaluation metrics over the previous best approach.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the identified statistical factors (co-occurrence, uncertainty during decoding, and later text position) are sufficient to guide reliable post-hoc reconstruction of less hallucinatory descriptions across diverse LVLMs and image domains.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"LURE reduces object hallucination in LVLMs by 23% via post-hoc revision informed by co-occurrence, uncertainty, and text position analysis.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A post-hoc algorithm called LURE reduces object hallucinations in large vision-language models by reconstructing descriptions based on co-occurrence, uncertainty, and text position.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"5e613e487c5f3b805438b73816607e3997ab63e0e860e26c71832a40ade1e4ac"},"source":{"id":"2310.00754","kind":"arxiv","version":2},"verdict":{"id":"8a42bbf5-61ea-40cf-a1c7-54e9a25cada0","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T22:42:30.899600Z","strongest_claim":"We evaluate LURE on six open-source LVLMs, achieving a 23% improvement in general object hallucination evaluation metrics over the previous best approach.","one_line_summary":"LURE reduces object hallucination in LVLMs by 23% via post-hoc revision informed by co-occurrence, uncertainty, and text position analysis.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the identified statistical factors (co-occurrence, uncertainty during decoding, and later text position) are sufficient to guide reliable post-hoc reconstruction of less hallucinatory descriptions across diverse LVLMs and image domains.","pith_extraction_headline":"A post-hoc algorithm called LURE reduces object hallucinations in large vision-language models by reconstructing descriptions based on co-occurrence, uncertainty, and text position."},"references":{"count":36,"sample":[{"doi":"","year":2016,"title":"Spice: Semantic propo- sitional image caption evaluation","work_id":"d8bdc1f3-0501-47d7-bf83-5a51413210ef","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1901,"title":"Language models are few-shot learners","work_id":"677093e0-2019-45af-8c52-d9b33dec7e3d","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"PaLM: Scaling Language Modeling with Pathways","work_id":"a94f3ef7-2c49-4445-93fe-6ec16aafd966","ref_index":3,"cited_arxiv_id":"2204.02311","is_internal_anchor":true},{"doi":"","year":2024,"title":"Imagenet: A large-scale hi- erarchical image database","work_id":"92fd5d54-2aba-4ce2-9641-8dea59b064ff","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Beam Search Strategies for Neural Machine Translation","work_id":"00b2d35f-c6ed-4c4d-bbf1-93f0a15d12c9","ref_index":5,"cited_arxiv_id":"1702.01806","is_internal_anchor":true}],"resolved_work":36,"snapshot_sha256":"d4101dddd528ceba37ca9cc9e6fb245cbfb1d907a10fc9f18c50040393f1b52c","internal_anchors":11},"formal_canon":{"evidence_count":2,"snapshot_sha256":"8f43d22835356f2f212bb2c30b8d97890589be87b3664013d23192697ec83327"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"8a42bbf5-61ea-40cf-a1c7-54e9a25cada0"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:12Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"0oKDEbC6oAiua2sWdbFeLD56jLGnxf+cPDS6EE9NWHnP7nLgCiIWVqDghnQQ/7vVVTHL0AqF1wNda78nEeGwDg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-03T22:13:25.378450Z"},"content_sha256":"7f80b4b0db25be98fbc09402922394c0381b98b15ad16712bfee709cd687f244","schema_version":"1.0","event_id":"sha256:7f80b4b0db25be98fbc09402922394c0381b98b15ad16712bfee709cd687f244"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/GNLFWKJ6EAKKF2KU4DKSIFJGP7/bundle.json","state_url":"https://pith.science/pith/GNLFWKJ6EAKKF2KU4DKSIFJGP7/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/GNLFWKJ6EAKKF2KU4DKSIFJGP7/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-03T22:13:25Z","links":{"resolver":"https://pith.science/pith/GNLFWKJ6EAKKF2KU4DKSIFJGP7","bundle":"https://pith.science/pith/GNLFWKJ6EAKKF2KU4DKSIFJGP7/bundle.json","state":"https://pith.science/pith/GNLFWKJ6EAKKF2KU4DKSIFJGP7/state.json","well_known_bundle":"https://pith.science/.well-known/pith/GNLFWKJ6EAKKF2KU4DKSIFJGP7/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2023:GNLFWKJ6EAKKF2KU4DKSIFJGP7","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"43b0362344d3c7cb05687a67c322cdeed04d966c35d846ac64a635b188facd47","cross_cats_sorted":["cs.CL","cs.CV"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-10-01T18:10:53Z","title_canon_sha256":"710315a327cf2970c1ef42a0ea943f384d1b99c568a91f16ec64db2926c4ad7e"},"schema_version":"1.0","source":{"id":"2310.00754","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2310.00754","created_at":"2026-05-17T23:38:12Z"},{"alias_kind":"arxiv_version","alias_value":"2310.00754v2","created_at":"2026-05-17T23:38:12Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.00754","created_at":"2026-05-17T23:38:12Z"},{"alias_kind":"pith_short_12","alias_value":"GNLFWKJ6EAKK","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_16","alias_value":"GNLFWKJ6EAKKF2KU","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_8","alias_value":"GNLFWKJ6","created_at":"2026-05-18T12:33:33Z"}],"graph_snapshots":[{"event_id":"sha256:7f80b4b0db25be98fbc09402922394c0381b98b15ad16712bfee709cd687f244","target":"graph","created_at":"2026-05-17T23:38:12Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"We evaluate LURE on six open-source LVLMs, achieving a 23% improvement in general object hallucination evaluation metrics over the previous best approach."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the identified statistical factors (co-occurrence, uncertainty during decoding, and later text position) are sufficient to guide reliable post-hoc reconstruction of less hallucinatory descriptions across diverse LVLMs and image domains."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"LURE reduces object hallucination in LVLMs by 23% via post-hoc revision informed by co-occurrence, uncertainty, and text position analysis."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"A post-hoc algorithm called LURE reduces object hallucinations in large vision-language models by reconstructing descriptions based on co-occurrence, uncertainty, and text position."}],"snapshot_sha256":"5e613e487c5f3b805438b73816607e3997ab63e0e860e26c71832a40ade1e4ac"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"8f43d22835356f2f212bb2c30b8d97890589be87b3664013d23192697ec83327"},"paper":{"abstract_excerpt":"Large vision-language models (LVLMs) have shown remarkable abilities in understanding visual information with human languages. However, LVLMs still suffer from object hallucination, which is the problem of generating descriptions that include objects that do not actually exist in the images. This can negatively impact many vision-language tasks, such as visual summarization and reasoning. To address this issue, we propose a simple yet powerful algorithm, LVLM Hallucination Revisor (LURE), to post-hoc rectify object hallucination in LVLMs by reconstructing less hallucinatory descriptions. LURE ","authors_text":"Chelsea Finn, Chenhang Cui, Huaxiu Yao, Jaehong Yoon, Linjun Zhang, Mohit Bansal, Yiyang Zhou, Zhun Deng","cross_cats":["cs.CL","cs.CV"],"headline":"A post-hoc algorithm called LURE reduces object hallucinations in large vision-language models by reconstructing descriptions based on co-occurrence, uncertainty, and text position.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-10-01T18:10:53Z","title":"Analyzing and Mitigating Object Hallucination in Large Vision-Language Models"},"references":{"count":36,"internal_anchors":11,"resolved_work":36,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Spice: Semantic propo- sitional image caption evaluation","work_id":"d8bdc1f3-0501-47d7-bf83-5a51413210ef","year":2016},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Language models are few-shot learners","work_id":"677093e0-2019-45af-8c52-d9b33dec7e3d","year":1901},{"cited_arxiv_id":"2204.02311","doi":"","is_internal_anchor":true,"ref_index":3,"title":"PaLM: Scaling Language Modeling with Pathways","work_id":"a94f3ef7-2c49-4445-93fe-6ec16aafd966","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Imagenet: A large-scale hi- erarchical image database","work_id":"92fd5d54-2aba-4ce2-9641-8dea59b064ff","year":2024},{"cited_arxiv_id":"1702.01806","doi":"","is_internal_anchor":true,"ref_index":5,"title":"Beam Search Strategies for Neural Machine Translation","work_id":"00b2d35f-c6ed-4c4d-bbf1-93f0a15d12c9","year":null}],"snapshot_sha256":"d4101dddd528ceba37ca9cc9e6fb245cbfb1d907a10fc9f18c50040393f1b52c"},"source":{"id":"2310.00754","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-17T22:42:30.899600Z","id":"8a42bbf5-61ea-40cf-a1c7-54e9a25cada0","model_set":{"reader":"grok-4.3"},"one_line_summary":"LURE reduces object hallucination in LVLMs by 23% via post-hoc revision informed by co-occurrence, uncertainty, and text position analysis.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"A post-hoc algorithm called LURE reduces object hallucinations in large vision-language models by reconstructing descriptions based on co-occurrence, uncertainty, and text position.","strongest_claim":"We evaluate LURE on six open-source LVLMs, achieving a 23% improvement in general object hallucination evaluation metrics over the previous best approach.","weakest_assumption":"That the identified statistical factors (co-occurrence, uncertainty during decoding, and later text position) are sufficient to guide reliable post-hoc reconstruction of less hallucinatory descriptions across diverse LVLMs and image domains."}},"verdict_id":"8a42bbf5-61ea-40cf-a1c7-54e9a25cada0"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:a235d7a84730b0b505c2bdb41a097b7bf448a357f383633e52352caca6f8f514","target":"record","created_at":"2026-05-17T23:38:12Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"43b0362344d3c7cb05687a67c322cdeed04d966c35d846ac64a635b188facd47","cross_cats_sorted":["cs.CL","cs.CV"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-10-01T18:10:53Z","title_canon_sha256":"710315a327cf2970c1ef42a0ea943f384d1b99c568a91f16ec64db2926c4ad7e"},"schema_version":"1.0","source":{"id":"2310.00754","kind":"arxiv","version":2}},"canonical_sha256":"33565b293e2014a2e954e0d52415267fc9491b4f96356c11127a5610940790a1","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"33565b293e2014a2e954e0d52415267fc9491b4f96356c11127a5610940790a1","first_computed_at":"2026-05-17T23:38:12.767088Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:12.767088Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"jLU4NaKEOqvI2MXE+/bEMwYRKe6HM72Npd+omkxQTn6LzRKqJM2fwAuCzXKmddc+LdjwrMb1Ti+Mxpw/zdAMDw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:12.768111Z","signed_message":"canonical_sha256_bytes"},"source_id":"2310.00754","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:a235d7a84730b0b505c2bdb41a097b7bf448a357f383633e52352caca6f8f514","sha256:7f80b4b0db25be98fbc09402922394c0381b98b15ad16712bfee709cd687f244"],"state_sha256":"958a292b8296a19e428dbe1017cd1d2453b0a4a1b264faa945b9d8345505be1d"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"RLEeLczA65F5Zu+AHuhc5Hcm7K7WCe2xuEOhcuUpv5Aa+FmcvWQ0nVYwm+X4WqiIwdyOZI86k0v7jTVBVdnABg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-03T22:13:25.380835Z","bundle_sha256":"accf79c16be0a683100749472222898e6771f125ffc86092f5b321abe91a2c96"}}