{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:WKEPV2Y7VQYCAVACXDMCVWY3BO","short_pith_number":"pith:WKEPV2Y7","schema_version":"1.0","canonical_sha256":"b288faeb1fac30205402b8d82adb1b0b8d1bac4ef57e28c4364c752df8a739d2","source":{"kind":"arxiv","id":"2505.21547","version":1},"attestation_state":"computed","paper":{"title":"Image Tokens Matter: Mitigating Hallucination in Discrete Tokenizer-based Large Vision-Language Models via Latent Editing","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Christoph Meinel, Gerard de Melo, Haojin Yang, Jindong Gu, Rui Cao, Weixing Wang, Zifeng Ding","submitted_at":"2025-05-24T22:36:15Z","abstract_excerpt":"Large Vision-Language Models (LVLMs) with discrete image tokenizers unify multimodal representations by encoding visual inputs into a finite set of tokens. Despite their effectiveness, we find that these models still hallucinate non-existent objects. We hypothesize that this may be due to visual priors induced during training: When certain image tokens frequently co-occur in the same spatial regions and represent shared objects, they become strongly associated with the verbalizations of those objects. As a result, the model may hallucinate by evoking visually absent tokens that often co-occur "},"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":"2505.21547","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-05-24T22:36:15Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"949bc0bf7c32a362508ec0bb7bff0723a2a8717ce80999856b484ed618fa5b59","abstract_canon_sha256":"49d98e28fcb0816d6ca006cb58bb0a1a3af5c1ae2fff4f7c2e7f8cc55163cd2e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:10:59.721711Z","signature_b64":"EoLpdZOg0c6ZG/mZn3P+BsPtDodIVn6A5rR96bdS08rM54mHV9gasj08g3XPYHEAZUPKN3h6vWMpC/cMoXh/Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b288faeb1fac30205402b8d82adb1b0b8d1bac4ef57e28c4364c752df8a739d2","last_reissued_at":"2026-07-05T11:10:59.721245Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:10:59.721245Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Image Tokens Matter: Mitigating Hallucination in Discrete Tokenizer-based Large Vision-Language Models via Latent Editing","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Christoph Meinel, Gerard de Melo, Haojin Yang, Jindong Gu, Rui Cao, Weixing Wang, Zifeng Ding","submitted_at":"2025-05-24T22:36:15Z","abstract_excerpt":"Large Vision-Language Models (LVLMs) with discrete image tokenizers unify multimodal representations by encoding visual inputs into a finite set of tokens. Despite their effectiveness, we find that these models still hallucinate non-existent objects. We hypothesize that this may be due to visual priors induced during training: When certain image tokens frequently co-occur in the same spatial regions and represent shared objects, they become strongly associated with the verbalizations of those objects. As a result, the model may hallucinate by evoking visually absent tokens that often co-occur "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.21547","kind":"arxiv","version":1},"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/2505.21547/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":"2505.21547","created_at":"2026-07-05T11:10:59.721300+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.21547v1","created_at":"2026-07-05T11:10:59.721300+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.21547","created_at":"2026-07-05T11:10:59.721300+00:00"},{"alias_kind":"pith_short_12","alias_value":"WKEPV2Y7VQYC","created_at":"2026-07-05T11:10:59.721300+00:00"},{"alias_kind":"pith_short_16","alias_value":"WKEPV2Y7VQYCAVAC","created_at":"2026-07-05T11:10:59.721300+00:00"},{"alias_kind":"pith_short_8","alias_value":"WKEPV2Y7","created_at":"2026-07-05T11:10:59.721300+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.03893","citing_title":"FeynmanBench: Benchmarking Multimodal LLMs on Diagrammatic Physics Reasoning","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02735","citing_title":"Visual Latents Know More Than They Say: Unsilencing Latent Reasoning in MLLMs","ref_index":32,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WKEPV2Y7VQYCAVACXDMCVWY3BO","json":"https://pith.science/pith/WKEPV2Y7VQYCAVACXDMCVWY3BO.json","graph_json":"https://pith.science/api/pith-number/WKEPV2Y7VQYCAVACXDMCVWY3BO/graph.json","events_json":"https://pith.science/api/pith-number/WKEPV2Y7VQYCAVACXDMCVWY3BO/events.json","paper":"https://pith.science/paper/WKEPV2Y7"},"agent_actions":{"view_html":"https://pith.science/pith/WKEPV2Y7VQYCAVACXDMCVWY3BO","download_json":"https://pith.science/pith/WKEPV2Y7VQYCAVACXDMCVWY3BO.json","view_paper":"https://pith.science/paper/WKEPV2Y7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.21547&json=true","fetch_graph":"https://pith.science/api/pith-number/WKEPV2Y7VQYCAVACXDMCVWY3BO/graph.json","fetch_events":"https://pith.science/api/pith-number/WKEPV2Y7VQYCAVACXDMCVWY3BO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WKEPV2Y7VQYCAVACXDMCVWY3BO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WKEPV2Y7VQYCAVACXDMCVWY3BO/action/storage_attestation","attest_author":"https://pith.science/pith/WKEPV2Y7VQYCAVACXDMCVWY3BO/action/author_attestation","sign_citation":"https://pith.science/pith/WKEPV2Y7VQYCAVACXDMCVWY3BO/action/citation_signature","submit_replication":"https://pith.science/pith/WKEPV2Y7VQYCAVACXDMCVWY3BO/action/replication_record"}},"created_at":"2026-07-05T11:10:59.721300+00:00","updated_at":"2026-07-05T11:10:59.721300+00:00"}