{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:WLE4AD3S3JKLXMDBE5U7RVJDZV","short_pith_number":"pith:WLE4AD3S","schema_version":"1.0","canonical_sha256":"b2c9c00f72da54bbb0612769f8d523cd7ce46b2306524b7ff8ebcb89f282dab9","source":{"kind":"arxiv","id":"2504.10020","version":4},"attestation_state":"computed","paper":{"title":"The Mirage of Performance Gains: Why Contrastive Decoding Fails to Mitigate Object Hallucinations in MLLMs?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.CL","authors_text":"Guangzong Si, Hao Yin, Zilei Wang","submitted_at":"2025-04-14T09:25:37Z","abstract_excerpt":"Contrastive decoding strategies are widely used to reduce object hallucinations in multimodal large language models (MLLMs). These methods work by constructing contrastive samples to induce hallucinations and then suppressing them in the output distribution. However, this paper demonstrates that such approaches fail to effectively mitigate the hallucination problem. The performance improvements observed on POPE Benchmark are largely driven by two misleading factors: (1) crude, unidirectional adjustments to the model's output distribution and (2) the adaptive plausibility constraint, which redu"},"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":"2504.10020","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-04-14T09:25:37Z","cross_cats_sorted":["cs.AI","cs.CV"],"title_canon_sha256":"53dbb9ed203a8d3e42158d20d3e00f3ea7e11dd0379cb2dbfcb8f90a33e9863e","abstract_canon_sha256":"681fc74fcb701ad9ace0320a4f1d2271ed2fa56b735b702f73e59302da6dc204"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-05T01:15:13.285473Z","signature_b64":"Ah3/uFS2nL+x3Y72JsrOWdr9teWgdEzqAXE+7q2DYDJzH0wbhvMEzAzm/QEgwgI+X+57j8YCluHA0o4jVfH7Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b2c9c00f72da54bbb0612769f8d523cd7ce46b2306524b7ff8ebcb89f282dab9","last_reissued_at":"2026-06-05T01:15:13.284559Z","signature_status":"signed_v1","first_computed_at":"2026-06-05T01:15:13.284559Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Mirage of Performance Gains: Why Contrastive Decoding Fails to Mitigate Object Hallucinations in MLLMs?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.CL","authors_text":"Guangzong Si, Hao Yin, Zilei Wang","submitted_at":"2025-04-14T09:25:37Z","abstract_excerpt":"Contrastive decoding strategies are widely used to reduce object hallucinations in multimodal large language models (MLLMs). These methods work by constructing contrastive samples to induce hallucinations and then suppressing them in the output distribution. However, this paper demonstrates that such approaches fail to effectively mitigate the hallucination problem. The performance improvements observed on POPE Benchmark are largely driven by two misleading factors: (1) crude, unidirectional adjustments to the model's output distribution and (2) the adaptive plausibility constraint, which redu"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.10020","kind":"arxiv","version":4},"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/2504.10020/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":"2504.10020","created_at":"2026-06-05T01:15:13.284675+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.10020v4","created_at":"2026-06-05T01:15:13.284675+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.10020","created_at":"2026-06-05T01:15:13.284675+00:00"},{"alias_kind":"pith_short_12","alias_value":"WLE4AD3S3JKL","created_at":"2026-06-05T01:15:13.284675+00:00"},{"alias_kind":"pith_short_16","alias_value":"WLE4AD3S3JKLXMDB","created_at":"2026-06-05T01:15:13.284675+00:00"},{"alias_kind":"pith_short_8","alias_value":"WLE4AD3S","created_at":"2026-06-05T01:15:13.284675+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2604.14363","citing_title":"The Cost of Language: Centroid Erasure Exposes and Exploits Modal Competition in Multimodal Language Models","ref_index":22,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WLE4AD3S3JKLXMDBE5U7RVJDZV","json":"https://pith.science/pith/WLE4AD3S3JKLXMDBE5U7RVJDZV.json","graph_json":"https://pith.science/api/pith-number/WLE4AD3S3JKLXMDBE5U7RVJDZV/graph.json","events_json":"https://pith.science/api/pith-number/WLE4AD3S3JKLXMDBE5U7RVJDZV/events.json","paper":"https://pith.science/paper/WLE4AD3S"},"agent_actions":{"view_html":"https://pith.science/pith/WLE4AD3S3JKLXMDBE5U7RVJDZV","download_json":"https://pith.science/pith/WLE4AD3S3JKLXMDBE5U7RVJDZV.json","view_paper":"https://pith.science/paper/WLE4AD3S","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.10020&json=true","fetch_graph":"https://pith.science/api/pith-number/WLE4AD3S3JKLXMDBE5U7RVJDZV/graph.json","fetch_events":"https://pith.science/api/pith-number/WLE4AD3S3JKLXMDBE5U7RVJDZV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WLE4AD3S3JKLXMDBE5U7RVJDZV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WLE4AD3S3JKLXMDBE5U7RVJDZV/action/storage_attestation","attest_author":"https://pith.science/pith/WLE4AD3S3JKLXMDBE5U7RVJDZV/action/author_attestation","sign_citation":"https://pith.science/pith/WLE4AD3S3JKLXMDBE5U7RVJDZV/action/citation_signature","submit_replication":"https://pith.science/pith/WLE4AD3S3JKLXMDBE5U7RVJDZV/action/replication_record"}},"created_at":"2026-06-05T01:15:13.284675+00:00","updated_at":"2026-06-05T01:15:13.284675+00:00"}