{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7LEFEJ6IX75C4YXPZVLX2T7UQF","short_pith_number":"pith:7LEFEJ6I","schema_version":"1.0","canonical_sha256":"fac85227c8bffa2e62efcd577d4ff4816e7908174bbddcfc633ba9a80233f7d3","source":{"kind":"arxiv","id":"2506.10128","version":1},"attestation_state":"computed","paper":{"title":"ViCrit: A Verifiable Reinforcement Learning Proxy Task for Visual Perception in VLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Chao Feng, Chung-Ching Lin, Furong Huang, Kevin Lin, Lijuan Wang, Linjie Li, Ming Li, Xiaoyu Liu, Xiyao Wang, Yongyuan Liang, Yuhang Zhou, Zhengyuan Yang, Ziyi Zang","submitted_at":"2025-06-11T19:16:54Z","abstract_excerpt":"Reinforcement learning (RL) has shown great effectiveness for fine-tuning large language models (LLMs) using tasks that are challenging yet easily verifiable, such as math reasoning or code generation. However, extending this success to visual perception in vision-language models (VLMs) has been impeded by the scarcity of vision-centric tasks that are simultaneously challenging and unambiguously verifiable. To this end, we introduce ViCrit (Visual Caption Hallucination Critic), an RL proxy task that trains VLMs to localize a subtle, synthetic visual hallucination injected into paragraphs of hu"},"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":"2506.10128","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-06-11T19:16:54Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"17086dede3ee12f2ce4d0593d162bde12693e05716a623a225b06e108345efd6","abstract_canon_sha256":"f4501609caad5f4383ae48464cdf2cd0d3c526d10843700c63535d4fb37df38a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:20:17.026304Z","signature_b64":"8qB9VcR39uotjDV+9i1SIEYLovwPzr3G7lNFXJhYLoDdhFwEZBOpQLOZCf91ukoqmSjc9IUssoAX3aYL9dvXBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fac85227c8bffa2e62efcd577d4ff4816e7908174bbddcfc633ba9a80233f7d3","last_reissued_at":"2026-07-05T11:20:17.025809Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:20:17.025809Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ViCrit: A Verifiable Reinforcement Learning Proxy Task for Visual Perception in VLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Chao Feng, Chung-Ching Lin, Furong Huang, Kevin Lin, Lijuan Wang, Linjie Li, Ming Li, Xiaoyu Liu, Xiyao Wang, Yongyuan Liang, Yuhang Zhou, Zhengyuan Yang, Ziyi Zang","submitted_at":"2025-06-11T19:16:54Z","abstract_excerpt":"Reinforcement learning (RL) has shown great effectiveness for fine-tuning large language models (LLMs) using tasks that are challenging yet easily verifiable, such as math reasoning or code generation. However, extending this success to visual perception in vision-language models (VLMs) has been impeded by the scarcity of vision-centric tasks that are simultaneously challenging and unambiguously verifiable. To this end, we introduce ViCrit (Visual Caption Hallucination Critic), an RL proxy task that trains VLMs to localize a subtle, synthetic visual hallucination injected into paragraphs of hu"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.10128","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/2506.10128/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":"2506.10128","created_at":"2026-07-05T11:20:17.025867+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.10128v1","created_at":"2026-07-05T11:20:17.025867+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.10128","created_at":"2026-07-05T11:20:17.025867+00:00"},{"alias_kind":"pith_short_12","alias_value":"7LEFEJ6IX75C","created_at":"2026-07-05T11:20:17.025867+00:00"},{"alias_kind":"pith_short_16","alias_value":"7LEFEJ6IX75C4YXP","created_at":"2026-07-05T11:20:17.025867+00:00"},{"alias_kind":"pith_short_8","alias_value":"7LEFEJ6I","created_at":"2026-07-05T11:20:17.025867+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.28023","citing_title":"VCap: Hypergeometric Rewards for Weak-to-Strong Visual Captioning","ref_index":49,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7LEFEJ6IX75C4YXPZVLX2T7UQF","json":"https://pith.science/pith/7LEFEJ6IX75C4YXPZVLX2T7UQF.json","graph_json":"https://pith.science/api/pith-number/7LEFEJ6IX75C4YXPZVLX2T7UQF/graph.json","events_json":"https://pith.science/api/pith-number/7LEFEJ6IX75C4YXPZVLX2T7UQF/events.json","paper":"https://pith.science/paper/7LEFEJ6I"},"agent_actions":{"view_html":"https://pith.science/pith/7LEFEJ6IX75C4YXPZVLX2T7UQF","download_json":"https://pith.science/pith/7LEFEJ6IX75C4YXPZVLX2T7UQF.json","view_paper":"https://pith.science/paper/7LEFEJ6I","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.10128&json=true","fetch_graph":"https://pith.science/api/pith-number/7LEFEJ6IX75C4YXPZVLX2T7UQF/graph.json","fetch_events":"https://pith.science/api/pith-number/7LEFEJ6IX75C4YXPZVLX2T7UQF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7LEFEJ6IX75C4YXPZVLX2T7UQF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7LEFEJ6IX75C4YXPZVLX2T7UQF/action/storage_attestation","attest_author":"https://pith.science/pith/7LEFEJ6IX75C4YXPZVLX2T7UQF/action/author_attestation","sign_citation":"https://pith.science/pith/7LEFEJ6IX75C4YXPZVLX2T7UQF/action/citation_signature","submit_replication":"https://pith.science/pith/7LEFEJ6IX75C4YXPZVLX2T7UQF/action/replication_record"}},"created_at":"2026-07-05T11:20:17.025867+00:00","updated_at":"2026-07-05T11:20:17.025867+00:00"}