{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:ONEPLWCXTGBZ4WTBXPZO645SBB","short_pith_number":"pith:ONEPLWCX","schema_version":"1.0","canonical_sha256":"7348f5d85799839e5a61bbf2ef73b2084a89e91cf2618ecbc883947dc8d73624","source":{"kind":"arxiv","id":"2606.12412","version":1},"attestation_state":"computed","paper":{"title":"Reroute, Don't Remove: Recoverable Visual Token Routing for Vision-Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Cheng-Yu Yang, Shao-Yuan Lo, Yu-Lun Liu","submitted_at":"2026-06-10T17:59:57Z","abstract_excerpt":"Vision-language models (VLMs) project images into hundreds to thousands of visual tokens, making decoder inference expensive in both attention computation and KV-cache memory. Existing visual-token reduction methods largely follow a rank-and-remove paradigm: they score visual tokens, keep a compact subset, and permanently discard the rest. We show that this irreversible action is fragile because visual-token importance changes across decoder depth; tokens ranked low at one stage may become relevant in later layers, especially for grounding-sensitive queries. We propose Reroute, a training-free"},"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":"2606.12412","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-06-10T17:59:57Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"903b1a4a52c0897260729253341050bce03218789e141d71dcfbe5a06a790981","abstract_canon_sha256":"cceb8c25cd969bfc8457762221eb3dd55bc73c674940143e4b74f2747625e62a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-11T02:09:50.189753Z","signature_b64":"QezXn4+8yz89TYI/VbkNmvAKc8aqJ/zolrAnMHec5v4x+wcI6sjf9/hCX+y59Ap1Cyk5Z7v3qpGJH9eS5mFrDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7348f5d85799839e5a61bbf2ef73b2084a89e91cf2618ecbc883947dc8d73624","last_reissued_at":"2026-06-11T02:09:50.189326Z","signature_status":"signed_v1","first_computed_at":"2026-06-11T02:09:50.189326Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Reroute, Don't Remove: Recoverable Visual Token Routing for Vision-Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Cheng-Yu Yang, Shao-Yuan Lo, Yu-Lun Liu","submitted_at":"2026-06-10T17:59:57Z","abstract_excerpt":"Vision-language models (VLMs) project images into hundreds to thousands of visual tokens, making decoder inference expensive in both attention computation and KV-cache memory. Existing visual-token reduction methods largely follow a rank-and-remove paradigm: they score visual tokens, keep a compact subset, and permanently discard the rest. We show that this irreversible action is fragile because visual-token importance changes across decoder depth; tokens ranked low at one stage may become relevant in later layers, especially for grounding-sensitive queries. We propose Reroute, a training-free"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.12412","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/2606.12412/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":"2606.12412","created_at":"2026-06-11T02:09:50.189389+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.12412v1","created_at":"2026-06-11T02:09:50.189389+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.12412","created_at":"2026-06-11T02:09:50.189389+00:00"},{"alias_kind":"pith_short_12","alias_value":"ONEPLWCXTGBZ","created_at":"2026-06-11T02:09:50.189389+00:00"},{"alias_kind":"pith_short_16","alias_value":"ONEPLWCXTGBZ4WTB","created_at":"2026-06-11T02:09:50.189389+00:00"},{"alias_kind":"pith_short_8","alias_value":"ONEPLWCX","created_at":"2026-06-11T02:09:50.189389+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ONEPLWCXTGBZ4WTBXPZO645SBB","json":"https://pith.science/pith/ONEPLWCXTGBZ4WTBXPZO645SBB.json","graph_json":"https://pith.science/api/pith-number/ONEPLWCXTGBZ4WTBXPZO645SBB/graph.json","events_json":"https://pith.science/api/pith-number/ONEPLWCXTGBZ4WTBXPZO645SBB/events.json","paper":"https://pith.science/paper/ONEPLWCX"},"agent_actions":{"view_html":"https://pith.science/pith/ONEPLWCXTGBZ4WTBXPZO645SBB","download_json":"https://pith.science/pith/ONEPLWCXTGBZ4WTBXPZO645SBB.json","view_paper":"https://pith.science/paper/ONEPLWCX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.12412&json=true","fetch_graph":"https://pith.science/api/pith-number/ONEPLWCXTGBZ4WTBXPZO645SBB/graph.json","fetch_events":"https://pith.science/api/pith-number/ONEPLWCXTGBZ4WTBXPZO645SBB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ONEPLWCXTGBZ4WTBXPZO645SBB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ONEPLWCXTGBZ4WTBXPZO645SBB/action/storage_attestation","attest_author":"https://pith.science/pith/ONEPLWCXTGBZ4WTBXPZO645SBB/action/author_attestation","sign_citation":"https://pith.science/pith/ONEPLWCXTGBZ4WTBXPZO645SBB/action/citation_signature","submit_replication":"https://pith.science/pith/ONEPLWCXTGBZ4WTBXPZO645SBB/action/replication_record"}},"created_at":"2026-06-11T02:09:50.189389+00:00","updated_at":"2026-06-11T02:09:50.189389+00:00"}