{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:QS225IU3C6UFFQTKIM5KQRQMSM","short_pith_number":"pith:QS225IU3","schema_version":"1.0","canonical_sha256":"84b5aea29b17a852c26a433aa8460c9312340f494905e73dd4713e82a1b894c0","source":{"kind":"arxiv","id":"2507.07990","version":1},"attestation_state":"computed","paper":{"title":"Multi-Granular Spatio-Temporal Token Merging for Training-Free Acceleration of Video LLMs","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Dongyoon Wee, Inwoong Lee, Jeongseok Hyun, Joon-Young Lee, Minho Shim, Seon Joo Kim, Su Ho Han, Sukjun Hwang, Taeoh Kim","submitted_at":"2025-07-10T17:59:02Z","abstract_excerpt":"Video large language models (LLMs) achieve strong video understanding by leveraging a large number of spatio-temporal tokens, but suffer from quadratic computational scaling with token count. To address this, we propose a training-free spatio-temporal token merging method, named STTM. Our key insight is to exploit local spatial and temporal redundancy in video data which has been overlooked in prior work. STTM first transforms each frame into multi-granular spatial tokens using a coarse-to-fine search over a quadtree structure, then performs directed pairwise merging across the temporal dimens"},"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":"2507.07990","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2025-07-10T17:59:02Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"e4af9d9c2ec174483634602892f977caee50ec6ca672930ff1f15d8faf323873","abstract_canon_sha256":"ee9838ea1e4735dbaa27885d81053e6a840488f270752992ca11ae50d4a87004"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:35:08.213798Z","signature_b64":"3u8S+Tu68a4vQsqeYef36qcAUNUUA8OPe9gQ7DQeIxJy0KaQWQngSUaDSD2RAv+ONVOklgT4NR8xwAblppLsDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"84b5aea29b17a852c26a433aa8460c9312340f494905e73dd4713e82a1b894c0","last_reissued_at":"2026-07-05T11:35:08.213247Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:35:08.213247Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Multi-Granular Spatio-Temporal Token Merging for Training-Free Acceleration of Video LLMs","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Dongyoon Wee, Inwoong Lee, Jeongseok Hyun, Joon-Young Lee, Minho Shim, Seon Joo Kim, Su Ho Han, Sukjun Hwang, Taeoh Kim","submitted_at":"2025-07-10T17:59:02Z","abstract_excerpt":"Video large language models (LLMs) achieve strong video understanding by leveraging a large number of spatio-temporal tokens, but suffer from quadratic computational scaling with token count. To address this, we propose a training-free spatio-temporal token merging method, named STTM. Our key insight is to exploit local spatial and temporal redundancy in video data which has been overlooked in prior work. STTM first transforms each frame into multi-granular spatial tokens using a coarse-to-fine search over a quadtree structure, then performs directed pairwise merging across the temporal dimens"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.07990","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/2507.07990/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":"2507.07990","created_at":"2026-07-05T11:35:08.213306+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.07990v1","created_at":"2026-07-05T11:35:08.213306+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.07990","created_at":"2026-07-05T11:35:08.213306+00:00"},{"alias_kind":"pith_short_12","alias_value":"QS225IU3C6UF","created_at":"2026-07-05T11:35:08.213306+00:00"},{"alias_kind":"pith_short_16","alias_value":"QS225IU3C6UFFQTK","created_at":"2026-07-05T11:35:08.213306+00:00"},{"alias_kind":"pith_short_8","alias_value":"QS225IU3","created_at":"2026-07-05T11:35:08.213306+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.19322","citing_title":"DynaTok: Temporally Adaptive and Positional Bias-Aware Token Compression for Video-LLMs","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2509.08016","citing_title":"Video Parallel Scaling: Aggregating Diverse Frame Subsets for VideoLLMs","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03351","citing_title":"VLMaxxing through FrameMogging Training-Free Anti-Recomputation for Video Vision-Language Models","ref_index":34,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/QS225IU3C6UFFQTKIM5KQRQMSM","json":"https://pith.science/pith/QS225IU3C6UFFQTKIM5KQRQMSM.json","graph_json":"https://pith.science/api/pith-number/QS225IU3C6UFFQTKIM5KQRQMSM/graph.json","events_json":"https://pith.science/api/pith-number/QS225IU3C6UFFQTKIM5KQRQMSM/events.json","paper":"https://pith.science/paper/QS225IU3"},"agent_actions":{"view_html":"https://pith.science/pith/QS225IU3C6UFFQTKIM5KQRQMSM","download_json":"https://pith.science/pith/QS225IU3C6UFFQTKIM5KQRQMSM.json","view_paper":"https://pith.science/paper/QS225IU3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.07990&json=true","fetch_graph":"https://pith.science/api/pith-number/QS225IU3C6UFFQTKIM5KQRQMSM/graph.json","fetch_events":"https://pith.science/api/pith-number/QS225IU3C6UFFQTKIM5KQRQMSM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QS225IU3C6UFFQTKIM5KQRQMSM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QS225IU3C6UFFQTKIM5KQRQMSM/action/storage_attestation","attest_author":"https://pith.science/pith/QS225IU3C6UFFQTKIM5KQRQMSM/action/author_attestation","sign_citation":"https://pith.science/pith/QS225IU3C6UFFQTKIM5KQRQMSM/action/citation_signature","submit_replication":"https://pith.science/pith/QS225IU3C6UFFQTKIM5KQRQMSM/action/replication_record"}},"created_at":"2026-07-05T11:35:08.213306+00:00","updated_at":"2026-07-05T11:35:08.213306+00:00"}