{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:JCMH23DTKT5O4YY6DFUOJOW526","short_pith_number":"pith:JCMH23DT","schema_version":"1.0","canonical_sha256":"48987d6c7354faee631e1968e4baddd7a541a5ae3bdbf944febef51fa9d750ee","source":{"kind":"arxiv","id":"2410.16267","version":2},"attestation_state":"computed","paper":{"title":"xGen-MM-Vid (BLIP-3-Video): You Only Need 32 Tokens to Represent a Video Even in VLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.LG"],"primary_cat":"cs.CV","authors_text":"Caiming Xiong, Can Qin, Honglu Zhou, Jongwoo Park, Juan Carlos Niebles, Kanchana Ranasinghe, Le Xue, Manli Shu, Michael S. Ryoo, Ran Xu, Shrikant Kendre, Silvio Savarese","submitted_at":"2024-10-21T17:59:11Z","abstract_excerpt":"We present xGen-MM-Vid (BLIP-3-Video): a multimodal language model for videos, particularly designed to efficiently capture temporal information over multiple frames. BLIP-3-Video takes advantage of the 'temporal encoder' in addition to the conventional visual tokenizer, which maps a sequence of tokens over multiple frames into a compact set of visual tokens. This enables BLIP3-Video to use much fewer visual tokens than its competing models (e.g., 32 vs. 4608 tokens). We explore different types of temporal encoders, including learnable spatio-temporal pooling as well as sequential models like "},"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":"2410.16267","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2024-10-21T17:59:11Z","cross_cats_sorted":["cs.AI","cs.CL","cs.LG"],"title_canon_sha256":"0b34f367eb156eb813dfd086ff02df3e73d28fc74ed5b0e48ac25bff9030979c","abstract_canon_sha256":"6b30ac4ab06f9adfd44ecf4e630541d932dd4301e8cc2b92a9a1252fbd2a86ee"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:18:39.888123Z","signature_b64":"Ba8bO++IwdHl9uc9fVuM8SP8TRA8H7FdHHl00+DwaFuuPc6ypqdrNL6qeGVr6Oe6Jfq8I9nbNhsCfgzfu+4xDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"48987d6c7354faee631e1968e4baddd7a541a5ae3bdbf944febef51fa9d750ee","last_reissued_at":"2026-07-05T11:18:39.887633Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:18:39.887633Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"xGen-MM-Vid (BLIP-3-Video): You Only Need 32 Tokens to Represent a Video Even in VLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.LG"],"primary_cat":"cs.CV","authors_text":"Caiming Xiong, Can Qin, Honglu Zhou, Jongwoo Park, Juan Carlos Niebles, Kanchana Ranasinghe, Le Xue, Manli Shu, Michael S. Ryoo, Ran Xu, Shrikant Kendre, Silvio Savarese","submitted_at":"2024-10-21T17:59:11Z","abstract_excerpt":"We present xGen-MM-Vid (BLIP-3-Video): a multimodal language model for videos, particularly designed to efficiently capture temporal information over multiple frames. BLIP-3-Video takes advantage of the 'temporal encoder' in addition to the conventional visual tokenizer, which maps a sequence of tokens over multiple frames into a compact set of visual tokens. This enables BLIP3-Video to use much fewer visual tokens than its competing models (e.g., 32 vs. 4608 tokens). We explore different types of temporal encoders, including learnable spatio-temporal pooling as well as sequential models like "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.16267","kind":"arxiv","version":2},"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/2410.16267/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":"2410.16267","created_at":"2026-07-05T11:18:39.887690+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.16267v2","created_at":"2026-07-05T11:18:39.887690+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.16267","created_at":"2026-07-05T11:18:39.887690+00:00"},{"alias_kind":"pith_short_12","alias_value":"JCMH23DTKT5O","created_at":"2026-07-05T11:18:39.887690+00:00"},{"alias_kind":"pith_short_16","alias_value":"JCMH23DTKT5O4YY6","created_at":"2026-07-05T11:18:39.887690+00:00"},{"alias_kind":"pith_short_8","alias_value":"JCMH23DT","created_at":"2026-07-05T11:18:39.887690+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.09904","citing_title":"TOC-Bench: A Temporal Object Consistency Benchmark for Video Large Language Models","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09904","citing_title":"TOC-Bench: A Temporal Object Consistency Benchmark for Video Large Language Models","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07568","citing_title":"Tracing the Arrow of Time: Diagnosing Temporal Information Flow in Video-LLMs","ref_index":39,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JCMH23DTKT5O4YY6DFUOJOW526","json":"https://pith.science/pith/JCMH23DTKT5O4YY6DFUOJOW526.json","graph_json":"https://pith.science/api/pith-number/JCMH23DTKT5O4YY6DFUOJOW526/graph.json","events_json":"https://pith.science/api/pith-number/JCMH23DTKT5O4YY6DFUOJOW526/events.json","paper":"https://pith.science/paper/JCMH23DT"},"agent_actions":{"view_html":"https://pith.science/pith/JCMH23DTKT5O4YY6DFUOJOW526","download_json":"https://pith.science/pith/JCMH23DTKT5O4YY6DFUOJOW526.json","view_paper":"https://pith.science/paper/JCMH23DT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.16267&json=true","fetch_graph":"https://pith.science/api/pith-number/JCMH23DTKT5O4YY6DFUOJOW526/graph.json","fetch_events":"https://pith.science/api/pith-number/JCMH23DTKT5O4YY6DFUOJOW526/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JCMH23DTKT5O4YY6DFUOJOW526/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JCMH23DTKT5O4YY6DFUOJOW526/action/storage_attestation","attest_author":"https://pith.science/pith/JCMH23DTKT5O4YY6DFUOJOW526/action/author_attestation","sign_citation":"https://pith.science/pith/JCMH23DTKT5O4YY6DFUOJOW526/action/citation_signature","submit_replication":"https://pith.science/pith/JCMH23DTKT5O4YY6DFUOJOW526/action/replication_record"}},"created_at":"2026-07-05T11:18:39.887690+00:00","updated_at":"2026-07-05T11:18:39.887690+00:00"}