{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:6IBFJNAXNW4OUVLS5K5RP5AVE4","short_pith_number":"pith:6IBFJNAX","schema_version":"1.0","canonical_sha256":"f20254b4176db8ea5572eabb17f4152710f369b9eed13edd27067d5e0d52e04a","source":{"kind":"arxiv","id":"2105.09996","version":3},"attestation_state":"computed","paper":{"title":"VLM: Task-agnostic Video-Language Model Pre-training for Video Understanding","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CV","authors_text":"Christoph Feichtenhofer, Florian Metze, Gargi Ghosh, Hu Xu, Luke Zettlemoyer, Masoumeh Aminzadeh, Po-Yao Huang, Prahal Arora","submitted_at":"2021-05-20T19:13:27Z","abstract_excerpt":"We present a simplified, task-agnostic multi-modal pre-training approach that can accept either video or text input, or both for a variety of end tasks. Existing pre-training are task-specific by adopting either a single cross-modal encoder that requires both modalities, limiting their use for retrieval-style end tasks or more complex multitask learning with two unimodal encoders, limiting early cross-modal fusion. We instead introduce new pretraining masking schemes that better mix across modalities (e.g. by forcing masks for text to predict the closest video embeddings) while also maintainin"},"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":"2105.09996","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CV","submitted_at":"2021-05-20T19:13:27Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"b4d60d09d3bb26e68d145b29c94ce906697d4ce8dad927c207fa15ea93081df8","abstract_canon_sha256":"88556f0cff69468657a60a311b66803df95ececb4c371c9448277c8c07b60db6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:19:12.161878Z","signature_b64":"OPk1rvj3RecoW+Ty3BjySxNPW9x/vfl/Spo/gGIo8x74RpkJ9Xm2mf5Rj0TLnXujf1iVYKpqz2lAu5nnH7k9BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f20254b4176db8ea5572eabb17f4152710f369b9eed13edd27067d5e0d52e04a","last_reissued_at":"2026-07-05T03:19:12.161444Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:19:12.161444Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"VLM: Task-agnostic Video-Language Model Pre-training for Video Understanding","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CV","authors_text":"Christoph Feichtenhofer, Florian Metze, Gargi Ghosh, Hu Xu, Luke Zettlemoyer, Masoumeh Aminzadeh, Po-Yao Huang, Prahal Arora","submitted_at":"2021-05-20T19:13:27Z","abstract_excerpt":"We present a simplified, task-agnostic multi-modal pre-training approach that can accept either video or text input, or both for a variety of end tasks. Existing pre-training are task-specific by adopting either a single cross-modal encoder that requires both modalities, limiting their use for retrieval-style end tasks or more complex multitask learning with two unimodal encoders, limiting early cross-modal fusion. We instead introduce new pretraining masking schemes that better mix across modalities (e.g. by forcing masks for text to predict the closest video embeddings) while also maintainin"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2105.09996","kind":"arxiv","version":3},"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/2105.09996/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":"2105.09996","created_at":"2026-07-05T03:19:12.161507+00:00"},{"alias_kind":"arxiv_version","alias_value":"2105.09996v3","created_at":"2026-07-05T03:19:12.161507+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2105.09996","created_at":"2026-07-05T03:19:12.161507+00:00"},{"alias_kind":"pith_short_12","alias_value":"6IBFJNAXNW4O","created_at":"2026-07-05T03:19:12.161507+00:00"},{"alias_kind":"pith_short_16","alias_value":"6IBFJNAXNW4OUVLS","created_at":"2026-07-05T03:19:12.161507+00:00"},{"alias_kind":"pith_short_8","alias_value":"6IBFJNAX","created_at":"2026-07-05T03:19:12.161507+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21607","citing_title":"T-MOR: Learning Motion-Aware Skeleton Representations for Human Action Recognition","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31144","citing_title":"A Modular Vision-Language-Action Robotics Framework for Indoor Environments","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2412.14058","citing_title":"What Matters in Building Vision-Language-Action Models for Generalist Robots","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2204.14198","citing_title":"Flamingo: a Visual Language Model for Few-Shot Learning","ref_index":133,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17473","citing_title":"Dual-Anchoring: Addressing State Drift in Vision-Language Navigation","ref_index":135,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6IBFJNAXNW4OUVLS5K5RP5AVE4","json":"https://pith.science/pith/6IBFJNAXNW4OUVLS5K5RP5AVE4.json","graph_json":"https://pith.science/api/pith-number/6IBFJNAXNW4OUVLS5K5RP5AVE4/graph.json","events_json":"https://pith.science/api/pith-number/6IBFJNAXNW4OUVLS5K5RP5AVE4/events.json","paper":"https://pith.science/paper/6IBFJNAX"},"agent_actions":{"view_html":"https://pith.science/pith/6IBFJNAXNW4OUVLS5K5RP5AVE4","download_json":"https://pith.science/pith/6IBFJNAXNW4OUVLS5K5RP5AVE4.json","view_paper":"https://pith.science/paper/6IBFJNAX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2105.09996&json=true","fetch_graph":"https://pith.science/api/pith-number/6IBFJNAXNW4OUVLS5K5RP5AVE4/graph.json","fetch_events":"https://pith.science/api/pith-number/6IBFJNAXNW4OUVLS5K5RP5AVE4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6IBFJNAXNW4OUVLS5K5RP5AVE4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6IBFJNAXNW4OUVLS5K5RP5AVE4/action/storage_attestation","attest_author":"https://pith.science/pith/6IBFJNAXNW4OUVLS5K5RP5AVE4/action/author_attestation","sign_citation":"https://pith.science/pith/6IBFJNAXNW4OUVLS5K5RP5AVE4/action/citation_signature","submit_replication":"https://pith.science/pith/6IBFJNAXNW4OUVLS5K5RP5AVE4/action/replication_record"}},"created_at":"2026-07-05T03:19:12.161507+00:00","updated_at":"2026-07-05T03:19:12.161507+00:00"}