{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:2KLVNDVXPBNW24AC7B2QZFXOVV","short_pith_number":"pith:2KLVNDVX","schema_version":"1.0","canonical_sha256":"d297568eb7785b6d7002f8750c96eead684adcb51fc675616e138bdf224c4ccc","source":{"kind":"arxiv","id":"2603.14482","version":3},"attestation_state":"computed","paper":{"title":"V-JEPA 2.1: Unlocking Dense Features in Video Self-Supervised Learning","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Adrien Bardes, Amir Bar, Koustuv Sinha, Lorenzo Mur-Labadia, Matthew Muckley, Mido Assran, Mike Rabbat, Nicolas Ballas, Yann LeCun","submitted_at":"2026-03-15T17:02:40Z","abstract_excerpt":"We present V-JEPA 2.1, a family of self-supervised models that learn dense, high-quality visual representations for both images and videos while retaining strong global scene understanding. The approach combines four key components. First, a dense predictive loss uses a masking-based objective in which both visible and masked tokens contribute to the training signal, encouraging explicit spatial and temporal grounding. Second, deep self-supervision applies the self-supervised objective hierarchically across multiple intermediate encoder layers to improve representation quality. Third, multi-mo"},"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":"2603.14482","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CV","submitted_at":"2026-03-15T17:02:40Z","cross_cats_sorted":[],"title_canon_sha256":"a1f0099e712ff1a613b8ee811befcb33877f2de80909a59b234dbf96b97b7bf6","abstract_canon_sha256":"4d18dd647d91aebf279d0879623ad54998fee9d49339c508f776fd2ee6036a8f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-12T01:09:25.038360Z","signature_b64":"q2KJvNOJtsheMpgIElq9zoxnXvIydPcdNkqrSP/us5ZGuhn/NSMS506nvf/a/dyGTGVrok4iAJyUzIMICC2mBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d297568eb7785b6d7002f8750c96eead684adcb51fc675616e138bdf224c4ccc","last_reissued_at":"2026-06-12T01:09:25.037705Z","signature_status":"signed_v1","first_computed_at":"2026-06-12T01:09:25.037705Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"V-JEPA 2.1: Unlocking Dense Features in Video Self-Supervised Learning","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Adrien Bardes, Amir Bar, Koustuv Sinha, Lorenzo Mur-Labadia, Matthew Muckley, Mido Assran, Mike Rabbat, Nicolas Ballas, Yann LeCun","submitted_at":"2026-03-15T17:02:40Z","abstract_excerpt":"We present V-JEPA 2.1, a family of self-supervised models that learn dense, high-quality visual representations for both images and videos while retaining strong global scene understanding. The approach combines four key components. First, a dense predictive loss uses a masking-based objective in which both visible and masked tokens contribute to the training signal, encouraging explicit spatial and temporal grounding. Second, deep self-supervision applies the self-supervised objective hierarchically across multiple intermediate encoder layers to improve representation quality. Third, multi-mo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.14482","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/2603.14482/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":"2603.14482","created_at":"2026-06-12T01:09:25.037781+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.14482v3","created_at":"2026-06-12T01:09:25.037781+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.14482","created_at":"2026-06-12T01:09:25.037781+00:00"},{"alias_kind":"pith_short_12","alias_value":"2KLVNDVXPBNW","created_at":"2026-06-12T01:09:25.037781+00:00"},{"alias_kind":"pith_short_16","alias_value":"2KLVNDVXPBNW24AC","created_at":"2026-06-12T01:09:25.037781+00:00"},{"alias_kind":"pith_short_8","alias_value":"2KLVNDVX","created_at":"2026-06-12T01:09:25.037781+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":18,"internal_anchor_count":18,"sample":[{"citing_arxiv_id":"2605.20388","citing_title":"How You Move Tells What You'll Do: Trajectory-Conditioned Egocentric Prediction","ref_index":19,"is_internal_anchor":true},{"citing_arxiv_id":"2605.20674","citing_title":"Modular Multimodal Classification Without Fine-Tuning: A Simple Compositional Approach","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2605.20901","citing_title":"VISTA: Technical Report for the Ego4D Short-Term Object Interaction Anticipation at EgoVis 2026","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2605.20904","citing_title":"JFAA: Technical Report for the EPIC-KITCHENS-100 Action Anticipation Challenge at EgoVis 2026","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2604.16503","citing_title":"Motif-Video 2B: Technical Report","ref_index":25,"is_internal_anchor":true},{"citing_arxiv_id":"2605.15618","citing_title":"Latent Video Prediction Learns Better World Models","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2605.17165","citing_title":"Factorized Latent Dynamics for Video JEPA: An Empirical Study of Auxiliary Objectives","ref_index":35,"is_internal_anchor":true},{"citing_arxiv_id":"2605.15394","citing_title":"Representation Without Reward: A JEPA Audit for LLM Fine-Tuning","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2605.15477","citing_title":"EgoExo-WM: Unlocking Exo Video for Ego World Models","ref_index":55,"is_internal_anchor":true},{"citing_arxiv_id":"2604.26182","citing_title":"Lifting Embodied World Models for Planning and Control","ref_index":26,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06487","citing_title":"3D MRI Image Pretraining via Controllable 2D Slice Navigation Task","ref_index":19,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06388","citing_title":"Reconstruction or Semantics? What Makes a Latent Space Useful for Robotic World Models","ref_index":38,"is_internal_anchor":true},{"citing_arxiv_id":"2605.00078","citing_title":"Being-H0.7: A Latent World-Action Model from Egocentric Videos","ref_index":110,"is_internal_anchor":true},{"citing_arxiv_id":"2604.16503","citing_title":"Motif-Video 2B: Technical Report","ref_index":25,"is_internal_anchor":true},{"citing_arxiv_id":"2605.01694","citing_title":"Latent State Design for World Models under Sufficiency Constraints","ref_index":48,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07800","citing_title":"SARA: Semantically Adaptive Relational Alignment for Video Diffusion Models","ref_index":29,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06809","citing_title":"LookWhen? Fast Video Recognition by Learning When, Where, and What to Compute","ref_index":60,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07079","citing_title":"Learning Visual Feature-Based World Models via Residual Latent Action","ref_index":37,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2KLVNDVXPBNW24AC7B2QZFXOVV","json":"https://pith.science/pith/2KLVNDVXPBNW24AC7B2QZFXOVV.json","graph_json":"https://pith.science/api/pith-number/2KLVNDVXPBNW24AC7B2QZFXOVV/graph.json","events_json":"https://pith.science/api/pith-number/2KLVNDVXPBNW24AC7B2QZFXOVV/events.json","paper":"https://pith.science/paper/2KLVNDVX"},"agent_actions":{"view_html":"https://pith.science/pith/2KLVNDVXPBNW24AC7B2QZFXOVV","download_json":"https://pith.science/pith/2KLVNDVXPBNW24AC7B2QZFXOVV.json","view_paper":"https://pith.science/paper/2KLVNDVX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.14482&json=true","fetch_graph":"https://pith.science/api/pith-number/2KLVNDVXPBNW24AC7B2QZFXOVV/graph.json","fetch_events":"https://pith.science/api/pith-number/2KLVNDVXPBNW24AC7B2QZFXOVV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2KLVNDVXPBNW24AC7B2QZFXOVV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2KLVNDVXPBNW24AC7B2QZFXOVV/action/storage_attestation","attest_author":"https://pith.science/pith/2KLVNDVXPBNW24AC7B2QZFXOVV/action/author_attestation","sign_citation":"https://pith.science/pith/2KLVNDVXPBNW24AC7B2QZFXOVV/action/citation_signature","submit_replication":"https://pith.science/pith/2KLVNDVXPBNW24AC7B2QZFXOVV/action/replication_record"}},"created_at":"2026-06-12T01:09:25.037781+00:00","updated_at":"2026-06-12T01:09:25.037781+00:00"}