{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:WEXOUBOEAOQ67IZEFDHGSIWCVP","short_pith_number":"pith:WEXOUBOE","schema_version":"1.0","canonical_sha256":"b12eea05c403a1efa32428ce6922c2abeedcc43a43935101aea759c9d3f9c646","source":{"kind":"arxiv","id":"2502.08769","version":3},"attestation_state":"computed","paper":{"title":"Cluster and Predict Latent Patches for Improved Masked Image Modeling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Federico Baldassarre, Julien Mairal, Maxime Oquab, Piotr Bojanowski, Timoth\\'ee Darcet","submitted_at":"2025-02-12T20:17:10Z","abstract_excerpt":"Masked Image Modeling (MIM) offers a promising approach to self-supervised representation learning, however existing MIM models still lag behind the state-of-the-art. In this paper, we systematically analyze target representations, loss functions, and architectures, to introduce CAPI - a novel pure-MIM framework that relies on the prediction of latent clusterings. Our approach leverages a clustering-based loss, which is stable to train, and exhibits promising scaling properties. Our ViT-L backbone, CAPI, achieves 83.8% accuracy on ImageNet and 32.1% mIoU on ADE20K with simple linear probes, su"},"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":"2502.08769","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-02-12T20:17:10Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"64ad90edbc64c0173b81e7a243ac70b606586ced9c44bc6a02f7d158dbaab7c2","abstract_canon_sha256":"2ce86c158ca02a0e6c8d7fa80316d538bb1d7063c19705ac02205de781ae1c8d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:29:06.905923Z","signature_b64":"fSIJRln3drzSdvxj0DJo0bd7YoC9AYG2jFIysxlWF97gMsRqrhvNztBty50ogHE1Q5q3M8J/473cELXKuAIjDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b12eea05c403a1efa32428ce6922c2abeedcc43a43935101aea759c9d3f9c646","last_reissued_at":"2026-07-05T11:29:06.905413Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:29:06.905413Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Cluster and Predict Latent Patches for Improved Masked Image Modeling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Federico Baldassarre, Julien Mairal, Maxime Oquab, Piotr Bojanowski, Timoth\\'ee Darcet","submitted_at":"2025-02-12T20:17:10Z","abstract_excerpt":"Masked Image Modeling (MIM) offers a promising approach to self-supervised representation learning, however existing MIM models still lag behind the state-of-the-art. In this paper, we systematically analyze target representations, loss functions, and architectures, to introduce CAPI - a novel pure-MIM framework that relies on the prediction of latent clusterings. Our approach leverages a clustering-based loss, which is stable to train, and exhibits promising scaling properties. Our ViT-L backbone, CAPI, achieves 83.8% accuracy on ImageNet and 32.1% mIoU on ADE20K with simple linear probes, su"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.08769","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/2502.08769/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":"2502.08769","created_at":"2026-07-05T11:29:06.905473+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.08769v3","created_at":"2026-07-05T11:29:06.905473+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.08769","created_at":"2026-07-05T11:29:06.905473+00:00"},{"alias_kind":"pith_short_12","alias_value":"WEXOUBOEAOQ6","created_at":"2026-07-05T11:29:06.905473+00:00"},{"alias_kind":"pith_short_16","alias_value":"WEXOUBOEAOQ67IZE","created_at":"2026-07-05T11:29:06.905473+00:00"},{"alias_kind":"pith_short_8","alias_value":"WEXOUBOE","created_at":"2026-07-05T11:29:06.905473+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23503","citing_title":"UniverSat: Resolution- and Modality-Agnostic Transformers for Earth Observation","ref_index":64,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03245","citing_title":"Text-Conditional JEPA for Learning Semantically Rich Visual Representations","ref_index":6,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WEXOUBOEAOQ67IZEFDHGSIWCVP","json":"https://pith.science/pith/WEXOUBOEAOQ67IZEFDHGSIWCVP.json","graph_json":"https://pith.science/api/pith-number/WEXOUBOEAOQ67IZEFDHGSIWCVP/graph.json","events_json":"https://pith.science/api/pith-number/WEXOUBOEAOQ67IZEFDHGSIWCVP/events.json","paper":"https://pith.science/paper/WEXOUBOE"},"agent_actions":{"view_html":"https://pith.science/pith/WEXOUBOEAOQ67IZEFDHGSIWCVP","download_json":"https://pith.science/pith/WEXOUBOEAOQ67IZEFDHGSIWCVP.json","view_paper":"https://pith.science/paper/WEXOUBOE","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.08769&json=true","fetch_graph":"https://pith.science/api/pith-number/WEXOUBOEAOQ67IZEFDHGSIWCVP/graph.json","fetch_events":"https://pith.science/api/pith-number/WEXOUBOEAOQ67IZEFDHGSIWCVP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WEXOUBOEAOQ67IZEFDHGSIWCVP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WEXOUBOEAOQ67IZEFDHGSIWCVP/action/storage_attestation","attest_author":"https://pith.science/pith/WEXOUBOEAOQ67IZEFDHGSIWCVP/action/author_attestation","sign_citation":"https://pith.science/pith/WEXOUBOEAOQ67IZEFDHGSIWCVP/action/citation_signature","submit_replication":"https://pith.science/pith/WEXOUBOEAOQ67IZEFDHGSIWCVP/action/replication_record"}},"created_at":"2026-07-05T11:29:06.905473+00:00","updated_at":"2026-07-05T11:29:06.905473+00:00"}