{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:P3M3NDEHV7P3CM7726QMF6D4FV","short_pith_number":"pith:P3M3NDEH","schema_version":"1.0","canonical_sha256":"7ed9b68c87afdfb133ffd7a0c2f87c2d6292e627a0af5191d237666a0d99ef67","source":{"kind":"arxiv","id":"2503.18931","version":2},"attestation_state":"computed","paper":{"title":"CoMP: Continual Multimodal Pre-training for Vision Foundation Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Lingchen Meng, Wujian Peng, Yitong Chen, Yu-Gang Jiang, Zuxuan Wu","submitted_at":"2025-03-24T17:52:47Z","abstract_excerpt":"Pre-trained Vision Foundation Models (VFMs) provide strong visual representations for a wide range of applications. In this paper, we continually pre-train prevailing VFMs in a multimodal manner such that they can effortlessly process visual inputs of varying sizes and produce visual representations that are more aligned with language representations, regardless of their original pre-training process. To this end, we introduce CoMP, a carefully designed multimodal pre-training pipeline. CoMP uses a Continual Rotary Position Embedding to accommodate visual inputs with different resolutions, and"},"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":"2503.18931","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-03-24T17:52:47Z","cross_cats_sorted":[],"title_canon_sha256":"9216da5eb0e46dd1cb8cab1fc2bac4eed6351777b5855b5c4e95828fbf16e319","abstract_canon_sha256":"a72d6dff2004cf1c041ff184b7cb1b0836b61bf15806a96216d24216c7cbc575"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:03:58.565353Z","signature_b64":"eKNX8vwY0vuRbuDO1OLyfOULsE6AXNknnN15TxVt6Ch6aFfbdYEXx0n3LSRBiah+Ex5DPuK1BvFnm2GJ3Y6GCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7ed9b68c87afdfb133ffd7a0c2f87c2d6292e627a0af5191d237666a0d99ef67","last_reissued_at":"2026-07-05T11:03:58.564823Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:03:58.564823Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"CoMP: Continual Multimodal Pre-training for Vision Foundation Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Lingchen Meng, Wujian Peng, Yitong Chen, Yu-Gang Jiang, Zuxuan Wu","submitted_at":"2025-03-24T17:52:47Z","abstract_excerpt":"Pre-trained Vision Foundation Models (VFMs) provide strong visual representations for a wide range of applications. In this paper, we continually pre-train prevailing VFMs in a multimodal manner such that they can effortlessly process visual inputs of varying sizes and produce visual representations that are more aligned with language representations, regardless of their original pre-training process. To this end, we introduce CoMP, a carefully designed multimodal pre-training pipeline. CoMP uses a Continual Rotary Position Embedding to accommodate visual inputs with different resolutions, and"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.18931","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/2503.18931/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":"2503.18931","created_at":"2026-07-05T11:03:58.564890+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.18931v2","created_at":"2026-07-05T11:03:58.564890+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.18931","created_at":"2026-07-05T11:03:58.564890+00:00"},{"alias_kind":"pith_short_12","alias_value":"P3M3NDEHV7P3","created_at":"2026-07-05T11:03:58.564890+00:00"},{"alias_kind":"pith_short_16","alias_value":"P3M3NDEHV7P3CM77","created_at":"2026-07-05T11:03:58.564890+00:00"},{"alias_kind":"pith_short_8","alias_value":"P3M3NDEH","created_at":"2026-07-05T11:03:58.564890+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":5,"sample":[{"citing_arxiv_id":"2606.18249","citing_title":"Unified Multimodal Autoregressive Modeling with Shared Context-Visual Tokenizer is Key to Unification","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2606.11096","citing_title":"IDEAL: In-DEpth ALignment Makes A Discrete Representation AutoEncoder","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08384","citing_title":"jina-embeddings-v5-omni: Geometry-preserving Embeddings via Locked Aligned Towers","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08384","citing_title":"jina-embeddings-v5-omni: Geometry-preserving Embeddings via Locked Aligned Towers","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08384","citing_title":"jina-embeddings-v5-omni: Geometry-preserving Embeddings via Locked Aligned Towers","ref_index":7,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/P3M3NDEHV7P3CM7726QMF6D4FV","json":"https://pith.science/pith/P3M3NDEHV7P3CM7726QMF6D4FV.json","graph_json":"https://pith.science/api/pith-number/P3M3NDEHV7P3CM7726QMF6D4FV/graph.json","events_json":"https://pith.science/api/pith-number/P3M3NDEHV7P3CM7726QMF6D4FV/events.json","paper":"https://pith.science/paper/P3M3NDEH"},"agent_actions":{"view_html":"https://pith.science/pith/P3M3NDEHV7P3CM7726QMF6D4FV","download_json":"https://pith.science/pith/P3M3NDEHV7P3CM7726QMF6D4FV.json","view_paper":"https://pith.science/paper/P3M3NDEH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.18931&json=true","fetch_graph":"https://pith.science/api/pith-number/P3M3NDEHV7P3CM7726QMF6D4FV/graph.json","fetch_events":"https://pith.science/api/pith-number/P3M3NDEHV7P3CM7726QMF6D4FV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/P3M3NDEHV7P3CM7726QMF6D4FV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/P3M3NDEHV7P3CM7726QMF6D4FV/action/storage_attestation","attest_author":"https://pith.science/pith/P3M3NDEHV7P3CM7726QMF6D4FV/action/author_attestation","sign_citation":"https://pith.science/pith/P3M3NDEHV7P3CM7726QMF6D4FV/action/citation_signature","submit_replication":"https://pith.science/pith/P3M3NDEHV7P3CM7726QMF6D4FV/action/replication_record"}},"created_at":"2026-07-05T11:03:58.564890+00:00","updated_at":"2026-07-05T11:03:58.564890+00:00"}