{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:RY4PNJH3S3E6JF7X4AWO7KXTLC","short_pith_number":"pith:RY4PNJH3","schema_version":"1.0","canonical_sha256":"8e38f6a4fb96c9e497f7e02cefaaf358810a038e20ab0d59e85867d324758f84","source":{"kind":"arxiv","id":"2506.04997","version":1},"attestation_state":"computed","paper":{"title":"Towards Storage-Efficient Visual Document Retrieval: An Empirical Study on Reducing Patch-Level Embeddings","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.IR","authors_text":"Aixin Sun, Haodong Duan, Jiaqi Wang, Jinsong Li, Pan Zhang, Xiaobao Wu, Xiaoyi Dong, Yixin Cao, Yubo Ma, Yuhang Cao, Yuhang Zang","submitted_at":"2025-06-05T13:06:01Z","abstract_excerpt":"Despite the strong performance of ColPali/ColQwen2 in Visualized Document Retrieval (VDR), it encodes each page into multiple patch-level embeddings and leads to excessive memory usage. This empirical study investigates methods to reduce patch embeddings per page at minimum performance degradation. We evaluate two token-reduction strategies: token pruning and token merging. Regarding token pruning, we surprisingly observe that a simple random strategy outperforms other sophisticated pruning methods, though still far from satisfactory. Further analysis reveals that pruning is inherently unsuita"},"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":"2506.04997","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.IR","submitted_at":"2025-06-05T13:06:01Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"1f19adb2ad78196cb756f0e3f9606463ed06024521c9a8a60982acaf5ec9737f","abstract_canon_sha256":"8f8ffb75819278d0a0e13b504aa5336ac64e6996c6b6ce3a0d64eec28b605310"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:16:40.574795Z","signature_b64":"kbusaizMv/hfkv43ThimDdavQYXDlyZPYdA1i57EyU93p0jkkmaG8/2OAYtMKHEnDpUqCNv150Cof60ddaXtBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8e38f6a4fb96c9e497f7e02cefaaf358810a038e20ab0d59e85867d324758f84","last_reissued_at":"2026-07-05T11:16:40.573982Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:16:40.573982Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards Storage-Efficient Visual Document Retrieval: An Empirical Study on Reducing Patch-Level Embeddings","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.IR","authors_text":"Aixin Sun, Haodong Duan, Jiaqi Wang, Jinsong Li, Pan Zhang, Xiaobao Wu, Xiaoyi Dong, Yixin Cao, Yubo Ma, Yuhang Cao, Yuhang Zang","submitted_at":"2025-06-05T13:06:01Z","abstract_excerpt":"Despite the strong performance of ColPali/ColQwen2 in Visualized Document Retrieval (VDR), it encodes each page into multiple patch-level embeddings and leads to excessive memory usage. This empirical study investigates methods to reduce patch embeddings per page at minimum performance degradation. We evaluate two token-reduction strategies: token pruning and token merging. Regarding token pruning, we surprisingly observe that a simple random strategy outperforms other sophisticated pruning methods, though still far from satisfactory. Further analysis reveals that pruning is inherently unsuita"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.04997","kind":"arxiv","version":1},"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/2506.04997/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":"2506.04997","created_at":"2026-07-05T11:16:40.574045+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.04997v1","created_at":"2026-07-05T11:16:40.574045+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.04997","created_at":"2026-07-05T11:16:40.574045+00:00"},{"alias_kind":"pith_short_12","alias_value":"RY4PNJH3S3E6","created_at":"2026-07-05T11:16:40.574045+00:00"},{"alias_kind":"pith_short_16","alias_value":"RY4PNJH3S3E6JF7X","created_at":"2026-07-05T11:16:40.574045+00:00"},{"alias_kind":"pith_short_8","alias_value":"RY4PNJH3","created_at":"2026-07-05T11:16:40.574045+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2601.21262","citing_title":"CausalEmbed: Auto-Regressive Multi-Vector Generation in Latent Space for Visual Document Embedding","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10167","citing_title":"Visual Late Chunking: An Empirical Study of Contextual Chunking for Efficient Visual Document Retrieval","ref_index":21,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RY4PNJH3S3E6JF7X4AWO7KXTLC","json":"https://pith.science/pith/RY4PNJH3S3E6JF7X4AWO7KXTLC.json","graph_json":"https://pith.science/api/pith-number/RY4PNJH3S3E6JF7X4AWO7KXTLC/graph.json","events_json":"https://pith.science/api/pith-number/RY4PNJH3S3E6JF7X4AWO7KXTLC/events.json","paper":"https://pith.science/paper/RY4PNJH3"},"agent_actions":{"view_html":"https://pith.science/pith/RY4PNJH3S3E6JF7X4AWO7KXTLC","download_json":"https://pith.science/pith/RY4PNJH3S3E6JF7X4AWO7KXTLC.json","view_paper":"https://pith.science/paper/RY4PNJH3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.04997&json=true","fetch_graph":"https://pith.science/api/pith-number/RY4PNJH3S3E6JF7X4AWO7KXTLC/graph.json","fetch_events":"https://pith.science/api/pith-number/RY4PNJH3S3E6JF7X4AWO7KXTLC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RY4PNJH3S3E6JF7X4AWO7KXTLC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RY4PNJH3S3E6JF7X4AWO7KXTLC/action/storage_attestation","attest_author":"https://pith.science/pith/RY4PNJH3S3E6JF7X4AWO7KXTLC/action/author_attestation","sign_citation":"https://pith.science/pith/RY4PNJH3S3E6JF7X4AWO7KXTLC/action/citation_signature","submit_replication":"https://pith.science/pith/RY4PNJH3S3E6JF7X4AWO7KXTLC/action/replication_record"}},"created_at":"2026-07-05T11:16:40.574045+00:00","updated_at":"2026-07-05T11:16:40.574045+00:00"}