{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:EXQNZOXAQJUXJ7PEJANKKHMECX","short_pith_number":"pith:EXQNZOXA","schema_version":"1.0","canonical_sha256":"25e0dcbae0826974fde4481aa51d8415f60ebce0be03c0dd063fa5a3b846d0e2","source":{"kind":"arxiv","id":"2605.16981","version":1},"attestation_state":"computed","paper":{"title":"Rethinking the State Update Gate for Long-Sequence Recurrent 3D Reconstruction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A scalar frame-level gate computed from internal feature changes extends effective memory in recurrent 3D reconstruction without added cost or training.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Kejun Ren, Lei Jin, Lianming Xu, Li Wang, Tianxin Huang","submitted_at":"2026-05-16T13:00:49Z","abstract_excerpt":"Streaming 3D reconstruction under a strict constant-memory budget hinges on how the recurrent state is updated as the stream evolves. We profile TTT3R-style per-token gates across five benchmarks and discover a structural bottleneck: the gate is intrinsically bounded in magnitude (median $0.31$; never exceeding $0.6$) and nearly frame-invariant, yielding an effective memory horizon of only $\\sim$3 frames per state token, which serves as the structural origin of long-sequence drift. We trace this to a missing axis: existing inference-time methods modulate updates only at the per-token, intra-fr"},"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":true,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.16981","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-16T13:00:49Z","cross_cats_sorted":[],"title_canon_sha256":"6b5db4fd1aa168ba6b99db5e48d4ff6e681c78435c524d78409314b11bf27f5d","abstract_canon_sha256":"1c3b015c024990a672519fde5c56113184486eac6fd37de8d5ff14fc82dcb68c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:03:34.301516Z","signature_b64":"rHcDMFb1xsKRKOFCnP5Ifk6zCzz3awsW7TyNJ07CoyRM9yNw36mwrJwG8NiQM6E9PDI3oFcAime3+CvYUezmBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"25e0dcbae0826974fde4481aa51d8415f60ebce0be03c0dd063fa5a3b846d0e2","last_reissued_at":"2026-05-20T00:03:34.300713Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:03:34.300713Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Rethinking the State Update Gate for Long-Sequence Recurrent 3D Reconstruction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A scalar frame-level gate computed from internal feature changes extends effective memory in recurrent 3D reconstruction without added cost or training.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Kejun Ren, Lei Jin, Lianming Xu, Li Wang, Tianxin Huang","submitted_at":"2026-05-16T13:00:49Z","abstract_excerpt":"Streaming 3D reconstruction under a strict constant-memory budget hinges on how the recurrent state is updated as the stream evolves. We profile TTT3R-style per-token gates across five benchmarks and discover a structural bottleneck: the gate is intrinsically bounded in magnitude (median $0.31$; never exceeding $0.6$) and nearly frame-invariant, yielding an effective memory horizon of only $\\sim$3 frames per state token, which serves as the structural origin of long-sequence drift. We trace this to a missing axis: existing inference-time methods modulate updates only at the per-token, intra-fr"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our gate cuts ATE by 51% on long TUM-RGBD pose sequences, reduces AbsRel by 12.8% on Bonn video depth, and on KITTI long-sequence pose estimation surpasses both LongStream and Keyframe-VO, while retaining strictly constant memory at zero training cost.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"Frame-to-frame changes of internal features can be used to derive in closed form a scalar α_t that correctly determines how strongly each frame should contribute to the recurrent state, serving as a content-independent continuous relaxation of classical SLAM keyframe selection.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A closed-form scalar frame-level gate α_t derived from internal feature changes extends effective memory in recurrent 3D reconstruction and improves accuracy on long sequences up to 4541 frames.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A scalar frame-level gate computed from internal feature changes extends effective memory in recurrent 3D reconstruction without added cost or training.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b1926b38b3cab2df24ba795e86f1fd66001fc45c883b453e1d9d263f03cef9f2"},"source":{"id":"2605.16981","kind":"arxiv","version":1},"verdict":{"id":"7d53eb50-aeea-491f-b580-76e8d022e579","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T20:09:53.510754Z","strongest_claim":"Our gate cuts ATE by 51% on long TUM-RGBD pose sequences, reduces AbsRel by 12.8% on Bonn video depth, and on KITTI long-sequence pose estimation surpasses both LongStream and Keyframe-VO, while retaining strictly constant memory at zero training cost.","one_line_summary":"A closed-form scalar frame-level gate α_t derived from internal feature changes extends effective memory in recurrent 3D reconstruction and improves accuracy on long sequences up to 4541 frames.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"Frame-to-frame changes of internal features can be used to derive in closed form a scalar α_t that correctly determines how strongly each frame should contribute to the recurrent state, serving as a content-independent continuous relaxation of classical SLAM keyframe selection.","pith_extraction_headline":"A scalar frame-level gate computed from internal feature changes extends effective memory in recurrent 3D reconstruction without added cost or training."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.16981/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T20:31:19.036098Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T20:21:39.464394Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"citation_quote_validity","ran_at":"2026-05-19T19:50:06.726217Z","status":"skipped","version":"0.1.0","findings_count":0},{"name":"cited_work_retraction","ran_at":"2026-05-19T19:24:11.670191Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T18:41:56.214928Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.302550Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"9e1d96f1ca0c235f6ca0daf8138b3bac2a06501e4c61e8dae00a91cbc5d6f820"},"references":{"count":31,"sample":[{"doi":"","year":2022,"title":"Neural rgb-d surface reconstruction","work_id":"6d621e0a-3d83-421a-bde8-d6e715851bf6","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Orb- slam3: An accurate open-source library for visual, visual–inertial, and multimap slam.IEEE transactions on robotics, 37(6):1874–1890, 2021","work_id":"f78cefb5-6d76-4b28-9d3b-90805439c727","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"TTT3R: 3D Reconstruction as Test-Time Training","work_id":"a2dcce9c-5c33-49e2-989b-e0438645054f","ref_index":3,"cited_arxiv_id":"2509.26645","is_internal_anchor":true},{"doi":"","year":2025,"title":"Long3r: Long sequence streaming 3d reconstruction","work_id":"3010298f-dff4-41a9-946b-cb8bdc8cc6b2","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2026,"title":"Longstream: Long-sequence streaming autoregressive visual geometry","work_id":"16ddac87-9858-4ac3-9335-d3f9513c2e63","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":31,"snapshot_sha256":"b8046857bce9f1cf18212771b2296145b87c74114d8b7c14795c53268715b866","internal_anchors":7},"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":"2605.16981","created_at":"2026-05-20T00:03:34.300851+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.16981v1","created_at":"2026-05-20T00:03:34.300851+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.16981","created_at":"2026-05-20T00:03:34.300851+00:00"},{"alias_kind":"pith_short_12","alias_value":"EXQNZOXAQJUX","created_at":"2026-05-20T00:03:34.300851+00:00"},{"alias_kind":"pith_short_16","alias_value":"EXQNZOXAQJUXJ7PE","created_at":"2026-05-20T00:03:34.300851+00:00"},{"alias_kind":"pith_short_8","alias_value":"EXQNZOXA","created_at":"2026-05-20T00:03:34.300851+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EXQNZOXAQJUXJ7PEJANKKHMECX","json":"https://pith.science/pith/EXQNZOXAQJUXJ7PEJANKKHMECX.json","graph_json":"https://pith.science/api/pith-number/EXQNZOXAQJUXJ7PEJANKKHMECX/graph.json","events_json":"https://pith.science/api/pith-number/EXQNZOXAQJUXJ7PEJANKKHMECX/events.json","paper":"https://pith.science/paper/EXQNZOXA"},"agent_actions":{"view_html":"https://pith.science/pith/EXQNZOXAQJUXJ7PEJANKKHMECX","download_json":"https://pith.science/pith/EXQNZOXAQJUXJ7PEJANKKHMECX.json","view_paper":"https://pith.science/paper/EXQNZOXA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.16981&json=true","fetch_graph":"https://pith.science/api/pith-number/EXQNZOXAQJUXJ7PEJANKKHMECX/graph.json","fetch_events":"https://pith.science/api/pith-number/EXQNZOXAQJUXJ7PEJANKKHMECX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EXQNZOXAQJUXJ7PEJANKKHMECX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EXQNZOXAQJUXJ7PEJANKKHMECX/action/storage_attestation","attest_author":"https://pith.science/pith/EXQNZOXAQJUXJ7PEJANKKHMECX/action/author_attestation","sign_citation":"https://pith.science/pith/EXQNZOXAQJUXJ7PEJANKKHMECX/action/citation_signature","submit_replication":"https://pith.science/pith/EXQNZOXAQJUXJ7PEJANKKHMECX/action/replication_record"}},"created_at":"2026-05-20T00:03:34.300851+00:00","updated_at":"2026-05-20T00:03:34.300851+00:00"}