{"paper":{"title":"3D Reconstruction with Spatial Memory","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Spann3R maintains an external spatial memory to regress per-image pointmaps directly in a global coordinate system from ordered or unordered image collections.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Hengyi Wang, Lourdes Agapito","submitted_at":"2024-08-28T18:01:00Z","abstract_excerpt":"We present Spann3R, a novel approach for dense 3D reconstruction from ordered or unordered image collections. Built on the DUSt3R paradigm, Spann3R uses a transformer-based architecture to directly regress pointmaps from images without any prior knowledge of the scene or camera parameters. Unlike DUSt3R, which predicts per image-pair pointmaps each expressed in its local coordinate frame, Spann3R can predict per-image pointmaps expressed in a global coordinate system, thus eliminating the need for optimization-based global alignment. The key idea of Spann3R is to manage an external spatial mem"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Spann3R can predict per-image pointmaps expressed in a global coordinate system, thus eliminating the need for optimization-based global alignment.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That an external spatial memory can reliably retain and retrieve all relevant prior 3D information across arbitrary ordered or unordered image collections without drift or loss of consistency.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Spann3R uses a learned spatial memory to regress per-image pointmaps directly in a shared global coordinate system, removing the need for optimization-based alignment after per-pair predictions.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Spann3R maintains an external spatial memory to regress per-image pointmaps directly in a global coordinate system from ordered or unordered image collections.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"08b1de9de7d50e31b5f29fea6f6e147f7b64c733e38e0b207377e78b468dbc35"},"source":{"id":"2408.16061","kind":"arxiv","version":1},"verdict":{"id":"fe694b86-e570-45e5-b5f6-051412618e53","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T07:25:46.136038Z","strongest_claim":"Spann3R can predict per-image pointmaps expressed in a global coordinate system, thus eliminating the need for optimization-based global alignment.","one_line_summary":"Spann3R uses a learned spatial memory to regress per-image pointmaps directly in a shared global coordinate system, removing the need for optimization-based alignment after per-pair predictions.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That an external spatial memory can reliably retain and retrieve all relevant prior 3D information across arbitrary ordered or unordered image collections without drift or loss of consistency.","pith_extraction_headline":"Spann3R maintains an external spatial memory to regress per-image pointmaps directly in a global coordinate system from ordered or unordered image collections."},"references":{"count":95,"sample":[{"doi":"","year":2016,"title":"Large-scale data for multiple-view stereopsis","work_id":"59e8513c-0002-466c-9df6-149362b24863","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2009,"title":"Building rome in a day","work_id":"f5d6ddbf-5d90-485b-bab8-845195de1221","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2010,"title":"Bundle adjustment in the large","work_id":"7999b7bd-140d-481a-8fdd-ed245119639c","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Map-free visual relocalization: Metric pose relative to a single image","work_id":"f07669bc-42a2-4775-95d6-0fc178a4997a","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Human mem- ory: A proposed system and its control processes","work_id":"882e21c1-2606-4802-b75c-f35f0905d0b7","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":95,"snapshot_sha256":"62782dd1bf249ee604bb4810249d5001510c54f98dc5bf6ee4499f89dcff2c27","internal_anchors":1},"formal_canon":{"evidence_count":3,"snapshot_sha256":"bc859328ff2e8efb18e5322841e646239961417c7494790a949d1080b5d7a736"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}