{"paper":{"title":"VGGT-Long: Chunk it, Loop it, Align it -- Pushing VGGT's Limits on Kilometer-scale Long RGB Sequences","license":"http://creativecommons.org/licenses/by/4.0/","headline":"By dividing long video sequences into chunks and aligning their overlaps with lightweight loop closure, a foundation 3D model can produce accurate monocular reconstructions and trajectories over kilometer-scale outdoor paths without camera,","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Jian Yang, Jiawei Xu, Jin Xie, Kai Deng, Zexin Ti","submitted_at":"2025-07-22T10:39:04Z","abstract_excerpt":"Foundation models for 3D vision have recently demonstrated remarkable capabilities in 3D perception. However, extending these models to large-scale RGB stream 3D reconstruction remains challenging due to memory limitations. In this work, we propose VGGT-Long, a simple yet effective system that pushes the limits of monocular 3D reconstruction to kilometer-scale, unbounded outdoor environments. Our approach addresses the scalability bottlenecks of existing models through a chunk-based processing strategy combined with overlapping alignment and lightweight loop closure optimization. Without requi"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"VGGT-Long achieves trajectory and reconstruction performance comparable to traditional methods on KITTI, Waymo, and Virtual KITTI datasets without requiring camera calibration, depth supervision or model retraining.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That chunk-wise alignment plus lightweight loop closure will maintain global consistency and metric accuracy over kilometer-scale trajectories without drift or scale drift that would require additional constraints or supervision.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"VGGT-Long extends VGGT with chunking, overlap alignment, and loop closure to produce consistent kilometer-scale 3D reconstructions from monocular RGB sequences without retraining or extra supervision.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"By dividing long video sequences into chunks and aligning their overlaps with lightweight loop closure, a foundation 3D model can produce accurate monocular reconstructions and trajectories over kilometer-scale outdoor paths without camera,","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"0264686dd511248d824ef796d324c7d6ab96ae3858d2ee19806907c42cf3e82d"},"source":{"id":"2507.16443","kind":"arxiv","version":2},"verdict":{"id":"2744134d-a077-4d81-bc40-501d77b106ff","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T08:41:36.975777Z","strongest_claim":"VGGT-Long achieves trajectory and reconstruction performance comparable to traditional methods on KITTI, Waymo, and Virtual KITTI datasets without requiring camera calibration, depth supervision or model retraining.","one_line_summary":"VGGT-Long extends VGGT with chunking, overlap alignment, and loop closure to produce consistent kilometer-scale 3D reconstructions from monocular RGB sequences without retraining or extra supervision.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That chunk-wise alignment plus lightweight loop closure will maintain global consistency and metric accuracy over kilometer-scale trajectories without drift or scale drift that would require additional constraints or supervision.","pith_extraction_headline":"By dividing long video sequences into chunks and aligning their overlaps with lightweight loop closure, a foundation 3D model can produce accurate monocular reconstructions and trajectories over kilometer-scale outdoor paths without camera,"},"references":{"count":47,"sample":[{"doi":"","year":2011,"title":"Building rome in a day.Communications of the ACM, 54 (10):105–112, 2011","work_id":"9cb7cc10-d733-4f1a-947a-632a3dfc22c5","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Learning to match features with seeded graph matching network","work_id":"a15eecc5-1a30-423a-bb06-bf39ded9dc23","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Soft2: Stereo visual odometry for road vehicles based on a point- to-epipolar-line metric.IEEE Transactions on Robotics, 39 (1):273–288, 2022","work_id":"e55539dc-2f72-4581-91f7-807cd501547f","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"FlashAttention-2: Faster attention with better par- allelism and work partitioning","work_id":"d339dd1a-7759-4a93-ae6c-cea84710cbd6","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Fu, Stefano Ermon, Atri Rudra, and Christopher R´e","work_id":"b96d29be-63c4-4b92-b53f-6d6583896943","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":47,"snapshot_sha256":"95b649b6a2702c2068cb26816984e64621f040a1a8a70d3b02d22f5bbc7b0c2c","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"6c4cdb246c715d5cf3733dc6b854b9448cbcc7a06c2316c916bc11e8809469e8"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}