pith. sign in

arxiv: 2605.30310 · v1 · pith:PWUSTAG2new · submitted 2026-05-28 · 💻 cs.CV · cs.AI· cs.GR

City-Mesh3R: Simulation-Ready City-Scale 3D Mesh Reconstruction from Multi-View Images

Sayan Paul , Sourav Ghosh , Siddharth Katageri , Soumyadip Maity , Sanjana Sinha , Brojeshwar Bhowmick This is my paper

Pith reviewed 2026-06-29 08:03 UTC · model grok-4.3

classification 💻 cs.CV cs.AIcs.GR
keywords city-scale 3D reconstructionwatertight meshmulti-view imagesdivide-and-conquerSfMsurface reconstructionsimulation-ready meshesmesh stitching
0
0 comments X

The pith

A divide-and-conquer pipeline reconstructs watertight city-scale 3D meshes directly from unordered photo collections.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents City-Mesh3R as a framework that turns large sets of urban images into high-fidelity watertight meshes ready for simulation. It splits the work by first clustering images topologically, running independent sparse SfM on each cluster, and merging the maps. The scene is then partitioned spatially for dense reconstruction, curvature-aware remeshing, and final stitching of the parts. This matters because prior methods either fail to produce solid meshes or cannot handle city-sized data without excessive computation. If the approach holds, it would make practical 3D city models feasible from standard photo collections.

Core claim

City-Mesh3R reconstructs watertight surface meshes from large unordered image collections by first performing topological image clustering, cluster-wise independent sparse SfM and map merging to build a sparse city map, then spatially partitioning it for geometry-aware camera selection, dense surface reconstruction and curvature-aware adaptive vertex density remeshing, and finally stitching the partition meshes to produce the global mesh.

What carries the argument

The divide-and-conquer pipeline of topological image clustering for sparse SfM, spatial partitioning for dense reconstruction, curvature-aware remeshing, and partition stitching.

If this is right

  • Reconstruction becomes feasible for arbitrarily large urban scenes through distributed local processing.
  • The output meshes exhibit regular geometry and fine surface details suitable for 3D simulation.
  • The pipeline avoids exhaustive global feature matching while still recovering complete geometry.
  • It produces simulation-ready meshes unlike incomplete or noisy results from radiance-field methods.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The local-to-global consistency pattern might apply to other distributed geometry tasks such as large terrain modeling.
  • Incremental updates to clusters could enable dynamic city reconstruction if the merging step supports it.
  • Success at city scale would suggest that careful boundary handling in stitching can substitute for full global optimization.

Load-bearing premise

Independent local SfM maps and partitioned dense meshes can be merged and stitched to produce globally consistent geometry without seams, holes, or topological errors at city scale.

What would settle it

Visual or geometric inspection of the stitched output on a city-scale dataset showing discontinuities, holes, or topology errors at partition boundaries would disprove global consistency.

Figures

Figures reproduced from arXiv: 2605.30310 by Brojeshwar Bhowmick, Sanjana Sinha, Sayan Paul, Siddharth Katageri, Soumyadip Maity, Sourav Ghosh.

Figure 1
Figure 1. Figure 1: Our method reconstructs simulation-ready watertight 3D meshes with high geometric fidelity, smooth surface normals, from [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Our proposed pipeline: Starting from an unordered input image set, we first build a sparse SfM representation of the entire scene [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative Comparison with recent city-scale surface reconstruction methods (CityGS-v2[ [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: (A) Ablation Study of ”Poisson Only” vs ”Poisson + MeshOpt/Remesh” across different scenes of GauU-Scene dataset [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Area Partitioning Visualizer: The selected partition’s sparse points are highlighted by orange and all other points are shown in gray. All the candidate cameras of a partition are denoted by yellow and the top-M ranked cameras are denoted by blue. Camera score aggregation. Initialize Si = 0 for all i ∈ Icand r,c . To avoid over-counting many near-equivalent pairs for one point, we retain only the top-K sco… view at source ↗
Figure 6
Figure 6. Figure 6: Extra Qualitative Comparison Results with recent city-scale surface reconstruction methods (CityGS-v2 and CityGS-X) on [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Mesh Stitching Result: (A) Two adjacent partition meshes with seam-region points highlighted in green. (B) Sur￾face normals of the final stitched mesh. Our seam stitching yields a topologically consistent and geometrically smooth merge, pre￾serving clean boundaries and normal continuity across partitions. and flipped when doing so improves local triangle quality and valence regularity, subject to the same … view at source ↗
read the original abstract

City-scale 3D surface reconstruction from multiview images for downstream 3D simulation, poses highly challenging problems due to the scale and complexity of urban scenes. Existing city-scale 3D reconstruction methods based on NeRF, Gaussian Splatting etc. often fail to recover 3D meshes ready for simulation due to incomplete/missing geometry and irregular, noisy surfaces. Scaling existing small-scale 3D reconstruction methods to arbitrarily large urban scenes is highly infeasible due to their computational complexity. We present City-Mesh3R, a scalable framework for reconstructing watertight surface meshes directly from large unordered image collections. Unlike recent methods which use global sparse SfM point-cloud initialization followed by a distributed 3D dense reconstruction of large-scale scenes, our method follows an end-to-end images-to-mesh 3D reconstruction approach using a divide-and-conquer strategy. The sparse city map is reconstructed via topological image clustering, cluster-wise independent sparse SfM and map merging, without need for exhaustive image feature matching. Then this map is partitioned spatially to perform geometry-aware camera selection, followed by dense surface reconstruction and surface refinement using curvature-aware adaptive vertex density remeshing. These partition meshes are then stitched together to produce the global mesh of the city. The proposed end-to-end framework is evaluated on city-scale reconstruction datasets. As demonstrated by our qualitative and quantitative results, our proposed method yields high-fidelity watertight 3D meshes with regular geometry, capturing fine surface details, and is suitable for scaling to arbitrarily large scenes owing to the end-to-end processing in a distributed setting.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper introduces City-Mesh3R, a divide-and-conquer pipeline for city-scale 3D mesh reconstruction from unordered multi-view images. It performs topological image clustering followed by independent per-cluster SfM and map merging, then spatially partitions the map for per-partition dense reconstruction, curvature-aware remeshing, and final stitching to produce a single watertight simulation-ready mesh. The abstract asserts that this yields high-fidelity meshes with regular geometry and scales to arbitrarily large scenes, supported by qualitative and quantitative results on city-scale datasets.

Significance. If the stitching and merging stages can be shown to enforce global consistency, the method would address a practical gap between existing NeRF/GS approaches (which rarely output simulation-ready meshes) and small-scale mesh pipelines (which do not scale). The end-to-end distributed design and curvature-aware remeshing are potentially useful contributions for urban simulation applications.

major comments (2)
  1. [Abstract] Abstract: The central claim that the pipeline 'yields high-fidelity watertight 3D meshes ... suitable for scaling to arbitrarily large scenes' depends on the map-merging and partition-stitching stages producing globally consistent geometry without seams, holes, or topological errors. No mechanism for boundary alignment, drift correction, or seam removal is described, and no quantitative metrics, error bars, ablation studies, or dataset details are supplied to support this.
  2. [Abstract] The abstract states that results are demonstrated by 'qualitative and quantitative results,' yet supplies none of the latter (no tables of accuracy, completeness, or runtime metrics; no comparison to baselines). This absence is load-bearing for the scalability and fidelity assertions.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their detailed review and constructive criticism. We address each major comment point by point below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that the pipeline 'yields high-fidelity watertight 3D meshes ... suitable for scaling to arbitrarily large scenes' depends on the map-merging and partition-stitching stages producing globally consistent geometry without seams, holes, or topological errors. No mechanism for boundary alignment, drift correction, or seam removal is described, and no quantitative metrics, error bars, ablation studies, or dataset details are supplied to support this.

    Authors: The referee correctly notes that the abstract provides no description of mechanisms for boundary alignment, drift correction, or seam removal. The manuscript body describes the topological clustering, independent SfM per cluster, map merging, spatial partitioning, and final stitching, but we acknowledge that explicit details on consistency enforcement across boundaries are not sufficiently elaborated to support the central claim. We will revise the manuscript to add a dedicated subsection on boundary handling and consistency in the method, along with quantitative seam-error metrics in the experiments. revision: yes

  2. Referee: [Abstract] The abstract states that results are demonstrated by 'qualitative and quantitative results,' yet supplies none of the latter (no tables of accuracy, completeness, or runtime metrics; no comparison to baselines). This absence is load-bearing for the scalability and fidelity assertions.

    Authors: The abstract references both qualitative and quantitative results, with the latter appearing in the experimental section of the manuscript. However, the referee is right that the abstract itself contains no supporting numbers, tables, or baseline comparisons, which weakens the scalability and fidelity claims as presented. We will revise the abstract to either remove the unqualified reference to quantitative results or include a concise statement of key metrics and comparisons. revision: partial

Circularity Check

0 steps flagged

No circularity: standard pipeline steps with no fitted predictions or self-definitional reductions

full rationale

The paper describes a divide-and-conquer pipeline consisting of topological image clustering, cluster-wise SfM with map merging, spatial partitioning, geometry-aware camera selection, dense reconstruction, curvature-aware remeshing, and final stitching. No equations, fitted parameters, or predictions are presented that reduce by construction to inputs. No self-citations are invoked as load-bearing uniqueness theorems or ansatzes. The claims rest on the described sequence of established vision operations evaluated on external datasets, rendering the derivation self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Review performed on abstract only; no explicit free parameters, invented entities, or non-standard axioms are stated.

axioms (1)
  • domain assumption Standard Structure-from-Motion assumptions hold when applied independently to image clusters at city scale.
    The method relies on cluster-wise independent sparse SfM and map merging.

pith-pipeline@v0.9.1-grok · 5854 in / 1248 out tokens · 25800 ms · 2026-06-29T08:03:46.066994+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

54 extracted references · 7 canonical work pages · 1 internal anchor

  1. [1]

    Building rome in a day.Communications of the ACM, 54 (10):105–112, 2011

    Sameer Agarwal, Yasutaka Furukawa, Noah Snavely, Ian Si- mon, Brian Curless, Steven M Seitz, and Richard Szeliski. Building rome in a day.Communications of the ACM, 54 (10):105–112, 2011. 2

    2011

  2. [2]

    Building rome in a day

    Sameer Agarwal, Yasutaka Furukawa, Noah Snavely, et al. Building rome in a day. InICCV, 2011. 2

    2011

  3. [3]

    Netvlad: Cnn architecture for weakly supervised place recognition, 2016

    Relja Arandjelovi ´c, Petr Gronat, Akihiko Torii, Tomas Pa- jdla, and Josef Sivic. Netvlad: Cnn architecture for weakly supervised place recognition, 2016. 3

    2016

  4. [4]

    Barron, Ben Mildenhall, Dor Verbin, Pratul P

    Jonathan T. Barron, Ben Mildenhall, Dor Verbin, Pratul P. Srinivasan, and Peter Hedman. Mip-nerf 360: Unbounded anti-aliased neural radiance fields.CVPR, 2022. 8

    2022

  5. [5]

    Kanishk Chaturvedi and Thomas Kolbe. Integrating dynamic data and sensors with semantic 3d city models in the con- text of smart cities.ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, IV-2/W1: 31–38, 2016. 1

    2016

  6. [6]

    Pgsr: Planar-based gaussian splatting for ef- ficient and high-fidelity surface reconstruction.IEEE Trans- actions on Visualization and Computer Graphics, 2024

    Danpeng Chen, Hai Li, Weicai Ye, Yifan Wang, Weijian Xie, Shangjin Zhai, Nan Wang, Haomin Liu, Hujun Bao, and Guofeng Zhang. Pgsr: Planar-based gaussian splatting for ef- ficient and high-fidelity surface reconstruction.IEEE Trans- actions on Visualization and Computer Graphics, 2024. 3

    2024

  7. [7]

    Gigags: 3d gaussian based planar representation for large-scene surface reconstruction

    Junyi Chen, Weicai Ye, Yifan Wang, Danpeng Chen, Di Huang, Wanli Ouyang, Guofeng Zhang, Yu Qiao, and Tong He. Gigags: 3d gaussian based planar representation for large-scene surface reconstruction. InProceedings of the AAAI Conference on Artificial Intelligence, pages 2088– 2096, 2025. 3

    2088

  8. [8]

    Dogs: Distributed-oriented gaus- sian splatting for large-scale 3d reconstruction via gaussian consensus.Advances in Neural Information Processing Sys- tems, 37:34487–34512, 2024

    Yu Chen and Gim Hee Lee. Dogs: Distributed-oriented gaus- sian splatting for large-scale 3d reconstruction via gaussian consensus.Advances in Neural Information Processing Sys- tems, 37:34487–34512, 2024. 3

    2024

  9. [9]

    Graph-based parallel large scale structure from motion,

    Yu Chen, Shuhan Shen, Yisong Chen, and Guoping Wang. Graph-based parallel large scale structure from motion,

  10. [10]

    Andreas Christodoulides, Gary K. L. Tam, James Clarke, Richard Smith, Jon Horgan, Nicholas Micallef, Jeremy Mor- ley, Nelly Villamizar, and Sean Walton. Survey on 3d recon- struction techniques: Large-scale urban city reconstruction and requirements.IEEE Transactions on Visualization and Computer Graphics, 31(10):9343–9367, 2025. 1

    2025

  11. [11]

    Mast3r- sfm: a fully-integrated solution for unconstrained structure- from-motion, 2024

    Bardienus Duisterhof, Lojze Zust, Philippe Weinzaepfel, Vincent Leroy, Yohann Cabon, and Jerome Revaud. Mast3r- sfm: a fully-integrated solution for unconstrained structure- from-motion, 2024. 5

    2024

  12. [12]

    Citygs- x: A scalable architecture for efficient and geometrically accurate large-scale scene reconstruction.arXiv preprint arXiv:2503.23044, 2025

    Yuanyuan Gao, Hao Li, Jiaqi Chen, Zhengyu Zou, Zhihang Zhong, Dingwen Zhang, Xiao Sun, and Junwei Han. Citygs- x: A scalable architecture for efficient and geometrically accurate large-scale scene reconstruction.arXiv preprint arXiv:2503.23044, 2025. 1, 3, 6, 7

    work page arXiv 2025

  13. [13]

    Surfacesplat: Connecting surface recon- struction and gaussian splatting

    Zihui Gao, Jia-Wang Bian, Guosheng Lin, Hao Chen, and Chunhua Shen. Surfacesplat: Connecting surface recon- struction and gaussian splatting. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 28525–28534, 2025. 3

    2025

  14. [14]

    Sugar: Surface- aligned gaussian splatting for efficient 3d mesh reconstruc- tion and high-quality mesh rendering

    Antoine Gu ´edon and Vincent Lepetit. Sugar: Surface- aligned gaussian splatting for efficient 3d mesh reconstruc- tion and high-quality mesh rendering. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5354–5363, 2024. 3

    2024

  15. [15]

    Milo: Mesh- in-the-loop gaussian splatting for detailed and efficient sur- face reconstruction.ACM Transactions on Graphics (TOG), 44(6):1–15, 2025

    Antoine Gu ´edon, Diego Gomez, Nissim Maruani, Bingchen Gong, George Drettakis, and Maks Ovsjanikov. Milo: Mesh- in-the-loop gaussian splatting for detailed and efficient sur- face reconstruction.ACM Transactions on Graphics (TOG), 44(6):1–15, 2025. 3, 7, 8

    2025

  16. [16]

    Lin, Marc Van Droogenbroeck, and Andrea Tagliasac- chi

    Jan Held, Sanghyun Son, Renaud Vandeghen, Daniel Re- bain, Matheus Gadelha, Yi Zhou, Anthony Cioppa, Ming C. Lin, Marc Van Droogenbroeck, and Andrea Tagliasac- chi. Meshsplatting: Differentiable rendering with opaque meshes.arXiv, 2025. 3, 7, 8

    2025

  17. [17]

    2d gaussian splatting for geometrically ac- curate radiance fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically ac- curate radiance fields. InACM SIGGRAPH 2024 conference papers, pages 1–11, 2024. 3

    2024

  18. [18]

    Robust image retrieval-based visual localization using kapture, 2022

    Martin Humenberger, Yohann Cabon, Nicolas Guerin, Julien Morat, Vincent Leroy, J´erˆome Revaud, Philippe Rerole, No´e Pion, Cesar de Souza, and Gabriela Csurka. Robust image retrieval-based visual localization using kapture, 2022. 3

    2022

  19. [19]

    Ex- ploring the 15-minutes city concept: Global challenges and opportunities in diverse urban contexts.Urban Science, 9(7): 252, 2025

    Asifa Iqbal, Humaira Nazir, and Ammad Waheed Qazi. Ex- ploring the 15-minutes city concept: Global challenges and opportunities in diverse urban contexts.Urban Science, 9(7): 252, 2025. 2

    2025

  20. [20]

    Horizon- gs: Unified 3d gaussian splatting for large-scale aerial-to- ground scenes

    Lihan Jiang, Kerui Ren, Mulin Yu, Linning Xu, Junting Dong, Tao Lu, Feng Zhao, Dahua Lin, and Bo Dai. Horizon- gs: Unified 3d gaussian splatting for large-scale aerial-to- ground scenes. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 26789–26799, 2025. 3

    2025

  21. [21]

    A hierarchical 3d gaussian representation for real-time ren- dering of very large datasets.ACM Transactions on Graphics (TOG), 43(4):1–15, 2024

    Bernhard Kerbl, Andreas Meuleman, Georgios Kopanas, Michael Wimmer, Alexandre Lanvin, and George Drettakis. A hierarchical 3d gaussian representation for real-time ren- dering of very large datasets.ACM Transactions on Graphics (TOG), 43(4):1–15, 2024. 3

    2024

  22. [22]

    Ground- ing image matching in 3d with mast3r, 2024

    Vincent Leroy, Yohann Cabon, and J´erˆome Revaud. Ground- ing image matching in 3d with mast3r, 2024. 4, 5

    2024

  23. [23]

    arXiv preprint arXiv:2509.18090 (2025)

    Jiahe Li, Jiawei Zhang, Youmin Zhang, Xiao Bai, Jin Zheng, Xiaohan Yu, and Lin Gu. Geosvr: Taming sparse vox- els for geometrically accurate surface reconstruction.arXiv preprint arXiv:2509.18090, 2025. 3

    work page arXiv 2025

  24. [24]

    Vastgaussian: Vast 3d gaussians for large scene reconstruction

    Jiaqi Lin, Zhihao Li, Xiao Tang, Jianzhuang Liu, Shiyong Liu, Jiayue Liu, Yangdi Lu, Xiaofei Wu, Songcen Xu, You- liang Yan, et al. Vastgaussian: Vast 3d gaussians for large scene reconstruction. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 5166–5175, 2024. 2, 3

    2024

  25. [25]

    Capturing, reconstructing, and simulating: the urbanscene3d dataset, 2022

    Liqiang Lin, Yilin Liu, Yue Hu, Xingguang Yan, Ke Xie, and Hui Huang. Capturing, reconstructing, and simulating: the urbanscene3d dataset, 2022. 6, 7

    2022

  26. [26]

    Citygaussian: Real-time high-quality large-scale scene rendering with gaussians

    Yang Liu, Chuanchen Luo, Lue Fan, Naiyan Wang, Jun- ran Peng, and Zhaoxiang Zhang. Citygaussian: Real-time high-quality large-scale scene rendering with gaussians. In European Conference on Computer Vision, pages 265–282. Springer, 2024. 2, 3

    2024

  27. [27]

    , author Luo, C

    Yang Liu, Chuanchen Luo, Zhongkai Mao, Junran Peng, and Zhaoxiang Zhang. Citygaussianv2: Efficient and geometri- cally accurate reconstruction for large-scale scenes.arXiv preprint arXiv:2411.00771, 2024. 1, 2, 3, 6, 7

    work page arXiv 2024

  28. [28]

    3dgsr: Implicit surface reconstruction with 3d gaussian splat- ting.ACM Transactions on Graphics (TOG), 43(6):1–12,

    Xiaoyang Lyu, Yang-Tian Sun, Yi-Hua Huang, Xiuzhe Wu, Ziyi Yang, Yilun Chen, Jiangmiao Pang, and Xiaojuan Qi. 3dgsr: Implicit surface reconstruction with 3d gaussian splat- ting.ACM Transactions on Graphics (TOG), 43(6):1–12,

  29. [29]

    Alexander Mai, Trevor Hedstrom, George Kopanas, Janne Kontkanen, Falko Kuester, and Jonathan T. Barron. Radiance meshes for volumetric reconstruction, 2025. 3, 7, 8

    2025

  30. [30]

    Dinov2: Learning robust visual features with- out supervision, 2024

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mah- moud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Herv ´e Je- gou, Julien Mairal, ...

    2024

  31. [31]

    Revisiting patchmatch multi-view stereo for urban 3d reconstruction

    Marco Orsingher, Paolo Zani, Paolo Medici, and Massimo Bertozzi. Revisiting patchmatch multi-view stereo for urban 3d reconstruction. In2022 IEEE Intelligent Vehicles Sympo- sium (IV), pages 190–196. IEEE, 2022. 2

    2022

  32. [32]

    Continuous remeshing for inverse render- ing.Computer Animation and Virtual Worlds, 33(5):e2101,

    Werner Palfinger. Continuous remeshing for inverse render- ing.Computer Animation and Virtual Worlds, 33(5):e2101,

  33. [33]

    Global Structure-from-Motion Revisited

    Linfei Pan, Daniel Barath, Marc Pollefeys, and Jo- hannes Lutz Sch ¨onberger. Global Structure-from-Motion Revisited. InEuropean Conference on Computer Vision (ECCV), 2024. 2

    2024

  34. [34]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V . Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay. Scikit-learn: Machine learning in Python.Journal of Machine Learning Research, 12:2825–2830, 2011. 3, 7

    2011

  35. [35]

    Learning with average preci- sion: Training image retrieval with a listwise loss, 2019

    Jerome Revaud, Jon Almazan, Rafael Sampaio de Rezende, and Cesar Roberto de Souza. Learning with average preci- sion: Training image retrieval with a listwise loss, 2019. 3

    2019

  36. [36]

    From coarse to fine: Robust hierarchical localization at large scale, 2019

    Paul-Edouard Sarlin, Cesar Cadena, Roland Siegwart, and Marcin Dymczyk. From coarse to fine: Robust hierarchical localization at large scale, 2019. 3

    2019

  37. [37]

    Sch ¨onberger and Jan-Michael Frahm

    Johannes L. Sch ¨onberger and Jan-Michael Frahm. Structure- from-motion revisited. InCVPR, 2016. 2, 4

    2016

  38. [38]

    Sparse voxels rasterization: Real- time high-fidelity radiance field rendering

    Cheng Sun, Jaesung Choe, Charles Loop, Wei-Chiu Ma, and Yu-Chiang Frank Wang. Sparse voxels rasterization: Real- time high-fidelity radiance field rendering. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 16187–16196, 2025. 3

    2025

  39. [39]

    Block-nerf: Scalable large scene neural view synthesis

    Matthew Tancik, Vincent Casser, Xinchen Yan, Sabeek Prad- han, Ben Mildenhall, Pratul P Srinivasan, Jonathan T Barron, and Henrik Kretzschmar. Block-nerf: Scalable large scene neural view synthesis. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 8248–8258, 2022. 2, 3

    2022

  40. [40]

    Mega-nerf: Scalable construction of large- scale nerfs for virtual fly-throughs

    Haithem Turki, Deva Ramanan, and Mahadev Satya- narayanan. Mega-nerf: Scalable construction of large- scale nerfs for virtual fly-throughs. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12922–12931, 2022. 2, 3

    2022

  41. [41]

    Moge: Unlocking accurate monocular geometry estimation for open-domain images with optimal training supervision, 2025

    Ruicheng Wang, Sicheng Xu, Cassie Dai, Jianfeng Xiang, Yu Deng, Xin Tong, and Jiaolong Yang. Moge: Unlocking accurate monocular geometry estimation for open-domain images with optimal training supervision, 2025. 5

    2025

  42. [42]

    SLPA: Uncovering Overlapping Communities in Social Networks via A Speaker-listener Interaction Dynamic Process

    Jierui Xie and Boleslaw K Szymanski. Slpa: Uncov- ering overlapping communities in social networks via a speaker-listener interaction dynamic process.arXiv preprint arXiv:1109.5720, 2011. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  43. [43]

    arXiv preprint arXiv:2401.14032 (2024)

    Butian Xiong, Zhuo Li, and Zhen Li. Gauu-scene: A scene reconstruction benchmark on large scale 3d recon- struction dataset using gaussian splatting.arXiv preprint arXiv:2401.14032, 2024. 7, 8

    work page arXiv 2024

  44. [44]

    Gauu-scene: A scene reconstruction benchmark on large scale 3d reconstruction dataset using gaussian splatting, 2024

    Butian Xiong, Zhuo Li, and Zhen Li. Gauu-scene: A scene reconstruction benchmark on large scale 3d reconstruction dataset using gaussian splatting, 2024. 1, 6

    2024

  45. [45]

    Sa-gs: Semantic-aware gaus- sian splatting for large scene reconstruction with geometry constrain.arXiv preprint arXiv:2405.16923, 2024

    Butian Xiong, Xiaoyu Ye, Tze Ho Elden Tse, Kai Han, Shuguang Cui, and Zhen Li. Sa-gs: Semantic-aware gaus- sian splatting for large scene reconstruction with geometry constrain.arXiv preprint arXiv:2405.16923, 2024. 3

    work page arXiv 2024

  46. [46]

    Scalable and high-quality neural implicit representation for 3d recon- struction.IEEE Transactions on Visualization and Computer Graphics, 2025

    Leyuan Yang, Bailin Deng, and Juyong Zhang. Scalable and high-quality neural implicit representation for 3d recon- struction.IEEE Transactions on Visualization and Computer Graphics, 2025. 3

    2025

  47. [47]

    Gaussian opacity fields: Efficient adaptive surface reconstruction in unbounded scenes.ACM Transactions on Graphics (ToG), 43(6):1–13, 2024

    Zehao Yu, Torsten Sattler, and Andreas Geiger. Gaussian opacity fields: Efficient adaptive surface reconstruction in unbounded scenes.ACM Transactions on Graphics (ToG), 43(6):1–13, 2024. 3

    2024

  48. [48]

    Supernerf: High-precision 3d reconstruction for large-scale scenes.IEEE Transactions on Geoscience and Remote Sens- ing, 2024

    Guangyun Zhang, Chaozhong Xue, and Rongting Zhang. Supernerf: High-precision 3d reconstruction for large-scale scenes.IEEE Transactions on Geoscience and Remote Sens- ing, 2024. 2, 3

    2024

  49. [49]

    Very large-scale global sfm by distributed motion averaging

    Siyu Zhu, Runze Zhang, Lei Zhou, Tianwei Shen, Tian Fang, Ping Tan, and Long Quan. Very large-scale global sfm by distributed motion averaging. InProceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 4568–4577, 2018. 2 City-Mesh3R: Simulation-Ready City-Scale 3D Mesh Reconstruction from Multi-View Images Supplementary Mate...

    2018

  50. [50]

    Support-plane parameterization and regular partition construction This subsection provides the implementation details omit- ted from the main paper (Sec

    Area partitioning of large-scale sparse SfM 6.1. Support-plane parameterization and regular partition construction This subsection provides the implementation details omit- ted from the main paper (Sec. 3.2) for constructing the spa- tial partitions used before camera selection. Dominant-support-plane parameterization.LetP= {Pp}denote the sparse 3D points...

  51. [51]

    For each partitioned area, we start from its sparse SfM reconstruction and retain the top-Mcameras from the rank- ing stage

    Dense Reconstruction and Surface Initial- ization This section provides the detailed formulation for the dense reconstruction and surface initialization module summa- rized in Section 3.2. For each partitioned area, we start from its sparse SfM reconstruction and retain the top-Mcameras from the rank- ing stage. Their intrinsics and poses,{K n, Rn, tn}M n...

  52. [52]

    Differentiable Rendering based Mesh Re- finement This section provides the full formulation and implementa- tion details omitted from Sec. 3.2. 8.1. Rendering Objective At iterationk, the current mesh isM k = (V k, F k). Given calibrated cameras{K j, Twc,j }N j=1, we optimize Φ(M k) = NX j=1 λn L(j) n (M k) +λ s L(j) sil (M k) +R(M k), (34) whereRis a lig...

  53. [53]

    Aspect Ratio (AR).For a triangular facef∈ F, leth f denote its longest edge length andr f its inradius

    Mesh Quality Metrics We assess mesh quality using the following metrics, which quantify complementary aspects of geometric validity, topological consistency, and structural regularity. Aspect Ratio (AR).For a triangular facef∈ F, leth f denote its longest edge length andr f its inradius. The face- wise aspect ratio is defined as AR(f) = hf 2rf ,(51) Metri...

    work page arXiv

  54. [54]

    Limitations Although our pipeline is highly scalable and produces clean, simulation-ready meshes, a few practical constraints re- main. Our method inherits any residual errors from up- stream SfM or pretrained depth prediction models, and ex- tremely texture-poor or reflective regions can still challenge local reconstruction quality. Seam merging is robus...