Pith. sign in

REVIEW 3 major objections 5 minor 65 references

Grouping city-scale camera views by orientation, not space, plus filling sparse point-cloud holes yields more accurate large-scene surfaces than prior 3DGS methods.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review

2026-07-12 00:03 UTC pith:NUF67U2M

load-bearing objection Orientation-first view clustering is a real, simple fix for large-scale 3DGS surfaces; gains hold even when thresholds move and completion is ablated. the 3 major comments →

arxiv 2607.03771 v1 pith:NUF67U2M submitted 2026-07-04 cs.CV

City-Level 3D Surface Reconstruction with Viewpoint Orientation Partitioning and Scene Completion

Liang Han , Wenyuan Zhang , Junsheng Zhou , Yu-Shen Liu , Zhizhong Han This is my paper
classification cs.CV
keywords 3D surface reconstructionlarge-scale scenes3D Gaussian Splattingviewpoint orientation partitioningpoint-cloud completionmulti-view consistencycity-scale geometry
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Large outdoor and urban scenes still produce coarse or incomplete surfaces under 3D Gaussian Splatting because spatial blocks mix opposing camera directions and because Structure-from-Motion leaves holes in poorly textured or sparsely observed regions. This paper claims that clustering cameras first by viewing direction (cosine distance) and then by position produces viewpoint-consistent groups that share high overlap, so multi-view photometric and geometric constraints recover accurate depth. A second, selective completion step detects images whose projected keypoints have low spatial entropy and densifies only those holes with a pretrained matcher. The resulting pipeline runs in parallel across ordinary GPUs and reports higher surface F1 scores than recent city-scale Gaussian methods on three public benchmarks. A reader who needs reliable city meshes for mapping or simulation cares because the gains come from a simple reordering of existing views rather than heavier models or denser captures.

Core claim

Viewpoint-orientation partitioning—DBSCAN on camera directions followed by position K-means that caps each group at 200 cameras—plus entropy-guided completion of COLMAP holes produces higher-fidelity surfaces than spatial-block 3DGS baselines, because views with similar orientations jointly constrain depth more reliably than views that merely lie near one another in space.

What carries the argument

Viewpoint Orientation Partitioning (VOP): two-stage clustering that first groups cameras by cosine distance of viewing directions (threshold ≈45°) then splits oversized groups by 3D position; source-view refinement and multi-view consistency losses are applied only inside each orientation-consistent group.

Load-bearing premise

A single fixed orientation threshold of roughly 45 degrees and a single entropy threshold of 0.7 on COLMAP keypoints correctly identify both the best view groups and the true geometric holes across every tested city-scale scene.

What would settle it

Re-run the identical training schedule on GauU-Scene and MatrixCity while sweeping the orientation threshold away from 45° (or the entropy threshold away from 0.7); if mean F1 falls below the CityGS-X baseline for most settings, the claim that these fixed thresholds are reliable collapses.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes VOP-GS for city-level multi-view surface reconstruction with 3D Gaussian Splatting. It partitions input views by camera orientation (DBSCAN on cosine distance of viewing directions, Eq. 1, with ε_dir=0.71) followed by position-based K-means splitting when clusters exceed M_max=200, then refines groups by adding favorable source views. It further detects holes in the COLMAP sparse point cloud via normalized Shannon entropy of projected keypoints on an N imes N grid (Eqs. 3–4, τ=0.7) and completes them by triangulating dense matches from a pretrained network (MINIMA). Each group is optimized independently with multi-view photometric/geometric losses from PGSR plus standard RGB and depth-normal terms (Eq. 5). Experiments on GauU-Scene, MatrixCity, and UrbanScene3D claim higher F1 surface accuracy than CityGaussian-V2 and CityGS-X, with competitive multi-GPU training time.

Significance. If the reported F1 gains hold under fair evaluation, the work provides a simple, geometry-oriented alternative to purely spatial block partitioning for large-scale 3DGS surface reconstruction. Orientation clustering directly targets the multi-view overlap that prior surface methods have shown improves depth estimation, while selective (not global) point-cloud completion avoids the cost of full MVS. Consistent quantitative improvements across three datasets, supported by ablations on angular threshold (Table 4) and entropy τ (Table 5), plus efficiency numbers (Table 3), make the approach practically useful for multi-GPU city-scale pipelines. The method is largely engineering but addresses a genuine gap: most large-scale 3DGS systems optimize for novel-view synthesis rather than metric surfaces.

major comments (3)
  1. [§4.4 / Table 1] §4.4 and Table 1: Quantitative claims rest on cropped central regions of the GauU-Scene LiDAR GT (following CityGaussian-V2) because edge regions are incomplete. This protocol is not neutral; completion of the initialization point cloud can occur outside the retained crop, and the paper itself notes that CityGaussian-V2 can look more complete yet score lower. Report uncropped F1 (or a fixed, publicly documented crop mask) so that completeness and accuracy can be assessed independently.
  2. [§4.4 / Table 2] §4.4 and Table 2: The CityGS-X numbers obtained by the authors (F1 0.527 on GauU-Scene, 0.581 on MatrixCity) diverge from those in the original CityGS-X paper, and the authors remark that the meshes “do not match those shown in its original paper.” Without a verified re-implementation or shared evaluation script, the SOTA ranking is not fully reliable. Either release the exact evaluation code/masks used for all baselines or re-run CityGS-X under identical conditions and document any hyper-parameter differences.
  3. [§3.1] §3.1 (Sub-cluster Refinement): The set S(v) of source views added to each group is described only as “favorable for geometry reconstruction”; later the implementation simply states “We select 4 source views.” The precise selection criterion (angular threshold, overlap score, COLMAP co-visibility, etc.) is load-bearing for the claim that orientation partitioning improves multi-view consistency, yet is not specified. Provide the exact algorithm so that the partitioning can be reproduced.
minor comments (5)
  1. [Fig. 4] Fig. 4 caption and surrounding text: the visual comparison of partitions is helpful, but the dashed lines of CityGaussian-V2 are hard to distinguish in grayscale; add a legend or color-code the blocks more clearly.
  2. [§3.2 / Fig. 8] §3.2 Hole Detection: the entropy threshold τ=0.7 is justified by the distribution in Fig. 8, yet the figure itself is never referenced in the main text of the ablation; a short pointer would help readers.
  3. [Table 3] Table 3: training times are given only as totals; reporting per-group wall-clock and peak GPU memory would make the efficiency claim more informative.
  4. [§4.4 / Fig. 7] UrbanScene3D results (Fig. 7) are purely qualitative; even a simple Chamfer or visual completeness score against the provided meshes would strengthen the multi-dataset claim.
  5. [throughout] Minor typos: “donates” → “denotes” (§3.1); “theshold” → “threshold” (Table 5 caption); inconsistent hyphenation of “point cloud” / “point-cloud”.

Circularity Check

0 steps flagged

No significant circularity: empirical partitioning + completion method evaluated against external LiDAR/GT point clouds; hyperparameters chosen by ablation, not by fitting the reported F1.

full rationale

The paper's load-bearing claims are (1) that viewpoint-orientation clustering (cosine DBSCAN + position K-means) plus selective hole completion produces higher surface F1 than CityGaussian-V2 / CityGS-X, and (2) that the multi-view photometric/geometric losses improve local coherence. Both are supported by direct quantitative comparison on held-out external ground-truth point clouds (GauU-Scene LiDAR, MatrixCity GT) and by ablations that vary the free parameters (angular threshold, entropy threshold au) rather than fitting them to the final metric. No equation defines a quantity in terms of the quantity it is later said to predict; no uniqueness theorem is imported from the authors' prior work to forbid alternatives; the cited multi-view consistency losses are taken from PGSR (an independent baseline) and applied inside each orientation group. Self-citations that appear (e.g., SparseRecon) are used only as supporting literature for the general observation that high-overlap views aid geometry, not as load-bearing premises that force the reported F1 numbers. The derivation chain is therefore self-contained against external benchmarks and contains no circular reduction.

Axiom & Free-Parameter Ledger

4 free parameters · 3 axioms · 0 invented entities

The method rests on a handful of hand-chosen clustering and entropy thresholds plus standard multi-view consistency losses taken from PGSR; no new physical entities are postulated.

free parameters (4)
  • ε_dir (orientation cosine threshold) = 0.71
    Set to 0.71 (45°) by ablation; controls the first-stage DBSCAN clusters.
  • M_max (max cameras per sub-cluster) = 200
    Hard limit of 200 used for K-means splitting; balances GPU load.
  • τ (normalized entropy threshold for hole detection) = 0.7
    Set to 0.7 after inspecting entropy histograms; decides which images trigger densification.
  • λ1, λ2, λ3 (loss weights) = 0.15, 0.03, 0.05
    Fixed to 0.15 / 0.03 / 0.05 following PGSR-style tuning.
axioms (3)
  • domain assumption Views with small angular difference and high overlap produce more accurate multi-view stereo depth estimates.
    Stated as key insight in Introduction and used to justify orientation clustering; supported by citations to classical MVS literature.
  • ad hoc to paper COLMAP 2-D keypoints that are non-uniformly distributed indicate geometric holes in the sparse point cloud.
    Introduced in Section 3.2; entropy on an N×N grid is the operational definition.
  • domain assumption A pretrained feature-matching network (MINIMA) yields geometrically consistent dense points that can be safely concatenated with SfM points.
    Assumed after geometric-consistency filtering; no independent verification of the matcher’s accuracy on the target city scenes is provided.

reviewed 2026-07-12 · how reviews work

0 comments
Cite this review

Pith. "Pith review of City-Level 3D Surface Reconstruction with Viewpoint Orientation Partitioning and Scene Completion." pith.science (2026). https://pith.science/paper/NUF67U2M

@misc{pith2026260703771,
  author       = {Pith},
  title        = {Pith review of: City-Level 3D Surface Reconstruction with Viewpoint Orientation Partitioning and Scene Completion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NUF67U2M}},
  note         = {Machine review of arXiv:2607.03771}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Multi-view 3D surface reconstruction is a longstanding challenge in computer vision. Although recent large-scale reconstruction methods based on 3D Gaussian Splatting (3DGS) achieve impressive novel-view synthesis, producing high-quality surfaces over large scenes remains difficult, due to complex geometry, long optimization, and limited memory. In this paper, we propose a novel yet simple partitioning method to efficiently and faithfully reconstruct large-scale scene surfaces. Our key insight lies in a scene partitioning method based on viewpoint orientation. This partitioning approach ensures that views with similar orientations are jointly involved for more accurate depth estimations, leading to precise surface reconstructions and balanced computation on multiple GPUs in parallel. In addition, we propose a strategy to detect and repair missing regions in the initial point cloud caused by sparse viewpoints or insufficient textures, thereby further improving the geometric quality. Extensive experiments on the GauU-Scene, MatrixCity, and UrbanScene3D datasets demonstrate that our method outperforms the state-of-the-art approaches in surface reconstruction for large-scale scenes. Project page: https://hanl2010.github.io/VOP-GS.

Figures

Figures reproduced from arXiv: 2607.03771 by Junsheng Zhou, Liang Han, Wenyuan Zhang, Yu-Shen Liu, Zhizhong Han.

Figure 1
Figure 1. Figure 1: Comparison of reconstruction on the MatrixCity dataset. Compared with state￾of-the-art large-scale reconstruction methods CityGaussian-V2 [31] and CityGS-X [10], Our method achieves more faithful surface reconstruction. Abstract. Multi-view 3D surface reconstruction is a longstanding chal￾lenge in computer vision. Although recent large-scale reconstruction meth￾ods based on 3D Gaussian Splatting (3DGS) ach… view at source ↗
Figure 2
Figure 2. Figure 2: Visual comparison of the point clouds (PC) and corresponding meshes before and after completion. The left column shows the reference image, the middle part presents the COLMAP point cloud and its corresponding mesh reconstruction, and the right part presents the completed point cloud and the corresponding reconstructed mesh. to their camera orientations and positions. Our method produces an optimal partiti… view at source ↗
Figure 3
Figure 3. Figure 3: The overview of our method. (a) We group all input views based on their camera orientations and positions, and we train 3DGS with multi-view consistency constraints in each group. (b) We detect regions with holes in the point cloud by analyzing the uniformity of valid keypoints, where the red points in the image indicate the valid keypoints. The corresponding images are then fed into a pretrained matching … view at source ↗
Figure 4
Figure 4. Figure 4: Visual comparison of scene partitioning in our method and CityGaussian￾V2. The dashed lines in the left figure indicate the nine spatial partitions used in CityGaussian-V2. Cameras belonging to the same group are shown in the same color. The difference is highlighted. ing to ensure that all views within each group have sufficient overlap for reliable geometry optimization, especially for views located near… view at source ↗
Figure 5
Figure 5. Figure 5: Visual comparison of reconstruction on the GauU-Scene dataset. 3.3 Training Loss Our loss function consists of the multi-view photometric consistency loss Lpatch and multi-view geometric consistency loss Lgeo from PGSR [2], as well as the color reconstruction loss Lrgb and depth-normal consistency loss Lnormal com￾monly used in 3DGS-based surface reconstruction methods [2, 15, 59]. The final loss function … view at source ↗
Figure 6
Figure 6. Figure 6: Visual comparison of reconstruction on the MatrixCity dataset. Failure ResidenceSci-Art PGSR 2DGS CityGaussian-V2 CityGS-X Ours [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visual comparison of reconstruction on the UrbanScene3D dataset. achieves the best performance on the HAV scene, which contains only 424 images and represents a relatively small-scale scene. This indicates that the non-block optimization strategy of PGSR is more advantageous for reconstructing compact scenes [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Distribution of normalized entropy on the Gau-Uscene and MatrixCity Datasets [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visualization of reconstruction with varying entropy threshold τ values. We then gradually increase the entropy threshold τ starting from 0.6 and perform reconstruction and quantitative evaluation for scenes in GauU-Scene and MatrixCity, with results summarized in Tab. 5. The results show that, for both GauU-Scene and MatrixCity, reconstruction quality progressively improves as the entropy threshold τ incr… view at source ↗

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

65 extracted references · 4 linked inside Pith

  1. [1]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Barron, J.T., Mildenhall, B., Verbin, D., Srinivasan, P.P., Hedman, P.: Mip- NeRF 360: Unbounded anti-aliased neural radiance fields. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 5470–5479 (2022) 2, 4

    2022

  2. [2]

    IEEE Transactions on Visualization and Computer Graph- ics31, 1–12 (01 2024) 4, 9, 10, 11, 14

    Chen, D., Li, H., Ye, W., Wang, Y., Xie, W., Zhai, S., Wang, N., Liu, H., Bao, H., Zhang, G.: PGSR: Planar-based gaussian splatting for efficient and high-fidelity surface reconstruction. IEEE Transactions on Visualization and Computer Graph- ics31, 1–12 (01 2024) 4, 9, 10, 11, 14

    2024

  3. [3]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Chen, J., Ye, W., Wang, Y., Chen, D., Huang, D., Ouyang, W., Zhang, G., Qiao, Y., He, T.: GigaGS: 3D gaussian based planar representation for large-scene surface reconstruction. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 39, pp. 2088–2096 (2025) 5 16 L. Han et al

    2088

  4. [4]

    Advances in Neural Information Pro- cessing Systems37, 34487–34512 (2024) 2, 5

    Chen, Y., Lee, G.H.: DoGS: Distributed-oriented gaussian splatting for large-scale 3D reconstruction via gaussian consensus. Advances in Neural Information Pro- cessing Systems37, 34487–34512 (2024) 2, 5

    2024

  5. [5]

    In: ACM SIGGRAPH 2024 Conference Proceed- ings (2024) 4

    Dai, P., Xu, J., Xie, W., Liu, X., Wang, H., Xu, W.: High-quality surface recon- struction using gaussian surfels. In: ACM SIGGRAPH 2024 Conference Proceed- ings (2024) 4

    2024

  6. [6]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Darmon, F., Bascle, B., Devaux, J.C., Monasse, P., Aubry, M.: Improving neural implicit surfaces geometry with patch warping. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 6260–6269 (2022) 2

    2022

  7. [7]

    Ester, M., Kriegel, H.P., Sander, J., Xu, X.: Density-based spatial clustering of ap- plications with noise. In: Int. Conf. knowledge discovery and data mining. vol. 240 (1996) 6

    1996

  8. [8]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Fan, J., Li, W., Han, Y., Dai, T., Tang, Y.: Momentum-GS: Momentum gaussian self-distillation for high-quality large scene reconstruction. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 25250–25260 (2025) 5

    2025

  9. [9]

    In: Proceedings of the Computer Vision and Pattern Recognition Con- ference

    Feng,G.,Chen,S.,Fu,R.,Liao,Z.,Wang,Y.,Liu,T.,Hu,B.,Xu,L.,Pei,Z.,Li,H., et al.: FlashGS: Efficient 3D gaussian splatting for large-scale and high-resolution rendering. In: Proceedings of the Computer Vision and Pattern Recognition Con- ference. pp. 26652–26662 (2025) 5

    2025

  10. [10]

    In: Proceedings of the IEEE/CVF International Conference on Com- puter Vision

    Gao,Y.,Li,H.,Chen,J.,Zou,Z.,Zhong,Z.,Zhang,D.,Sun,X.,Han,J.:CityGS-X: A scalable architecture for efficient and geometrically accurate large-scale scene re- construction. In: Proceedings of the IEEE/CVF International Conference on Com- puter Vision. pp. 27187–27196 (2025) 1, 5, 10, 11, 12, 14

    2025

  11. [11]

    In: Proceed- ings of the Computer Vision and Pattern Recognition Conference

    Guédon, A., Ichikawa, T., Yamashita, K., Nishino, K.: MAtCha Gaussians: Atlas of charts for high-quality geometry and photorealism from sparse views. In: Proceed- ings of the Computer Vision and Pattern Recognition Conference. pp. 6001–6011 (2025) 7

    2025

  12. [12]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Guédon, A., Lepetit, V.: SuGaR: Surface-aligned gaussian splatting for efficient 3D mesh reconstruction and high-quality mesh rendering. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 5354– 5363 (2024) 4

    2024

  13. [13]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (2025) 2

    Han, L., Zhang, X., Song, H., Shi, K., Liu, Y.S., Han, Z.: SparseRecon: Neural implicit surface reconstruction from sparse views with feature and depth consis- tencies. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (2025) 2

    2025

  14. [14]

    Advances in Neural Information Pro- cessing Systems37, 68595–68621 (2024) 7

    Han, L., Zhou, J., Liu, Y.S., Han, Z.: Binocular-guided 3D gaussian splatting with view consistency for sparse view synthesis. Advances in Neural Information Pro- cessing Systems37, 68595–68621 (2024) 7

    2024

  15. [15]

    In: ACM SIGGRAPH 2024 Conference Proceedings (2024) 4, 9, 10, 11, 14

    Huang, B., Yu, Z., Chen, A., Geiger, A., Gao, S.: 2D gaussian splatting for geomet- rically accurate radiance fields. In: ACM SIGGRAPH 2024 Conference Proceedings (2024) 4, 9, 10, 11, 14

    2024

  16. [16]

    In: Proceedings of the AAAI Conference on Artificial Intelli- gence (2025) 2

    Huang, H., Wu, Y., Deng, C., Gao, G., Gu, M., Liu, Y.S.: FatesGS: Fast and accurate sparse-view surface reconstruction using gaussian splatting with depth- feature consistency. In: Proceedings of the AAAI Conference on Artificial Intelli- gence (2025) 2

    2025

  17. [17]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Huang, H., Wu, Y., Zhou, J., Gao, G., Gu, M., Liu, Y.S.: NeuSurf: On-surface priors for neural surface reconstruction from sparse input views. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 38, pp. 2312–2320 (2024) 2 City-Level 3D Surface Reconstruction 17

    2024

  18. [18]

    ACM Transactions on Graphics42(4), 1–14 (2023) 2, 4

    Kerbl, B., Kopanas, G., Leimkühler, T., Drettakis, G.: 3D gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics42(4), 1–14 (2023) 2, 4

    2023

  19. [19]

    In: European Conference on Computer Vision

    Leroy, V., Cabon, Y., Revaud, J.: Grounding image matching in 3D with MASt3R. In: European Conference on Computer Vision. pp. 71–91. Springer (2024) 7

    2024

  20. [20]

    Computational Visual Media (2026) 4

    Li,J.,Pang,B.,Wang,P.S.:Jointpointcloudupsamplingandcleaningwithoctree- based cnns. Computational Visual Media (2026) 4

    2026

  21. [21]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Li, Q., Feng, H., Gong, X., Liu, Y.S.: Learning normals of noisy points by local gradient-aware surface filtering. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 28828–28838 (2025) 4

    2025

  22. [22]

    Advances in Neural Information Processing Systems38, 173867–173886 (2026) 4

    Li, Q., Feng, H., Gong, X., Liu, Y.S.: VA-GS: Enhancing the geometric represen- tation of gaussian splatting via view alignment. Advances in Neural Information Processing Systems38, 173867–173886 (2026) 4

    2026

  23. [23]

    IEEE Transactions on Visualization and Computer Graphics (2025) 4

    Li, Q., Feng, H., Shi, K., Gao, Y., Fang, Y., Liu, Y.S., Han, Z.: PFF-Net: Patch fea- ture fitting for point cloud normal estimation. IEEE Transactions on Visualization and Computer Graphics (2025) 4

    2025

  24. [24]

    Li, S., Liu, Y.S., Han, Z.: GaussianUDF: Inferring unsigned distance functions through3Dgaussiansplatting.In:ProceedingsoftheComputerVisionandPattern Recognition Conference. pp. 27113–27123 (2025) 4

    2025

  25. [25]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Li, Y., Jiang, L., Xu, L., Xiangli, Y., Wang, Z., Lin, D., Dai, B.: MatrixCity: A large-scale city dataset for city-scale neural rendering and beyond. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 3205–3215 (2023) 9, 11, 12, 13

    2023

  26. [26]

    In: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Lin, J., Li, Z., Tang, X., Liu, J., Liu, S., Liu, J., Lu, Y., Wu, X., Xu, S., Yan, Y., et al.: VastGaussian: Vast 3D gaussians for large scene reconstruction. In: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 5166–5175 (2024) 2, 4

    2024

  27. [27]

    In: European Conference on Computer Vision

    Lin, L., Liu, Y., Hu, Y., Yan, X., Xie, K., Huang, H.: Capturing, reconstructing, and simulating: the UrbanScene3D dataset. In: European Conference on Computer Vision. pp. 93–109. Springer (2022) 9, 11

    2022

  28. [28]

    In: Proceedings of the IEEE/CVF International Conference on Com- puter Vision

    Liu, S., Tang, X., Li, Z., He, Y., Ye, C., Liu, J., Huang, B., Zhou, S., Wu, X.: Oc- cluGaussian: Occlusion-aware gaussian splatting for large scene reconstruction and rendering. In: Proceedings of the IEEE/CVF International Conference on Com- puter Vision. pp. 26643–26652 (2025) 4

    2025

  29. [29]

    IEEE MultiMedia (2025) 5

    Liu, W., Guan, T., Zhu, B., Xu, L., Song, Z., Li, D., Wang, Y., Yang, W.: Effi- cientGS: Streamlining gaussian splatting for large-scale high-resolution scene rep- resentation. IEEE MultiMedia (2025) 5

    2025

  30. [30]

    In: European Conference on Computer Vision

    Liu, Y., Luo, C., Fan, L., Wang, N., Peng, J., Zhang, Z.: CityGaussian: Real-time high-quality large-scale scene rendering with gaussians. In: European Conference on Computer Vision. pp. 265–282. Springer (2024) 2, 5, 10

    2024

  31. [31]

    In: International Con- ference on Learning Representations

    Liu, Y., Luo, C., Mao, Z., Peng, J., Zhang, Z.: CityGaussianV2: Efficient and geometrically accurate reconstruction for large-scale scenes. In: International Con- ference on Learning Representations. vol. 2025, pp. 84324–84341 (2025) 1, 5, 7, 10, 11, 12, 14

    2025

  32. [32]

    ACM Trans- actions on Graphics (TOG)43(6), 1–12 (2024) 4

    Lyu, X., Sun, Y.T., Huang, Y.H., Wu, X., Yang, Z., Chen, Y., Pang, J., Qi, X.: 3DGSR: Implicit surface reconstruction with 3D gaussian splatting. ACM Trans- actions on Graphics (TOG)43(6), 1–12 (2024) 4

    2024

  33. [33]

    Com- munications of the ACM65(1), 99–106 (2021) 2, 4 18 L

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: NeRF: Representing scenes as neural radiance fields for view synthesis. Com- munications of the ACM65(1), 99–106 (2021) 2, 4 18 L. Han et al

    2021

  34. [34]

    ACM transactions on graphics (TOG)41(4), 1–15 (2022) 2, 4

    Müller,T.,Evans,A.,Schied,C.,Keller,A.:Instantneuralgraphicsprimitiveswith a multiresolution hash encoding. ACM transactions on graphics (TOG)41(4), 1–15 (2022) 2, 4

    2022

  35. [35]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Na,Y.,Kim,W.J.,Han,K.B.,Ha,S.,Yoon,S.E.:UFORecon:Generalizablesparse- view surface reconstruction from arbitrary and unfavorable sets. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 5094–5104 (2024) 2

    2024

  36. [36]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Ren,J.,Jiang,X.,Li,Z.,Liang,D.,Zhou,X.,Bai,X.:MINIMA:Modalityinvariant image matching. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 23059–23068 (2025) 3, 8

    2025

  37. [37]

    arXiv preprint arXiv:2403.17898 (2024) 4

    Ren, K., Jiang, L., Lu, T., Yu, M., Xu, L., Ni, Z., Dai, B.: Octree-GS: Towards consistent real-time rendering with lod-structured 3D gaussians. arXiv preprint arXiv:2403.17898 (2024) 4

    Pith/arXiv arXiv 2024

  38. [38]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Schonberger, J.L., Frahm, J.M.: Structure-from-motion revisited. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 4104–4113 (2016) 3

    2016

  39. [39]

    In: European Conference on Computer Vision (2016) 3, 7

    Schönberger, J.L., Zheng, E., Pollefeys, M., Frahm, J.M.: Pixelwise view selection for unstructured multi-view stereo. In: European Conference on Computer Vision (2016) 3, 7

    2016

  40. [40]

    Computational Visual Me- dia (2025) 4

    Shao, D., Jing, Y., Zhao, X., Mao, S., Lyu, L., Liu, X., Lu, X.: DS-MAE: Dual- siamese masked autoencoders for point cloud analysis. Computational Visual Me- dia (2025) 4

    2025

  41. [41]

    In: 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)

    Shao, M., Xiong, F., Zhang, H., Yang, S., Xu, M., Bian, W., Wang, X.: Global-guided focal neural radiance field for large-scale scene rendering. In: 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV). pp. 2558–2567. IEEE (2025) 4

    2025

  42. [42]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Su, M., Wang, Z., Au, H., Li, Y., Cao, X., Pan, C., Chen, Y., Wang, G.: HUG: Hierarchical urban gaussian splatting with block-based reconstruction for large- scale aerial scenes. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 28839–28848 (2025) 4

    2025

  43. [43]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Tancik, M., Casser, V., Yan, X., Pradhan, S., Mildenhall, B., Srinivasan, P.P., Bar- ron, J.T., Kretzschmar, H.: Block-NeRF: Scalable large scene neural view synthe- sis. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 8248–8258 (2022) 2, 4

    2022

  44. [44]

    In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition

    Turki, H., Ramanan, D., Satyanarayanan, M.: Mega-NeRF: Scalable construction of large-scale nerfs for virtual fly-throughs. In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition. pp. 12922–12931 (2022) 4

    2022

  45. [45]

    Advances in Neural Information Processing Systems (2021) 2, 4

    Wang, P., Liu, L., Liu, Y., Theobalt, C., Komura, T., Wang, W.: NeuS: Learn- ing neural implicit surfaces by volume rendering for multi-view reconstruction. Advances in Neural Information Processing Systems (2021) 2, 4

    2021

  46. [46]

    In: Proceedings of the 32nd ACM International Conference on Multimedia

    Wang, Y., Zhou, K., Zhang, W., Xiao, C.: MegaSurf: Scalable large scene neural surface reconstruction. In: Proceedings of the 32nd ACM International Conference on Multimedia. pp. 6414–6423 (2024) 2, 5

    2024

  47. [47]

    arXiv preprint arXiv:2405.16829 (2024) 4

    Wang, Z., Xu, D.: PyGS: Large-scale scene representation with pyramidal 3D gaus- sian splatting. arXiv preprint arXiv:2405.16829 (2024) 4

    Pith/arXiv arXiv 2024

  48. [48]

    IEEE Transactions on Pattern Analysis and Machine Intelligence45(1), 852–867 (2022) 4 City-Level 3D Surface Reconstruction 19

    Wen, X., Xiang, P., Han, Z., Cao, Y.P., Wan, P., Zheng, W., Liu, Y.S.: PMP- Net++: Point cloud completion by transformer-enhanced multi-step point moving paths. IEEE Transactions on Pattern Analysis and Machine Intelligence45(1), 852–867 (2022) 4 City-Level 3D Surface Reconstruction 19

    2022

  49. [49]

    arXiv preprint arXiv:2504.09048 (2025) 2, 4

    Wu, Y., Qi, Z., Shi, Z., Zou, Z.: BlockGaussian: Efficient large-scale scene novel view synthesis via adaptive block-based gaussian splatting. arXiv preprint arXiv:2504.09048 (2025) 2, 4

    Pith/arXiv arXiv 2025

  50. [50]

    IEEE Transactions on Pattern Analysis and Machine Intelligence 45(5), 6320–6338 (2022) 4

    Xiang, P., Wen, X., Liu, Y.S., Cao, Y.P., Wan, P., Zheng, W., Han, Z.: Snowflake point deconvolution for point cloud completion and generation with skip-transformer. IEEE Transactions on Pattern Analysis and Machine Intelligence 45(5), 6320–6338 (2022) 4

    2022

  51. [51]

    arXiv preprint arXiv:2401.14032 (2024) 9, 10, 11, 12, 13, 14

    Xiong, B., Li, Z., Li, Z.: Gauu-Scene: A scene reconstruction benchmark on large scale 3D reconstruction dataset using gaussian splatting. arXiv preprint arXiv:2401.14032 (2024) 9, 10, 11, 12, 13, 14

    Pith/arXiv arXiv 2024

  52. [52]

    Computational Visual Media10(1), 27–43 (2024) 4

    Xu, M., Zhou, Z., Wang, Y., Qiao, Y.: Towards robustness and generalization of point cloud representation: A geometry coding method and a large-scale object- level dataset. Computational Visual Media10(1), 27–43 (2024) 4

    2024

  53. [53]

    In: European Conference on Computer Vision

    Xu, W., Gao, H., Shen, S., Peng, R., Jiao, J., Wang, R.: MVPGS: Excavating multi-view priors for gaussian splatting from sparse input views. In: European Conference on Computer Vision. pp. 203–220. Springer (2024) 7

    2024

  54. [54]

    IEEE Transactions on Visualization and Computer Graphics (2025) 5

    Yang, L., Deng, B., Zhang, J.: Scalable and high-quality neural implicit represen- tation for 3D reconstruction. IEEE Transactions on Visualization and Computer Graphics (2025) 5

    2025

  55. [55]

    Computational Visual Media (2025) 4

    Yang, Y.Q., Guo, Y.X., Liu, Y.: Swin3D++: Effective multi-source pretraining for 3D indoor scene understanding. Computational Visual Media (2025) 4

    2025

  56. [56]

    In: Proceedings of the European conference on computer vision (ECCV)

    Yao, Y., Luo, Z., Li, S., Fang, T., Quan, L.: MVSNet: Depth inference for unstruc- tured multi-view stereo. In: Proceedings of the European conference on computer vision (ECCV). pp. 767–783 (2018) 2, 7

    2018

  57. [57]

    Advancesin Neural Information ProcessingSystems34,4805–4815(2021) 2, 4

    Yariv, L., Gu, J., Kasten, Y., Lipman, Y.: Volume rendering of neural implicit surfaces. Advancesin Neural Information ProcessingSystems34,4805–4815(2021) 2, 4

    2021

  58. [58]

    Circuits, systems and sig- nal processing28, 819–843 (2009) 9

    Yoo, J.C., Han, T.H.: Fast normalized cross-correlation. Circuits, systems and sig- nal processing28, 819–843 (2009) 9

    2009

  59. [59]

    ACM Transactions on Graphics43(6), 1–13 (2024) 4, 9

    Yu, Z., Sattler, T., Geiger, A.: Gaussian Opacity Fields: Efficient adaptive surface reconstruction in unbounded scenes. ACM Transactions on Graphics43(6), 1–13 (2024) 4, 9

    2024

  60. [60]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Yuan, Z., Huang, H., Xiong, Z., Wang, D., Yang, G.: Robust and efficient 3D gaus- sian splatting for urban scene reconstruction. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 26209–26219 (2025) 5

    2025

  61. [61]

    Scientific Reports15(1), 37295 (2025) 4

    Zhang, H., Xue, Y., Liao, M., Lao, Y.: BirdNeRF: fast neural reconstruction of large-scale scenes from aerial imagery. Scientific Reports15(1), 37295 (2025) 4

    2025

  62. [62]

    Computational Visual Media (2025) 4

    Zhang, J., Zhang, J., Ding, D., Ma, Z.: ARNet: Attribute artifact reduction for G-PCC compressed point clouds. Computational Visual Media (2025) 4

    2025

  63. [63]

    In: Pro- ceedings of the Computer Vision and Pattern Recognition Conference

    Zhang, W., Jia, E.Y.t., Zhou, J., Ma, B., Shi, K., Liu, Y.S., Han, Z.: NeRFPrior: Learning neural radiance field as a prior for indoor scene reconstruction. In: Pro- ceedings of the Computer Vision and Pattern Recognition Conference. pp. 11317– 11327 (2025) 2

    2025

  64. [64]

    IEEE Transactions on Pattern Analysis and Machine Intelligence (2026) 2

    Zhang, W., Wang, C., Shi, K., Liu, Y.S., Han, Z.: VRP-UDF: Towards unbiased learning of unsigned distance functions from multi-view images with volume ren- dering priors. IEEE Transactions on Pattern Analysis and Machine Intelligence (2026) 2

    2026

  65. [65]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Zhang, W., Yang, Y., Huang, H., Han, L., Shi, K., Liu, Y.S., Han, Z.: MonoIn- stance: Enhancing monocular priors via multi-view instance alignment for neural rendering and reconstruction. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 21642–21653 (2025) 2

    2025

This paper was first reviewed by grok-4.5 on July 12, 2026.