Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

CoSurfGS:Collaborative 3D Surface Gaussian Splatting with Distributed Learning for Large Scene Reconstruction

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read This paper claims that distributed local Gaussian models, compressed and aggregated by self-distillation, reconstruct large 3D surfaces with less memory and time than centralized training.

desk verdict A plausible distributed surface-GS framework, but the paper never explains how independently trained local models share one coordinate frame—without that, the fusion step is not well-defined. read the letter →

arxiv 2412.17612 v1 pith:6JDMDGQA submitted 2024-12-23 cs.CV

Yuanyuan Gao , Yalun Dai , Hao Li , Weicai Ye , Junyi Chen , Danpeng Chen , Dingwen Zhang , Tong He
show 2 more authors
Guofeng Zhang Junwei Han
This is my paper · ORCID
classification cs.CV
keywords 3DGaussiansplattingsurfacereconstructiondistributedlearninglarge-scalemodelcompressionself-distillationdevice-edge-cloudarchitecturemulti-agentcollaboration
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

CoSurfGS tries to make high-quality 3D surface reconstruction of large scenes practical by distributing the work across many devices instead of training one global model on a single GPU. It claims that a device-edge-cloud pipeline, in which each device trains a 3D Gaussian surface model on its own image subset, can be stitched into a single global model through two operations: local model compression (LMC), which prunes redundant boundary Gaussians, and a model aggregation scheme (MAS), which uses the local models as teachers to refine the global model. The paper reports that this reduces training time and GPU memory compared with centralized large-scene baselines while improving surface-reconstruction accuracy measured by mesh F-score, and stays competitive in novel-view rendering. A sympathetic reader would care because large outdoor scenes currently push surface-reconstruction methods past single-GPU memory limits, and the proposed pipeline converts that bottleneck into parallel, privacy-preserving local training.

What carries the argument

The load-bearing machinery is the two-stage local-to-global transition. LMC computes a per-Gaussian priority score from the frequency with which the Gaussian is hit by rays, its opacity, and a normalized volume factor, then prunes the fraction determined by how many cameras the local model shares with the global model, so only Gaussians with reliable multi-view agreement survive. MAS replaces naive set union with self-distillation: the compressed local models act as teachers, rendering RGB, depth, and normal maps from their own camera poses, and the global model is optimized against those teacher renderings plus its own flatness and single-view geometry losses, with no densification. Because the same LMC plus MAS pair is applied at the device-edge and edge-cloud levels, the scheme composes into a hierarchy that can scale the reconstructed region without changing the core operations.

What would settle it

Take a planar wall that spans two devices, each trained on its own half with a narrow shared strip of cameras. After LMC plus MAS fusion, extract the mesh and measure its deviation from a laser scan of the wall; if the strip shows a systematic kink, double surface, or inflated error while each half's interior is accurate, the assumption that self-distillation aligns rather than averages boundary geometry is broken.

Watch

Extended reading notes

Core claim

The paper's central claim is that a large scene's surface can be reconstructed more accurately and far more cheaply by converting the global optimization into a two-level distributed procedure. Each device trains a local planar-surface Gaussian model, following the PGSR approach, with flattening and single- and multi-view geometric constraints; before uploading, LMC prunes Gaussians that appear in few cameras, have low opacity, or sit in regions already covered by other models, using a priority score and a camera-overlap ratio. The edge or cloud then initializes a global model from the compressed local models and refines it with MAS, a self-distillation step where each local model renders RGB, depth, and normal images that supervise the global model, so the global surface is pulled toward the geometry each local model actually saw. The paper argues this avoids the blurry boundaries produced by naive unions of local models without expanding training regions or sharing neighboring images, and presents evaluations on aerial and MVS datasets showing better mesh F-scores, faster training, and lower memory than centralized and prior distributed methods.

Load-bearing premise

The whole pipeline depends on local models trained on separate chunks of the scene agreeing well enough at their borders that pruning and self-distillation can stitch them into one coherent surface without any explicit geometric alignment.

Editorial extensions

If this is right

  • Large scenes that previously caused out-of-memory errors on a single GPU can be reconstructed on consumer hardware by splitting the work among devices and aggregating only the compressed predictions.
  • Training time is cut substantially because all local models train in parallel and the global refinement lasts only a few distillation epochs, making iterative, practical use of large-scene reconstruction feasible.
  • Because only Gaussian models are uploaded and raw images stay on each device, the pipeline offers a privacy-preserving route to collaborative scene modeling across data owners.
  • The local-first strategy yields better surface geometry than a single global model, since each device's model converges on a smaller region and then passes its geometric knowledge to the global model through depth and normal supervision.
  • The same device-edge-cloud aggregation can be repeated over more hierarchy levels, so the method scales beyond the tested scene sizes by adding more devices or edges.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper, the same teacher-student aggregation could be applied to implicit neural surface representations, where local signed-distance or occupancy fields play the role of the compressed Gaussian teachers, if a differentiable renderer supplies depth and normal supervision.
  • The priority score used by LMC is a proxy for multi-view consistency; a testable extension is to replace it with a direct photometric consistency measure, for example reprojection error against neighboring views, and measure whether boundary artifacts shrink further.
  • A natural extension is to measure geometric error in a strip around partition seams separately from interior error, since the paper only reports whole-scene F-scores; this would show whether MAS aligns or averages boundary geometry.
  • The ablations show more partition blocks slightly reduce quality but save time, suggesting an adaptive partition that aligns block boundaries with low-texture or weakly observed regions could recover the quality loss while keeping the speed.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. CoSurfGS proposes a device-edge-cloud distributed learning framework for large-scale 3D surface reconstruction with Gaussian splatting. Each device trains a local PGSR-style flattened-Gaussian model, followed by Local Model Compression (LMC), which prunes Gaussians according to camera-overlap ratios and a priority score, and a Model Aggregation Scheme (MAS), which merges compressed local models and refines the global model via self-distillation from local RGB, depth, and normal renders. Experiments on BlendedMVS, Mill19, and UrbanScene3D report improved mesh F-scores, faster training times, and lower memory consumption relative to centralized and distributed baselines.

Significance. The work addresses a practically important problem: making 3DGS surface reconstruction feasible for large scenes under limited per-device memory. The distributed architecture is novel and the ablations in Tables 4 and 5 show that both LMC and MAS contribute to performance. If the coordinate-frame assumption is clarified and the experimental claims are appropriately qualified, the framework could be a useful contribution to large-scale reconstruction. The paper does not provide code, formal guarantees, or error analysis, so its significance rests mainly on the empirical validation, which currently has important gaps.

major comments (4)
  1. [Sec. 4.2, Fig. 2; Sec. 4.3 Eq. (11); Sec. 4.4 Eq. (13)] The paper states in the Figure 2 caption that each device performs its own SfM to initialize extrinsics and intrinsics. If each device runs SfM independently, the resulting local Gaussian models live in arbitrary similarity gauges (rotation, translation, and scale) that are not aligned to a common world frame. The LMC camera-overlap ratio in Eq. 11 and the MAS merging and self-distillation in Eq. 13 both presuppose a common coordinate frame; without an explicit alignment or registration step, the fusion is geometrically invalid. Even if the datasets provide globally consistent poses (as is typical for Mill19 and UrbanScene3D), the manuscript never states this and the device-side SfM description directly contradicts it. This is the central load-bearing issue and must be resolved, either by describing a global pose initialization or by adding an alignment step.
  2. [Abstract, Contributions, and Table 2] The contributions claim that the method 'surpasses all existing methods' in surface reconstruction and the abstract claims strong photorealistic rendering. In Table 2, however, the no-mesh baseline DOGS achieves higher PSNR than CoSurfGS on 4 of 5 scenes (Building 22.73 vs 22.40, Rubble 25.78 vs 25.39, Campus 24.01 vs 23.63, Sci-Art 24.42 vs 23.29). The state-of-the-art claim needs to be restricted to the 'With Mesh' surface-reconstruction group, and the text should be revised so that the 'comparable to NVS methods' statement is accurate with respect to PSNR.
  3. [Tables 1, 2, 4, 5, 6; Supp. 8.2] All reported metrics are from single runs with no error bars or variance estimation. Some differences are very small, e.g., the Scene-03 F-score gap of 0.0032 in Table 1. Additionally, the F-score error margins in the supplementary (0.5, 0.1, 0.5, 0.2 meters for the four BlendedMVS scenes) vary across scenes without justification, which hampers fair comparison. The quantitative support for the central claim would be substantially stronger with multiple seeds, standard deviations, or at least an explicit acknowledgment that the reported gaps may be within run-to-run noise.
  4. [Sec. 4.3 and 4.4] The paper does not analyze the case where overlapping local models disagree on surface geometry. The self-distillation loss in Eq. 13, with depth and normal supervision in Eq. 14, averages the teacher predictions; if the local surfaces are inconsistent near boundaries, the global model may average away fine structure instead of aligning it. The authors should provide a quantitative boundary-region analysis (e.g., geometry error as a function of distance to partition boundaries) or an experiment that isolates the effect of inconsistent local geometry. This is a broader correctness-risk concern that is not addressed by the current qualitative boundary visualizations.
minor comments (5)
  1. [Fig. 4 and Tables] Fed3DGS appears in the qualitative comparison in Fig. 4 but is not included in any quantitative table; either provide its results or remove it from the comparison.
  2. [Eq. (10)] The notation 'S_pro,k = M HWX' is unclear; the sum index and its range should be spelled out properly, and the definition of V(Σ) should be made less ambiguous.
  3. [Supp. Sec. 9 and Tab. 7] The supplementary material contains numerous typos and formatting errors, including 'Visionlization', 'comaprison', 'mehtod', 'acheives', 'campare', 'noraml', and the corrupted percentage markers in Table 7. The manuscript should be carefully proofread.
  4. [Sec. 5.2 and Table 2] The qualitative text says 'the only method that can model the detail of the telegraph pole' but this claim is not supported by a quantitative metric; consider adding a focused detail-evaluation metric or tempering the claim.
  5. [Sec. 4.1] The privacy statement ('by only uploading Gaussian models instead of raw images') is plausible but not evaluated; if privacy preservation is a claimed contribution, the paper should at least discuss model inversion resilience or refer to previous attacks.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the aggregation and compression schemes are training/distillation procedures, and the headline claims are checked against external ground-truth meshes, images, and independent baselines.

full rationale

The paper's load-bearing claims, Local Model Compression (LMC) and Model Aggregation Scheme (MAS), are not derived from their own outputs. LMC (Sec. 4.3) prunes Gaussians using a camera-overlap ratio and a priority score computed from camera sets; the prune ratio is not taken from the final quality metric. MAS (Sec. 4.4) uses local models as teachers to optimize the merged global model via self-distillation; this is a training objective, not a first-principles prediction, and the resulting global model is evaluated against external ground-truth depth, normal, mesh, and novel-view images. The device-side training borrows geometric constraints from PGSR through Eqs. (6)-(8); although several authors overlap with the PGSR paper, this is an explicit backbone choice rather than a hidden premise that forces the aggregation result. Ablations compare w/ MAS vs w/o MAS and w/ LMC vs w/o LMC on external rendering metrics, and final benchmarks include independent methods such as NeuS, Neuralangelo, 3DGS, and VastGaussian. Any concerns about coordinate-frame alignment at region boundaries or about comparing against a self-cited backbone are correctness or novelty risks, not cases where an equation reduces by definition to its own input. No circular step was found.

Assumptions & free parameters 9 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical or mathematical entities. The central claim rests on standard 3DGS machinery, PGSR geometry losses, and a set of hand-chosen hyperparameters and partition strategies. The most load-bearing assumption is that self-distillation can fuse independently trained local models into a coherent global surface without explicit geometric alignment.

free parameters (9)
  • prune percentage phi = 0.2
    Prune percentage during device training; controls how many low-priority Gaussians are removed (Sec. 4.2, Table 6).
  • single-view geometric loss weight beta2 = 0.01
    Maximum weight for single-view geometric loss in the smooth weighting schedule (Eq. 9).
  • multi-view loss weights beta3 = 0.05 geometric, 0.2 photometric
    Weights for multi-view geometric and photometric consistency losses (Eqs. 3, 9).
  • first training stage iterations tau = 7000
    Iteration threshold after which geometric losses are activated (Eq. 9).
  • scale loss weight lambda1 = 25
    Weight for the scale loss that flattens Gaussians (Eq. 7).
  • depth and normal distillation weights = 0.015
    Weights for depth and normal consistency losses in MAS (Eq. 14, Supp. 8.2).
  • reprojection threshold theta = 1
    Reprojection error threshold for pixel validity in multi-view consistency loss (Eq. 3, Supp. 8.2).
  • MAS distillation epochs = 5
    Number of optimization epochs for the global model in MAS (Supp. 8.2).
  • partition strategy = scene-specific, e.g. 2x2 devices and 4 edges
    Number of spatial partitions is chosen per scene and affects quality and time (Table 8).
assumptions (4)
  • standard math 3D Gaussian Splatting representation and rendering are differentiable and adequate for surface reconstruction.
    Adopted from [14] without re-derivation; all losses rely on this representation.
  • domain assumption PGSR's normal-based single-view and homography-based multi-view consistency losses improve geometric accuracy.
    Used as backbone; correctness is inherited from prior work by overlapping authors.
  • domain assumption Self-distillation from local teacher models yields a globally consistent model without exchanging raw data.
    Eq. 13 assumes teacher renderings of RGB, depth, and normal are sufficient and that no explicit global alignment step is needed.
  • ad hoc to paper Uniform spatial partition into equal areas gives local regions that are independent and consistent enough to merge.
    Supp. 7.2 admits results degrade with finer partitions; no principled partition criterion is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CoSurfGS:Collaborative 3D Surface Gaussian Splatting with Distributed Learning for Large Scene Reconstruction." pith.science (2026). https://pith.science/paper/6JDMDGQA

@misc{pith2026241217612,
  author       = {Pith},
  title        = {Pith review of: CoSurfGS:Collaborative 3D Surface Gaussian Splatting with Distributed Learning for Large Scene Reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6JDMDGQA}},
  note         = {Machine review of arXiv:2412.17612}
}
read the original abstract

3D Gaussian Splatting (3DGS) has demonstrated impressive performance in scene reconstruction. However, most existing GS-based surface reconstruction methods focus on 3D objects or limited scenes. Directly applying these methods to large-scale scene reconstruction will pose challenges such as high memory costs, excessive time consumption, and lack of geometric detail, which makes it difficult to implement in practical applications. To address these issues, we propose a multi-agent collaborative fast 3DGS surface reconstruction framework based on distributed learning for large-scale surface reconstruction. Specifically, we develop local model compression (LMC) and model aggregation schemes (MAS) to achieve high-quality surface representation of large scenes while reducing GPU memory consumption. Extensive experiments on Urban3d, MegaNeRF, and BlendedMVS demonstrate that our proposed method can achieve fast and scalable high-fidelity surface reconstruction and photorealistic rendering. Our project page is available at \url{https://gyy456.github.io/CoSurfGS}.

Figures

Figures reproduced from arXiv: 2412.17612 by the authors.

Figure 1
Figure 1. Our proposed CoSurfGS serves as a ”device-edge-cloud” distributed learning framework that enables multi-agent parallel training. Under this framework, we can achieve superior large-scene reconstruction performance w.r.t the novel view synthesis, depth rendering, and surface normal prediction results (see the bottom part). Meanwhile, this framework can also accelerate the whole modeling process while preserving the p… view at source ↗
Figure 2
Figure 2. Our CoSurfGS follows ”device-edge-cloud” three-layer distributed architecture. On the device side, each device is responsible for reconstructing an individual area by capturing images, performing SfM to initialize both extrinsic and intrinsic, and training the Gaussian models \protect \mathbf {G}^L . On the edge side, devices upload their Gaussian model to the edge followed by two-step aggregation techniques: 1) Loc… view at source ↗
Figure 3
Figure 3. 3D mesh comparison between our method and other surface reconstruction methods. The result of Scene-01, Scene-02, Scene-03, [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Qualitative results of our method and other methods in image and depth rendering, it shows the result of Rubble and Building, [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 6
Figure 6. Figure 6: Qualitative results of Local Model Compression module [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Qualitative results of our method and other methods, [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Qualitative results of our method and other methods in surface reconstruction datasets BlendedMVS [ [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Qualitative results of our method and other methods in large-scale reconstruction datasets Mill-19 [ [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]
Figure 10
Figure 10. Figure 10: The full mesh map of Blendmvs Scene-01, Scene-01, Scene-03. [PITH_FULL_IMAGE:figures/full_fig_p015_10.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Proxy-GS: Unified Occlusion Priors for Training and Inference in Structured 3D Gaussian Splatting

    cs.CV 2025-09 conditional novelty 5.0 of 10

    A proxy mesh rendered through hardware rasterization provides a cheap occlusion depth prior that culls hidden anchors at inference and guides densification at training, giving Octree-GS-like MLP splatting a 3 to 4x sp...

Reference graph

Works this paper leans on

50 extracted references · 32 canonical work pages · cited by 1 Pith paper

  1. [1]

    Building rome in a day

    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. 3

    2011

  2. [2]

    Patchmatch: A randomized correspon- dence algorithm for structural image editing

    Connelly Barnes, Eli Shechtman, Adam Finkelstein, and Dan B Goldman. Patchmatch: A randomized correspon- dence algorithm for structural image editing. ACM Trans. Graph., 28(3):24, 2009. 2

    work page 2009

  3. [3]

    Using multiple hypotheses to improve depth-maps for multi-view stereo

    Neill DF Campbell, George V ogiatzis, Carlos Hern ´andez, and Roberto Cipolla. Using multiple hypotheses to improve depth-maps for multi-view stereo. In Computer Vision– ECCV 2008: 10th European Conference on Computer Vi- sion, Marseille, France, October 12-18, 2008, Proceedings, Part I 10, pages 766–779. Springer, 2008. 2

    work page 2008

  4. [4]

    Pgsr: Planar-based gaussian splatting for efficient and high-fidelity surface reconstruction

    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 efficient and high-fidelity surface reconstruction. arXiv preprint arXiv:2406.06521, 2024. 2, 3, 4, 6, 7

    arXiv 2024

  5. [5]

    Dogaussian: Distributed- oriented gaussian splatting for large-scale 3d reconstruction via gaussian consensus

    Yu Chen and Gim Hee Lee. Dogaussian: Distributed- oriented gaussian splatting for large-scale 3d reconstruction via gaussian consensus. arXiv preprint arXiv:2405.13943 ,

    arXiv

  6. [6]

    Gaussianeditor: Swift and control- lable 3d editing with gaussian splatting

    Yiwen Chen, Zilong Chen, Chi Zhang, Feng Wang, Xi- aofeng Yang, Yikai Wang, Zhongang Cai, Lei Yang, Huaping Liu, and Guosheng Lin. Gaussianeditor: Swift and control- lable 3d editing with gaussian splatting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21476–21485, 2024. 2

    2024

  7. [7]

    An automated method for large-scale, ground-based city model acquisition

    Christian Fr ¨uh and Avideh Zakhor. An automated method for large-scale, ground-based city model acquisition. Inter- national Journal of Computer Vision, 60:5–24, 2004. 3

    work page 2004

  8. [8]

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

    Antoine Gu ´edon and Vincent Lepetit. Sugar: Surface- aligned gaussian splatting for efficient 3d mesh recon- struction and high-quality mesh rendering. arXiv preprint arXiv:2311.12775, 2023. 3, 6

    arXiv 2023

Show all 50 references
  1. [9]

    Cg-slam: Efficient dense rgb-d slam in a consistent uncertainty-aware 3d gaussian field, 2024

    Jiarui Hu, Xianhao Chen, Boyin Feng, Guanglin Li, Liangjing Yang, Hujun Bao, Guofeng Zhang, and Zhaopeng Cui. Cg-slam: Efficient dense rgb-d slam in a consistent uncertainty-aware 3d gaussian field, 2024. 2

    2024

  2. [10]

    2d gaussian splatting for geometrically accu- rate radiance fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically accu- rate radiance fields. arXiv preprint arXiv:2403.17888, 2024. 2, 3

    2024 arXiv

  3. [11]

    Adversarial generation of hi- erarchical gaussians for 3d generative model

    Sangeek Hyun and Jae-Pil Heo. Adversarial generation of hi- erarchical gaussians for 3d generative model. arXiv preprint arXiv:2406.02968, 2024. 6

    2024 arXiv

  4. [12]

    Splatam: Splat, track i& map 3d gaussians for dense rgb-d slam, 2024

    Nikhil Keetha, Jay Karhade, Krishna Murthy Jatavallabhula, Gengshan Yang, Sebastian Scherer, Deva Ramanan, and Jonathon Luiten. Splatam: Splat, track i& map 3d gaussians for dense rgb-d slam, 2024. 2

    2024

  5. [13]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics, 42 (4):1–14, 2023. 3

    2023

  6. [14]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics ,

  7. [15]

    A theory of shape by space carving

    Kiriakos N Kutulakos and Steven M Seitz. A theory of shape by space carving. International journal of computer vision , 38:199–218, 2000. 2

    2000

  8. [16]

    A quasi-dense approach to surface reconstruction from uncalibrated images

    Maxime Lhuillier and Long Quan. A quasi-dense approach to surface reconstruction from uncalibrated images. IEEE transactions on pattern analysis and machine intelligence , 27(3):418–433, 2005. 2

    2005

  9. [17]

    Neuralangelo: High-fidelity neural surface reconstruction

    Zhaoshuo Li, Thomas M ¨uller, Alex Evans, Russell H Tay- lor, Mathias Unberath, Ming-Yu Liu, and Chen-Hsuan Lin. Neuralangelo: High-fidelity neural surface reconstruction. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 8456–8465, 2023. 6

    2023

  10. [18]

    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. arXiv preprint arXiv:2402.17427 ,

    arXiv

  11. [19]

    Capturing, reconstructing, and simulating: the urbanscene3d dataset

    Liqiang Lin, Yilin Liu, Yue Hu, Xingguang Yan, Ke Xie, and Hui Huang. Capturing, reconstructing, and simulating: the urbanscene3d dataset. In European Conference on Computer Vision, pages 93–109. Springer, 2022. 6, 1, 4

    2022

  12. [20]

    Humangaus- sian: Text-driven 3d human generation with gaussian splat- ting, 2024

    Xian Liu, Xiaohang Zhan, Jiaxiang Tang, Ying Shan, Gang Zeng, Dahua Lin, Xihui Liu, and Ziwei Liu. Humangaus- sian: Text-driven 3d human generation with gaussian splat- ting, 2024. 2

    2024

  13. [21]

    Nerf: Representing scenes as neural radiance fields for view syn- thesis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view syn- thesis. Communications of the ACM , 65(1):99–106, 2021. 3

    2021

  14. [22]

    Mescheder, Michael Oechsle, and Andreas Geiger

    Michael Niemeyer, Lars M. Mescheder, Michael Oechsle, and Andreas Geiger. Differentiable volumetric rendering: Learning implicit 3D representations without 3D supervi- sion. In IEEE Conference on Computer Vision and Pattern Recognition, pages 3501–3512, 2020. 3

    2020

  15. [23]

    Newcombe, and Steven Lovegrove

    Jeong Joon Park, Peter Florence, Julian Straub, Richard A. Newcombe, and Steven Lovegrove. DeepSDF: Learning continuous signed distance functions for shape representa- 9 tion. In IEEE Conference on Computer Vision and Pattern Recognition, pages 165–174, 2019. 2

    2019

  16. [24]

    From coarse to fine: Robust hierarchical localization at large scale

    Paul-Edouard Sarlin, Cesar Cadena, Roland Siegwart, and Marcin Dymczyk. From coarse to fine: Robust hierarchical localization at large scale. In CVPR, 2019. 2

    2019

  17. [25]

    Structure- from-motion revisited

    Johannes L Schonberger and Jan-Michael Frahm. Structure- from-motion revisited. In Proceedings of the IEEE con- ference on computer vision and pattern recognition , pages 4104–4113, 2016. 3

    2016

  18. [26]

    Pixelwise view selection for un- structured multi-view stereo

    Johannes Lutz Sch ¨onberger, Enliang Zheng, Marc Pollefeys, and Jan-Michael Frahm. Pixelwise view selection for un- structured multi-view stereo. In European Conference on Computer Vision (ECCV), 2016. 2

    2016

  19. [27]

    Fed3dgs: Scalable 3d gaussian splatting with federated learning

    Teppei Suzuki. Fed3dgs: Scalable 3d gaussian splatting with federated learning. arXiv preprint arXiv:2403.11460, 2024. 3

    2024 arXiv

  20. [28]

    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 Reco...

    2022

  21. [29]

    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. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12922–12931, 2022. 3, 6, 7, 1, 4

    2022

  22. [30]

    Patchmatchnet: Learned multi-view patchmatch stereo

    Fangjinhua Wang, Silvano Galliani, Christoph V ogel, Pablo Speciale, and Marc Pollefeys. Patchmatchnet: Learned multi-view patchmatch stereo. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14194–14203, 2021. 2

    2021

  23. [31]

    Gaussianeditor: Editing 3d gaussians delicately with text instructions, 2024

    Junjie Wang, Jiemin Fang, Xiaopeng Zhang, Lingxi Xie, and Qi Tian. Gaussianeditor: Editing 3d gaussians delicately with text instructions, 2024. 2

    2024

  24. [32]

    Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction

    Peng Wang, Lingjie Liu, Yuan Liu, Christian Theobalt, Taku Komura, and Wenping Wang. Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction. arXiv preprint arXiv:2106.10689, 2021. 3, 6

    2021 arXiv

  25. [33]

    4d gaussian splatting for real-time dynamic scene rendering,

    Guanjun Wu, Taoran Yi, Jiemin Fang, Lingxi Xie, Xiaopeng Zhang, Wei Wei, Wenyu Liu, Qi Tian, and Xinggang Wang. 4d gaussian splatting for real-time dynamic scene rendering,

  26. [34]

    Bungeenerf: Progressive neural radiance field for extreme multi-scale scene rendering

    Yuanbo Xiangli, Linning Xu, Xingang Pan, Nanxuan Zhao, Anyi Rao, Christian Theobalt, Bo Dai, and Dahua Lin. Bungeenerf: Progressive neural radiance field for extreme multi-scale scene rendering. In European conference on computer vision, pages 106–122. Springer, 2022. 3

    2022

  27. [35]

    Grid-guided neural radiance fields for large urban scenes

    Linning Xu, Yuanbo Xiangli, Sida Peng, Xingang Pan, Nanxuan Zhao, Christian Theobalt, Bo Dai, and Dahua Lin. Grid-guided neural radiance fields for large urban scenes. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 8296–8306, 2023. 3

    2023

  28. [36]

    Gs-slam: Dense visual slam with 3d gaussian splatting, 2024

    Chi Yan, Delin Qu, Dan Xu, Bin Zhao, Zhigang Wang, Dong Wang, and Xuelong Li. Gs-slam: Dense visual slam with 3d gaussian splatting, 2024. 2

    2024

  29. [37]

    Street gaussians: Modeling dynamic urban scenes with gaussian splatting, 2024

    Yunzhi Yan, Haotong Lin, Chenxu Zhou, Weijie Wang, Haiyang Sun, Kun Zhan, Xianpeng Lang, Xiaowei Zhou, and Sida Peng. Street gaussians: Modeling dynamic urban scenes with gaussian splatting, 2024. 2

    2024

  30. [38]

    Multi-scale 3d gaussian splatting for anti-aliased rendering,

    Zhiwen Yan, Weng Fei Low, Yu Chen, and Gim Hee Lee. Multi-scale 3d gaussian splatting for anti-aliased rendering,

  31. [39]

    Blendedmvs: A large- scale dataset for generalized multi-view stereo networks

    Yao Yao, Zixin Luo, Shiwei Li, Jingyang Zhang, Yufan Ren, Lei Zhou, Tian Fang, and Long Quan. Blendedmvs: A large- scale dataset for generalized multi-view stereo networks. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 1790–1799,...

    2020

  32. [40]

    Gaussiandreamer: Fast generation from text to 3d gaussians by bridging 2d and 3d diffusion models, 2024

    Taoran Yi, Jiemin Fang, Junjie Wang, Guanjun Wu, Lingxi Xie, Xiaopeng Zhang, Wenyu Liu, Qi Tian, and Xinggang Wang. Gaussiandreamer: Fast generation from text to 3d gaussians by bridging 2d and 3d diffusion models, 2024. 2

    2024

  33. [41]

    Fast normalized cross- correlation

    Jae-Chern Yoo and Tae Hee Han. Fast normalized cross- correlation. Circuits, systems and signal processing, 28:819– 843, 2009. 3

    2009

  34. [42]

    Mip-splatting: Alias-free 3d gaussian splat- ting

    Zehao Yu, Anpei Chen, Binbin Huang, Torsten Sattler, and Andreas Geiger. Mip-splatting: Alias-free 3d gaussian splat- ting. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 19447–19456,

  35. [43]

    Gaussian opacity fields: Efficient and compact surface reconstruc- tion in unbounded scenes

    Zehao Yu, Torsten Sattler, and Andreas Geiger. Gaussian opacity fields: Efficient and compact surface reconstruc- tion in unbounded scenes. arXiv preprint arXiv:2404.10772,

    arXiv

  36. [44]

    3dmatch: Learning local geometric descriptors from rgb-d reconstruc- tions

    Andy Zeng, Shuran Song, Matthias Nießner, Matthew Fisher, Jianxiong Xiao, and Thomas Funkhouser. 3dmatch: Learning local geometric descriptors from rgb-d reconstruc- tions. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1802–1811, 2017. 6

    2017

  37. [45]

    Ef- ficient large-scale scene representation with a hybrid of high-resolution grid and plane features

    Yuqi Zhang, Guanying Chen, and Shuguang Cui. Ef- ficient large-scale scene representation with a hybrid of high-resolution grid and plane features. arXiv preprint arXiv:2303.03003, 2023. 3

    2023 arXiv

  38. [46]

    Switch-nerf: Learning scene de- composition with mixture of experts for large-scale neural radiance fields

    MI Zhenxing and Dan Xu. Switch-nerf: Learning scene de- composition with mixture of experts for large-scale neural radiance fields. In The Eleventh International Conference on Learning Representations, 2022. 3, 6, 7

    2022

  39. [47]

    Dreamscene360: Uncon- strained text-to-3d scene generation with panoramic gaussian splatting, 2024

    Shijie Zhou, Zhiwen Fan, Dejia Xu, Haoran Chang, Pradyumna Chari, Tejas Bharadwaj, Suya You, Zhangyang Wang, and Achuta Kadambi. Dreamscene360: Uncon- strained text-to-3d scene generation with panoramic gaussian splatting, 2024. 2 10 CoSurfGS:Collaborative 3D Surface Gaussian ...

    2024

  40. [48]

    Additional Ablation Studies 7.1. Geometry Supervision in Device Training The performance of device training is critical for the final results since it is the teacher model of the distillation-based aggregation procedure. Therefore, we carefully design a geometry supervision st...

  41. [49]

    Details 8.1. Datasets For surface reconstruction evaluation, we choose four scenes in BlendedMVS [39], Scene-01, Scene- 02, Scene-03, Scene-04 correspond to the scene 5bbb6eb2ea1cfa39f1af7e0c, 5b271079e0878c3816dacca4, 5b864d850d072a699b32f4ae, 5b60fa0c764f146feef84df0, each o...

    1940

  42. [50]

    Visualiztion 9.1. Additional depth Visionlization The comaprison of depth maps in other ohter datasets can be seen in 8, It’s obvious that our mehtod achieves the most accurate depth, as seen the images from Residence, we acheives the most smooth and consistant floor compare t...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.