Pith. sign in

REVIEW 4 major objections 6 minor 49 references

HaloGS: Loose Coupling of Compact Geometry and Gaussian Splats for 3D Scenes

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read HaloGS pairs learnable triangles with neural Gaussians to get compact meshes and photorealistic views in one model.

desk verdict A credible hybrid geometry-appearance system, but the headline geometry win is partly an artifact of comparing planar extraction on differently formatted inputs. read the letter →

arxiv 2505.20267 v1 pith:OQVI276G submitted 2025-05-26 cs.CV

Changjian Jiang , Kerui Ren , Linning Xu , Jiong Chen , Jiangmiao Pang , Yu Zhang , Bo Dai , Mulin Yu This is my paper
classification cs.CV
keywords 3DGaussianSplattingtriangledualrepresentationcompactmeshextractionplanarprimitivesmonoculardepthandnormalpriorsnovelviewsynthesislevel-of-detailrendering
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

This paper tries to show that a 3D scene can be represented as a set of learnable triangles that carry the geometry plus a cloud of neural Gaussians that carry the appearance, and that this loose split avoids the usual trade-off between compact, editable geometry and photorealistic rendering. The authors evaluate the method on 32 indoor and outdoor scenes and report lower Chamfer distance to ground-truth geometry than MVS and 2DGS baselines while keeping rendering quality on par with or above 3DGS-based methods. If the claim holds, one trained model provides a low-poly mesh for storage, editing, and collision detection, and Gaussian splats for real-time novel views. The key is that geometry and appearance are optimized in separate, mutually supportive stages rather than entangled in a single representation.

What carries the argument

The load-bearing object is the learnable triangle primitive: three movable vertices parameterized through a local tangent frame (Eq. 2), rasterized by a custom CUDA kernel that computes ray–triangle intersections and blends contributions with an edge-preserving kernel (Eq. 3) to render depth and normal maps. A coarse stage supervises these triangles with monocular depth from Depth-Anything-V2 and normals from StableNormal, aligned to SfM points by RANSAC; a fine stage decodes neural Gaussians from each triangle following the Scaffold-GS scheme, embeds them in an Octree-GS octree, and supervises the triangles through Gaussian-rendered depth/normal maps weighted by a wavelet edge map, alongside image losses and adaptive splitting and pruning of triangles. The optimized triangle soup is clustered by GoCoPP planar primitive detection into multi-scale LoD planes that seed both rendering and the assembly of a compact mesh.

What would settle it

Reconstruct a textureless white room or a low-light corridor with known ground-truth geometry, where monocular depth and normal priors are unreliable and SfM points are sparse. If HaloGS's triangle soup and compact mesh show Chamfer distance no better than 2DGS or MVS while its rendering PSNR stays high, the claim that accurate geometry and photorealistic appearance are jointly achieved would fail: appearance could be right while geometry is wrong.

Watch

Extended reading notes

Core claim

HaloGS's central claim is that decoupling geometry from appearance, instead of fusing them into one implicit field or forcing Gaussians onto a thin surface, makes a representation simultaneously compact and photorealistic. The paper argues that learnable triangle primitives, splatted through a custom rasterizer with an edge-preserving kernel, can be driven to accurate surfaces by monocular depth and normal priors in a coarse stage, and that neural Gaussians decoded from those triangles then render images while feeding finer depth and normal maps back into the triangles in a fine stage. This loose coupling, the authors claim, is why the extracted planar primitives and compact mesh beat dense reconstructions in both fidelity and simplicity on planar indoor and urban scenes.

Load-bearing premise

The pipeline stands on the assumption that monocular depth and normal predictions from pretrained diffusion models, after RANSAC alignment to sparse SfM points, are accurate enough to steer the triangle soup in the coarse stage, and that the fine stage's self-supervision from Gaussian-rendered depth and normals does not bake those priors' mistakes into the final geometry.

Editorial extensions

If this is right

  • A trained HaloGS model gives both a photorealistic renderer and an editable low-poly mesh from the same data, so storage and real-time streaming can use the mesh while rendering uses the Gaussians.
  • In textureless or low-light areas where SfM points are sparse, the monocular-prior-guided triangles provide a coherent surface that keeps rendering quality from collapsing.
  • Because the triangles are explicitly optimized for geometry and only loosely coupled to appearance, the planar abstraction step can discard redundant geometry without hurting novel-view quality.
  • The compact mesh and planar segments can serve downstream tasks such as collision detection, geometric editing, and semantic reasoning directly from the trained model.

Reading between the lines

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

  • If the monocular priors are systematically wrong, such as on mirror-like or highly reflective surfaces, the coarse stage may lock triangles onto a plausible-but-wrong surface; a worthwhile test is comparing mesh quality against ground-truth LiDAR in regions where the depth prior and SfM disagree.
  • The same loose-coupling idea might transfer to other appearance decoders, such as a lightweight radiance field or feature grid, suggesting the triangle scaffold, not the specific Gaussian decoder, is the core contribution.
  • The wavelet-weighted fine-stage loss predicts that high-frequency texture regions receive more geometric refinement; an ablation that removes the weight map would reveal whether it chiefly improves edges or also affects flat-surface accuracy.
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 / 6 minor

Summary. The paper proposes HaloGS, a dual representation that decouples geometry from appearance: coarse learnable triangle primitives provide explicit, compact geometry, while neural Gaussians decoded from triangle features are responsible for photorealistic rendering, organized in an Octree-GS-style level-of-detail structure. Training is coarse-to-fine: the coarse stage supervises triangles with monocular depth and normal priors (Depth-Anything-V2 and StableNormal, after RANSAC alignment with SfM points), and the fine stage adds photometric supervision together with Gaussian-rendered depth and normal maps as pseudo-ground truth for further geometric refinement. Adaptive density control (edge splitting, triangle pruning, opacity entropy regularization) is used to obtain clean reconstructions. Planar primitives are extracted from the optimized triangle soup with GoCoPP and stitched into compact meshes. Experiments cover 32 scenes from Deep Blending, ScanNet++, VR-NeRF, Zip-NeRF, MatrixCity, and FAST-LIVO2, with rendering metrics and geometric evaluations against ground-truth meshes and point clouds.

Significance. If the central claims hold, HaloGS is a useful point in the design space of hybrid scene representations: it offers an explicit low-poly geometry scaffold with photorealistic Gaussian appearance, compact storage, and downstream mesh usability, and its custom differentiable triangle rasterizer with subdivision and visibility correction is a concrete engineering contribution. Strengths include the broad evaluation across diverse indoor and outdoor datasets, per-scene result tables, and clean ablation studies for the adaptive density control and geometric detail enhancement. However, the headline geometry comparison is confounded by the planar-extraction protocol, the rendering results are not consistently above baselines, and a promised sensitivity analysis is missing from the supplementary material. The present evidence supports HaloGS as competitive and promising rather than as a demonstrated state-of-the-art improvement.

major comments (4)
  1. [§4.2, Table 3] The geometry comparison is confounded by the planar-extraction protocol. GoCoPP is applied to all outputs with the same fixed 10-iteration coarse-to-fine schedule (Table 5), but the method is sensitive to input density, noise, and normal coherence. Uniformly sampled points from perfectly planar triangles are a near-ideal input, whereas MVS point clouds and 2DGS or GFSGS outputs have different noise and sampling characteristics, so the reported Chamfer distances conflate representation quality with input-format quality. The per-scene primitive counts in Table 11 (e.g., Ours ranges from 1039 to 2902 on ScanNet++) further suggest that extraction settings dominate the reported numbers. Please add a direct geometric evaluation (e.g., mesh or point-cloud Chamfer distance against ground truth without planar abstraction) and/or per-method tuning of extraction parameters, and report how Table 3 changes when the GoCoPP schedule is varied.
  2. [§4.1, Tables 1–2; §A.2.1] The claim that HaloGS 'consistently outperforms baselines in rendering quality' is not supported by the reported aggregates. In Table 2, FAST-LIVO2 PSNR is 29.81 for Ours versus 29.96 for 3DGS; in Table 1, VR-NeRF SSIM is 0.916 for Ours versus 0.918 for 3DGS. The per-scene tables (e.g., Tables 8–9) also show many scenes where Ours is below 3DGS or Octree-GS on one or more metrics. No error bars, repeated-run statistics, or significance tests are provided. Please temper the wording and add variance or significance analysis to support any comparative claims.
  3. [§3.3, Eq. (6)] Equation (6) appears to contain a typo: the first Lgeo term writes W⊙Dgs twice, where the second argument should almost certainly be the Gaussian-rendered normal map W⊙Ngs. In addition, the wavelet-based weight map W is not fully specified (wavelet type, level, normalization), and its complement notation W=1−W is confusing. Because this loss is central to the fine-stage geometry refinement, please correct the equation and specify the weight map construction precisely.
  4. [§5 and §A.3] Section 5 promises that 'sensitivity to the chosen coarse-to-fine training schedule' is analyzed in the Supplementary, but Supplementary A.3 contains only qualitative limitations about transparent objects and sky/background regions, with no schedule-sensitivity analysis. This is a missing promised support. Given that the coarse stage relies on monocular priors (Eq. 5) and the fine stage uses Gaussian-rendered depth and normal maps as pseudo-labels (Eq. 6), please either provide the promised sensitivity analysis or remove the promise, and add an ablation that substitutes a different depth/normal prior (or ablates the RANSAC alignment) to show the geometric results are not tied to the specific priors.
minor comments (6)
  1. [§1] Typo: 'Benifiting' should be 'Benefiting'.
  2. [Fig. 1 caption] Typo: 'Aseembling' should be 'Assembling'; the same typo appears elsewhere in the figure.
  3. [§3.2] Typo: 'traingles' should be 'triangles'; also 'shapeness and planeness' is awkward and should be rephrased.
  4. [§2, Table 2] Zip-NeRF is described as an 'indoor-outdoor dataset' but its scenes are unbounded outdoor scenes; please correct the dataset description.
  5. [Table 4] The ablation row 'w/o enhancing geomtirc detail using Gaussians' contains a typo ('geomtirc' should be 'geometric').
  6. [§4.1] Please add a reproducibility statement indicating whether training or evaluation code will be released.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the Eq. (6) triangle-Gaussian feedback is an externally anchored co-training loop, Table 3 is checked against ground-truth meshes and LIDAR, and the self-cited components are independently benchmarked; the open issues are evaluation fairness and reproducibility, not definitional reduction.

full rationale

The HaloGS derivation chain contains no definitional reduction. In the coarse stage (Sec. 3.3, Eq. 5), triangle geometry is supervised by external monocular priors (Depth-Anything-V2 depth, StableNormal normals) after RANSAC alignment to SfM points, so the geometric signal originates outside the model. In the fine stage (Sec. 3.3, Eq. 6), neural Gaussians are decoded from the triangles and anchored by a ground-truth photometric term, and the optimized Gaussians' rendered depth/normal maps back-supervise the same triangles; this is the only candidate intra-model loop, since Dgs and Ngs are functions of Gaussian parameters that are themselves functions of the supervised triangles. That loop is not circular in the sense defined here: the target is not a fitted quantity repackaged as a prediction, because the Gaussian positions are constrained by ground-truth image supervision, and the final geometry claim in Table 3 is judged against external ScanNet++ ground-truth meshes and FAST-LIVO2 LIDAR point clouds rather than the model's own renders or the monocular priors. The planar-abstraction evaluation applies the authors' GoCoPP uniformly with the fixed Table 5 schedule to all inputs; the concern that uniformly sampled points with perfect normals from a triangle soup are better-conditioned GoCoPP input than noisy MVS point clouds is a genuine validity threat to the Table 3 gap (and the wide per-scene primitive counts in Table 11 suggest extraction settings dominate), but it is a measurement-fairness issue, not a reduction of a prediction to its inputs. Self-citations (Scaffold-GS, Octree-GS, GoCoPP, GSDF) supply architecture, renderer, and extraction components; they are published, independently benchmarked, and no uniqueness theorem from the authors' own work forces the design, so the self-citation chain provides evidence rather than circularity. Two manuscript-internal flags that do not change the verdict: Eq. (6) as printed repeats W⊙Dgs as the normal target where W⊙Ngs is evidently intended, making the written loss dimensionally inconsistent; and Sec. 5 promises 'see Supplementary for a detailed analysis' of coarse-to-fine schedule sensitivity, but Supplement A.3 discusses only transparency and sky limitations, so the promised analysis is absent. Both are reproducibility defects, not circular steps. Verdict: no significant circularity; score 2 reflects the pervasive self-authored infrastructure and the evaluation-instrument overlap, not a forced derivation.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central result relies on initialization from SfM, monocular priors, and several heuristic thresholds; these are not derived from first principles but are standard domain assumptions for 3DGS-style pipelines. No new physical entities are introduced.

free parameters (5)
  • Loss weights lambda_d, lambda_rgb, lambda_c, lambda_s = 10, 10, 0.2, 0.01
    Set by hand in Sec. 3.3; no sensitivity analysis or per-dataset tuning is reported.
  • Edge-splitting gradient threshold = 1e-5
    User-defined threshold in Eq. (7) that controls when triangles are split; no ablation of this value is given.
  • Triangle pruning thresholds = opacity < 0.5, contribution < 2.0
    Defined in Supp. A.1.2; these cutoff values are ad hoc and not swept.
  • Triangle subdivision edge-length threshold delta = not specified
    Introduced in Supp. A.1.1 to control depth-sorting accuracy; the numerical value is not reported.
  • GoCoPP plane-extraction parameters at three LoD levels = Table 5 parameter schedules
    Plane extraction uses tuned schedules for epsilon, sigma, and normal consistency; these directly affect the reported geometry metrics.
assumptions (6)
  • domain assumption SfM sparse point clouds from COLMAP provide sufficient initialization for triangle seeds.
    All triangles are initialized from SfM points in Sec. 3.3; if SfM is sparse or fails in textureless scenes, the geometry can be incomplete.
  • domain assumption Monocular depth and normal priors, after RANSAC depth calibration, are accurate enough to supervise triangle geometry.
    The coarse-stage loss in Eq. (5) uses Depth-Anything-V2 and StableNormal outputs; failure of these priors would propagate to the final geometry.
  • domain assumption The edge-splitting criterion based on the gradient of the 3DGS implicit field correctly identifies missing geometry.
    Eq. (7) is a heuristic; no proof or ablation of the threshold is provided.
  • domain assumption Triangle soup without 2-manifold connectivity is a sufficient geometric representation for the target scenes.
    The authors explicitly accept non-manifold output in Sec. 3, and the geometry claim depends on this loose connectivity being adequate for planar-heavy scenes.
  • domain assumption Ray-splat intersection and alpha-composited depth and normal maps provide unbiased geometric estimates.
    The triangle rasterizer follows 2DGS and 3DCS formulations in Sec. 3.2 and Supp. A.1.1, but unbiasedness is not proven and large triangles require subdivision to avoid sorting artifacts.
  • domain assumption GoCoPP planar extraction and kinetic shape reconstruction produce compact meshes faithful to the triangle soup.
    The reported geometry fidelity and compact meshes in Sec. 3.4 depend on these prior algorithms working as expected on triangle-soup samples.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HaloGS: Loose Coupling of Compact Geometry and Gaussian Splats for 3D Scenes." pith.science (2026). https://pith.science/paper/OQVI276G

@misc{pith2026250520267,
  author       = {Pith},
  title        = {Pith review of: HaloGS: Loose Coupling of Compact Geometry and Gaussian Splats for 3D Scenes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OQVI276G}},
  note         = {Machine review of arXiv:2505.20267}
}
read the original abstract

High fidelity 3D reconstruction and rendering hinge on capturing precise geometry while preserving photo realistic detail. Most existing methods either fuse these goals into a single cumbersome model or adopt hybrid schemes whose uniform primitives lead to a trade off between efficiency and fidelity. In this paper, we introduce HaloGS, a dual representation that loosely couples coarse triangles for geometry with Gaussian primitives for appearance, motivated by the lightweight classic geometry representations and their proven efficiency in real world applications. Our design yields a compact yet expressive model capable of photo realistic rendering across both indoor and outdoor environments, seamlessly adapting to varying levels of scene complexity. Experiments on multiple benchmark datasets demonstrate that our method yields both compact, accurate geometry and high fidelity renderings, especially in challenging scenarios where robust geometric structure make a clear difference.

Figures

Figures reproduced from arXiv: 2505.20267 by the authors.

Figure 1
Figure 1. HaloGS presents a dual-representation framework that disentangles geometry from appear [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of HaloGS. Our proposed dual-representation is illustrated in (a), where learnable triangles explicitly fit the scene geometry, and neural Gaussians decoded from these triangles render the appearance. In (b), we depict our coarse-to-fine training strategy: during the coarse stage, monocular geometric priors supervise the positions and shapes of the triangles. Subsequently, in the fine stage, neural Gaussian… view at source ↗
Figure 3
Figure 3. We evaluate our method against state-of-the-art approaches [ [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Geometric reconstruction comparison. We visualize the learned geometric representations our Triangle Soup, 2DGS meshes, our extracted planes, and 2DGS planes on two representative datasets. The top four rows show ScanNet++ indoor scenes with available ground-truth mesh…
Figure 5
Figure 5. Figure 5: Visualization of the compact mesh. We present two representative scenes: Raf_emptyroom (from VR-NeRF) and Garden (from MipNeRF-360) in both indoor and outdoor settings. 4 Experiments 4.1 Experimental Setup Datasets. We evaluate HaloGS on 32 real-world scenes drawn from…
Figure 6
Figure 6. Figure 6: Ablation results. Visualizations of triangle soup on DB [7] ablation studies. Numbered insets highlight where each omission degrades reconstruction quality. planar-primitive fitting and mesh extraction pipeline, HaloGS maximally compresses primitive counts while preser…
Figure 7
Figure 7. Figure 7: Two issues in forward rendering and our rendering pipeline. Forward Rendering Process. To achieve unbiased depth map rendering, we adopt the explicit ray-splat intersection algorithm [30], similar to 2DGS [9], which determines the corresponding planar coordinate of eac…
Figure 8
Figure 8. Figure 8: demonstrates the effectiveness of our forward rendering pipeline. The pipeline accurately reproduces depth and normal maps for both ground-truth meshes and our extracted planar primitives. The last two columns of [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: An illustration of the backward computation graph. This graph demonstrates the gradient propagation process from H, a, and n to the gradients with respect to three vertices of the triangle p0, p1, and p2. A.1.2 Triangle Pruning Mechanism In Sec. 3.3, an efficient trian…
Figure 10
Figure 10. Figure 10: The effectiveness of orienting triangle normals. The colors of the triangles are calculated based on their normals [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]
Figure 11
Figure 11. Figure 11: Qualitative rendering results. We present the rendering results of additional scenes. Our method showcases superior rendering effects on surfaces such as walls, floors, and ceilings [PITH_FULL_IMAGE:figures/full_fig_p018_11.png]
Figure 12
Figure 12. Figure 12: Qualitative geometric reconstruction comparisons. We provide geometric reconstruc￾tion comparisons of four more scenes from ScanNet++ [40] dataset. Scene Point Cloud Triangle Soup table apartment Scene Point Cloud Triangle Soup Scene Point Cloud Triangle Soup kitchen …
Figure 13
Figure 13. Figure 13: Qualitative geometric reconstruction results. We present geometric reconstruction results of four more scenes from VR-NeRF [35] dataset [PITH_FULL_IMAGE:figures/full_fig_p019_13.png]
Figure 14
Figure 14. Figure 14: We present additional geometric reconstruction results across diverse scenes from three [PITH_FULL_IMAGE:figures/full_fig_p020_14.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

49 extracted references · 23 canonical work pages

  1. [1]

    Mip- nerf 360: Unbounded anti-aliased neural radiance fields

    Jonathan T Barron, Ben Mildenhall, Dor Verbin, Pratul P Srinivasan, and Peter Hedman. Mip- nerf 360: Unbounded anti-aliased neural radiance fields. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5470–5479, 2022

    2022

  2. [2]

    Barron, Ben Mildenhall, Dor Verbin, Pratul P

    Jonathan T. Barron, Ben Mildenhall, Dor Verbin, Pratul P. Srinivasan, and Peter Hedman. Zip-nerf: Anti-aliased grid-based neural radiance fields.ICCV, 2023

    work page 2023

  3. [3]

    Kinetic shape reconstruction.ACM Transactions on Graphics (TOG), 39(5):1–14, 2020

    Jean-Philippe Bauchet and Florent Lafarge. Kinetic shape reconstruction.ACM Transactions on Graphics (TOG), 39(5):1–14, 2020

    work page 2020

  4. [4]

    Bsp-net: Generating compact meshes via binary space partitioning

    Zhiqin Chen, Andrea Tagliasacchi, and Hao Zhang. Bsp-net: Generating compact meshes via binary space partitioning. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 45–54, 2020

    work page 2020

  5. [5]

    Plenoxels: Radiance fields without neural networks

    Sara Fridovich-Keil, Alex Yu, Matthew Tancik, Qinhong Chen, Benjamin Recht, and Angjoo Kanazawa. Plenoxels: Radiance fields without neural networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5501–5510, 2022

    2022

  6. [6]

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

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

    2024

  7. [7]

    Deep blending for free-viewpoint image-based rendering.ACM Transactions on Graphics (ToG), 37(6):1–15, 2018

    Peter Hedman, Julien Philip, True Price, Jan-Michael Frahm, George Drettakis, and Gabriel Brostow. Deep blending for free-viewpoint image-based rendering.ACM Transactions on Graphics (ToG), 37(6):1–15, 2018

    2018

  8. [8]

    3d convex splatting: Radiance field rendering with 3d smooth convexes.arXiv preprint arXiv:2411.14974, 2024

    Jan Held, Renaud Vandeghen, Abdullah Hamdi, Adrien Deliege, Anthony Cioppa, Silvio Giancola, Andrea Vedaldi, Bernard Ghanem, and Marc Van Droogenbroeck. 3d convex splatting: Radiance field rendering with 3d smooth convexes.arXiv preprint arXiv:2411.14974, 2024

    arXiv 2024

Show all 49 references
  1. [9]

    2d gaussian splatting for geometrically accurate radiance fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically accurate radiance fields. InACM SIGGRAPH 2024 conference papers, pages 1–11, 2024. 10

    2024

  2. [10]

    Geometry field splatting with gaussian surfels.arXiv preprint arXiv:2411.17067, 2024

    Kaiwen Jiang, Venkataram Sivaram, Cheng Peng, and Ravi Ramamoorthi. Geometry field splatting with gaussian surfels.arXiv preprint arXiv:2411.17067, 2024

    2024 arXiv

  3. [11]

    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. arXiv preprint arXiv:2412.01745, 2024

    2024 arXiv

  4. [12]

    3d gaussian splatting for real-time radiance field rendering.ACM Trans

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Trans. Graph., 42(4):139–1, 2023

    2023

  5. [13]

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

    2024

  6. [14]

    Tanks and temples: Bench- marking large-scale scene reconstruction.ACM Transactions on Graphics (ToG), 36(4):1–13, 2017

    Arno Knapitsch, Jaesik Park, Qian-Yi Zhou, and Vladlen Koltun. Tanks and temples: Bench- marking large-scale scene reconstruction.ACM Transactions on Graphics (ToG), 36(4):1–13, 2017

    2017

  7. [15]

    Supervised fitting of geometric primitives to 3d point clouds

    Lingxiao Li, Minhyuk Sung, Anastasia Dubrovina, Li Yi, and Leonidas J Guibas. Supervised fitting of geometric primitives to 3d point clouds. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2652–2660, 2019

    2019

  8. [16]

    Matrixcity: A large-scale city dataset for city-scale neural rendering and beyond

    Yixuan Li, Lihan Jiang, Linning Xu, Yuanbo Xiangli, Zhenzhi Wang, Dahua Lin, and Bo Dai. Matrixcity: A large-scale city dataset for city-scale neural rendering and beyond. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 3205–3215, 2023

    2023

  9. [17]

    Neuralangelo: High-fidelity neural surface reconstruction

    Zhaoshuo Li, Thomas Müller, Alex Evans, Russell H Taylor, Mathias Unberath, Ming-Yu Liu, and Chen-Hsuan Lin. Neuralangelo: High-fidelity neural surface reconstruction. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8456–8465, 2023

    2023

  10. [18]

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

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

    2024

  11. [19]

    Marching cubes: A high resolution 3d surface construction algorithm

    William E Lorensen and Harvey E Cline. Marching cubes: A high resolution 3d surface construction algorithm. InSeminal graphics: pioneering efforts that shaped the field, pages 347–353. 1998

    1998

  12. [20]

    Scaffold- gs: Structured 3d gaussians for view-adaptive rendering

    Tao Lu, Mulin Yu, Linning Xu, Yuanbo Xiangli, Limin Wang, Dahua Lin, and Bo Dai. Scaffold- gs: Structured 3d gaussians for view-adaptive rendering. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20654–20664, 2024

    2024

  13. [21]

    3dgsr: Implicit surface reconstruction with 3d gaussian splatting.ACM Transactions on Graphics (TOG), 43(6):1–12, 2024

    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 splatting.ACM Transactions on Graphics (TOG), 43(6):1–12, 2024

    2024

  14. [22]

    Nerf: Representing scenes as neural radiance fields for view synthesis

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

    2021

  15. [23]

    Instant neural graphics primitives with a multiresolution hash encoding.ACM transactions on graphics (TOG), 41(4):1– 15, 2022

    Thomas Müller, Alex Evans, Christoph Schied, and Alexander Keller. Instant neural graphics primitives with a multiresolution hash encoding.ACM transactions on graphics (TOG), 41(4):1– 15, 2022

    2022

  16. [24]

    Polyfit: Polygonal surface reconstruction from point clouds

    Liangliang Nan and Peter Wonka. Polyfit: Polygonal surface reconstruction from point clouds. InProceedings of the IEEE International Conference on Computer Vision (ICCV), Oct 2017

    2017

  17. [25]

    Octree- gs: Towards consistent real-time rendering with lod-structured 3d gaussians.arXiv preprint arXiv:2403.17898, 2024

    Kerui Ren, Lihan Jiang, Tao Lu, Mulin Yu, Linning Xu, Zhangkai Ni, and Bo Dai. Octree- gs: Towards consistent real-time rendering with lod-structured 3d gaussians.arXiv preprint arXiv:2403.17898, 2024

    2024 arXiv

  18. [26]

    Permutosdf: Fast multi-view reconstruction with implicit surfaces using permutohedral lattices

    Radu Alexandru Rosu and Sven Behnke. Permutosdf: Fast multi-view reconstruction with implicit surfaces using permutohedral lattices. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8466–8475, 2023. 11

    2023

  19. [27]

    Structure-from-motion revisited

    Johannes Lutz Schönberger and Jan-Michael Frahm. Structure-from-motion revisited. In Conference on Computer Vision and Pattern Recognition (CVPR), 2016

    2016

  20. [28]

    Pixelwise view selection for unstructured multi-view stereo

    Johannes Lutz Schönberger, Enliang Zheng, Marc Pollefeys, and Jan-Michael Frahm. Pixelwise view selection for unstructured multi-view stereo. InEuropean Conference on Computer Vision (ECCV), 2016

    2016

  21. [29]

    Parsenet: A parametric surface fitting network for 3d point clouds

    Gopal Sharma, Difan Liu, Subhransu Maji, Evangelos Kalogerakis, Siddhartha Chaudhuri, and Radomír M ˇech. Parsenet: A parametric surface fitting network for 3d point clouds. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, P...

    2020

  22. [30]

    Gpu-based ray-casting of quadratic surfaces

    Christian Sigg, Tim Weyrich, Mario Botsch, and Markus H Gross. Gpu-based ray-casting of quadratic surfaces. InPBG@ SIGGRAPH, pages 59–65, 2006

    2006

  23. [31]

    Concise plane arrangements for low-poly surface and volume modelling

    Raphael Sulzer and Florent Lafarge. Concise plane arrangements for low-poly surface and volume modelling. InEuropean Conference on Computer Vision, pages 357–373. Springer, 2024

    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

    2021 arXiv

  25. [33]

    Neus2: Fast learning of neural implicit surfaces for multi-view reconstruction

    Yiming Wang, Qin Han, Marc Habermann, Kostas Daniilidis, Christian Theobalt, and Lingjie Liu. Neus2: Fast learning of neural implicit surfaces for multi-view reconstruction. InPro- ceedings of the IEEE/CVF International Conference on Computer Vision, pages 3295–3306, 2023

    2023

  26. [34]

    Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4):600– 612, 2004

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4):600– 612, 2004

    2004

  27. [35]

    VR-NeRF: High-fidelity virtualized walkable spaces

    Linning Xu, Vasu Agrawal, William Laney, Tony Garcia, Aayush Bansal, Changil Kim, Samuel Rota Bulò, Lorenzo Porzi, Peter Kontschieder, Aljaž Božiˇc, Dahua Lin, Michael Zollhöfer, and Christian Richardt. VR-NeRF: High-fidelity virtualized walkable spaces. InSIGGRAPH Asia Confer...

    2023

  28. [36]

    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. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8296–8306, 2023

    2023

  29. [37]

    Point-nerf: Point-based neural radiance fields

    Qiangeng Xu, Zexiang Xu, Julien Philip, Sai Bi, Zhixin Shu, Kalyan Sunkavalli, and Ulrich Neumann. Point-nerf: Point-based neural radiance fields. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5438–5448, 2022

    2022

  30. [38]

    Depth anything v2.Advances in Neural Information Processing Systems, 37:21875– 21911, 2024

    Lihe Yang, Bingyi Kang, Zilong Huang, Zhen Zhao, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything v2.Advances in Neural Information Processing Systems, 37:21875– 21911, 2024

    2024

  31. [39]

    Stablenormal: Reducing diffusion variance for stable and sharp normal.ACM Transactions on Graphics (TOG), 43(6):1–18, 2024

    Chongjie Ye, Lingteng Qiu, Xiaodong Gu, Qi Zuo, Yushuang Wu, Zilong Dong, Liefeng Bo, Yuliang Xiu, and Xiaoguang Han. Stablenormal: Reducing diffusion variance for stable and sharp normal.ACM Transactions on Graphics (TOG), 43(6):1–18, 2024

    2024

  32. [40]

    Scannet++: A high- fidelity dataset of 3d indoor scenes

    Chandan Yeshwanth, Yueh-Cheng Liu, Matthias Nießner, and Angela Dai. Scannet++: A high- fidelity dataset of 3d indoor scenes. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 12–22, 2023

    2023

  33. [41]

    Finding good configurations of planar primitives in unorganized point clouds

    Mulin Yu and Florent Lafarge. Finding good configurations of planar primitives in unorganized point clouds. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6367–6376, 2022. 12

    2022

  34. [42]

    Gsdf: 3dgs meets sdf for improved neural rendering and reconstruction.Advances in Neural Information Processing Systems, 37:129507–129530, 2024

    Mulin Yu, Tao Lu, Linning Xu, Lihan Jiang, Yuanbo Xiangli, and Bo Dai. Gsdf: 3dgs meets sdf for improved neural rendering and reconstruction.Advances in Neural Information Processing Systems, 37:129507–129530, 2024

    2024

  35. [43]

    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

    2024

  36. [44]

    The unrea- sonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unrea- sonable effectiveness of deep features as a perceptual metric. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 586–595, 2018

    2018

  37. [45]

    Fast-livo2: Fast, direct lidar-inertial-visual odometry.IEEE Transactions on Robotics, 2024

    Chunran Zheng, Wei Xu, Zuhao Zou, Tong Hua, Chongjian Yuan, Dongjiao He, Bingyang Zhou, Zheng Liu, Jiarong Lin, Fangcheng Zhu, et al. Fast-livo2: Fast, direct lidar-inertial-visual odometry.IEEE Transactions on Robotics, 2024. 13 A Supplementary Material The following sections...

    2024

  38. [46]

    A triangle is considered visible if at least one of its three vertices is visible

    Triangle Preprocessing.We implement a custom CUDA kernel to process triangle soup. A triangle is considered visible if at least one of its three vertices is visible. For each visible triangle, we define a local coordinate frame (Eq. 2 in the main text) and compute the transfor...

  39. [47]

    For each visible subdivided triangle, we project its vertices onto the image plane to compute the number of tiles it overlaps

    Subdivided Triangle Processing.We further process the subdivided triangles by first validating their visibility: a subdivided triangle is marked as visible if at least one of its three vertices is visible. For each visible subdivided triangle, we project its vertices onto the ...

  40. [48]

    We then perform an efficient GPU-based Radix sort to order the subdivided triangles based on these keys

    Depth Sorting.Following the approach of 2DGS [ 9], we assign each subdivided triangle a sorting key that encodes its view-space depth and the ID of the tile it overlaps. We then perform an efficient GPU-based Radix sort to order the subdivided triangles based on these keys

  41. [49]

    Rendering.During rendering, while subdivided triangles are used for depth sorting, intersection calculations are performed on the original triangles. Following the 2DGS [9] approach, we define the camera ray corresponding to a pixel x= (x, y)T as the intersection of two 4D hom...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.