Pith. sign in

REVIEW 2 major objections 6 minor 1 cited by

Geometry Field Splatting with Gaussian Surfels

T0 review · 2 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read For a geometry field built from Gaussian surfels, each kernel's footprint along a ray has a closed form, making splatting nearly exact instead of approximate.

desk verdict A genuinely useful closed-form footprint and a good continuity idea, but the 'almost exact' and 'guarantee' claims are stronger than the implemented approximations support. read the letter →

arxiv 2411.17067 v2 pith:5IZNVDZK submitted 2024-11-26 cs.GR cs.CV

Kaiwen Jiang , Venkataram Sivaram , Cheng Peng , Ravi Ramamoorthi This is my paper · ORCID
classification cs.GRcs.CV MSC 68U0568T45
keywords geometryfieldGaussiansurfelsvolumesplattingsurfacereconstructionstochasticdifferentiablerenderingradiancefieldsspecularsurfaces
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

The paper claims that the stochastic geometry field used to describe opaque solids can be parameterized directly by flat 2D Gaussian kernels, or Gaussian surfels, and that the resulting volume-splatting renderer can be made almost exact rather than heuristic. The key step is a closed form for each kernel's footprint along a ray, $\rho_i = -2\ln\Psi(c - f_i)$, which lets the renderer drop the Taylor-expansion and ignored-self-occlusion approximations of classical volume splatting. The paper also shows how to make the rendered color a continuous function of kernel positions by blending kernel colors, so that optimization can drive the stochastic geometry from a fuzzy cloud to a sharp surface. In the paper's comparisons on the DTU and BlendedMVS datasets, the reconstructed meshes achieve the best average Chamfer distance among neural and splatting baselines, at splatting speed. If the paper is right, the fuzziness about what surface a Gaussian-splatting model represents is removed while rendering stays efficient.

What carries the argument

The load-bearing object is the footprint identity $\rho_i = -2\ln\Psi(c - f_i)$, derived by extruding each surfel's intersection line into a linear ramp of geometry-field values, from $f_i - c$ down to $-c$ along the surfel normal (Eq. 13). That ramp makes the geometry-field-to-density map integrable in closed form and yields a footprint whose range exceeds $[0,1]$, which is why the refined splatting sum with self-attenuation and no Taylor expansion (Eq. 7) is required. Two supporting mechanisms carry the rest: the fusion operator $\sqcup$ defined by $a \sqcup b = S^{-1}(S(a) + S(b))$ with $S(u) = -2\ln\Psi(c - u)$, which merges fully coincident same-color kernels into one kernel with additive footprint, and the color-blend rule (Eqs. 20-21) that makes coincident kernels share a color, rendering the output a continuous function of kernel positions (Theorem 2).

What would settle it

Render a converged Gaussian-surfel scene and compare two ray integrations: the color from Eq. (16) with per-ray sorting, and a brute-force numerical integration of the unmodified density field $\sigma(x)$ from Eq. (9) with fine quadrature. A disagreement beyond numerical precision would show that the extrusion model itself, not global sorting, is the remaining approximation, and the closed form would need revision for sharp kernels or large $c$.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the footprint of the $i$-th Gaussian surfel along a ray — the integrated density that sets its opacity — is computable in closed form as $\rho_i = -2\ln\Psi(c - f_i)$, where $f_i$ is the weighted kernel value at the ray intersection and $\Psi$ is the standard normal CDF. This follows from a local extrusion model in which the geometry field $F(x)$ decays linearly from $f_i - c$ to $-c$ over a small width $h$ along the surfel normal, a modification that makes the density integral tractable and is equivalent to the unextruded case as $h \to 0$. Substituting this footprint into the refined splatting sum $C = \sum_i c_i(1-\exp(-\rho_i))\prod_{j<i}\exp(-\rho_j)$, which keeps self-attenuation and discards the Taylor expansion, renders the density field induced by the geometry field almost exactly. The paper proves (Theorem 1) that rendering is exact whenever kernels along a ray are disjoint or fully coincident with equal color, and argues that with the color blend of Eq. (20) forcing equal color at coincident intersections, the only remaining approximation is global sorting in place of per-ray sorting.

Load-bearing premise

The closed-form footprint rests on the assumption that the geometry field near a surfel decays linearly from $f_i - c$ down to $-c$ along the surfel normal over a small width $h$; that linear profile is a modeling choice made to keep the integral tractable, not a consequence of the Gaussian surfel representation, and a different decay would change $\rho_i$.

Editorial extensions

If this is right

  • Footprints may exceed 1 without breaking the renderer, because Eq. (7) keeps self-attenuation and drops the Taylor expansion, so the splatting sum is a faithful evaluation of the defined density field rather than a composite of heuristics.
  • Coincident kernels are forced to share a color, which makes the rendered color (and hence the view-synthesis loss) a continuous function of kernel positions, allowing stochastic geometry to settle onto a deterministic surface during optimization.
  • Expected depth is available in closed form (Eq. 18) for the depth-normal consistency loss, and the method reports the best average Chamfer distance among compared baselines on DTU and BlendedMVS with training times of roughly 10-11 minutes.
  • For specular surfaces, replacing SH-encoded colors with latent vectors decoded from SH-encoded ray and reflected-ray directions closes holes that appear with SH colors (e.g., DTU scan110 and BlendedMVS 'Dog' and 'Clock').
  • The paper's own ablations show per-ray sorting improves geometry slightly (0.57-0.61 versus 0.58-0.63) at 3-4x time and memory, and its supplementary evaluation reports that the geometry-focused losses reduce novel-view synthesis quality on Mip-NeRF 360 relative to general splatting methods, so the method trades view synthesis for geometry.

Reading between the lines

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

  • The linear-decay extrusion is a modeling assumption rather than a derived identity; numerically integrating the unmodified density near a recovered surface would reveal whether real optimized fields actually conform to that profile or whether some of the reported gain is an artifact of it.
  • The closed form is tied to the normal CDF; the same derivation would go through with any CDF whose density-to-survival ratio is simple, so other stochastic geometry models could inherit the almost-exact splatting result.
  • The color blend is computed with k-nearest surfels in 3D rather than per ray, so the continuity guarantee (Theorem 2) holds only approximately in the implemented system; a true per-ray blend would isolate how much of the quality gain comes from this remedy.
  • The paper states its model targets opaque solids, so transparent and fuzzy objects are outside its scope; a natural stress test is whether the continuity guarantee degrades gracefully when the opaque-solid density mapping is violated.
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

2 major / 6 minor

Summary. The paper proposes a method for surface reconstruction from calibrated images by parameterizing a stochastic geometry field with Gaussian surfels and rendering it via a refined volume-splatting algorithm. The main contributions are (i) a closed-form footprint ρ_i = -2 ln Ψ(c - f_i) (Eq. 15) for the geometry-field density, derived from a linear-decay extrusion model; (ii) a reformulation of the splatting equation (Eq. 7) that avoids the Taylor expansion and self-occlusion approximations of classic volume splatting; (iii) a color-blending scheme (Eqs. 20-21) intended to make the rendered color continuous under kernel reordering; and (iv) a latent MLP-based color representation for specular surfaces. Experiments on DTU, BlendedMVS, and Mip-NeRF 360 report improved Chamfer distances over 2DGS, RaDe-GS, and neural baselines, with ablations showing each component contributes.

Significance. If the claims hold, the paper is a valuable step toward combining the efficiency of Gaussian splatting with a clearly defined surface geometry. The derivation of the footprint integral is elegant and, as the supplementary shows, the result depends only on the boundary values of the geometry field along the ray, making it robust to the specific decay profile. The authors are transparent about the global-sorting approximation and provide a per-ray-sorting ablation. The reported gains on DTU/BlendedMVS are consistent, and the supplementary tables include full per-scene results. However, the central 'almost exact' and 'continuity guarantee' claims are stated more strongly than what the implemented algorithm actually delivers, which requires revision.

major comments (2)
  1. [Sec. 4.2, Theorem 2, Eqs. (20)-(21)] The theorem proves continuity of the rendered color for the per-ray color blend in Eq. (20) with per-ray sorting, but the implementation described in Sec. 4.2 and B.2 replaces this with the 3D-space blend in Eq. (21) over the k=10 nearest centers, and the default renderer uses global sorting rather than per-ray sorting. Under global sorting, order swaps occur when camera-space center depths cross, not when ray intersection depths coincide, and at such swaps the blended colors bc_i and bc_j are generally different, so the rendered color can jump; the k-closest neighbor set also changes discretely when centers move. The ablation in Table 4 (SH: 0.63 vs 0.61 with per-ray sorting; latent: 0.58 vs 0.57) shows that the global-sorting approximation is not negligible. The abstract's claim of 'guarantee[ing] that the rendered color is a continuous function ... irrespective of ordering' is therefore not established for the algorithm actually evaluated. Please either state the continuity theorem for the per-ray variant only, analyze the approximate scheme (e.g., bounded discontinuity magnitudes), or modify the implemented scheme to match the theorem.
  2. [Sec. 4.1, Sec. 5.3, B.1, Eqs. (38)-(41)] The paper repeatedly describes global sorting as 'the only approximation' (Sec. 4.1 Discussion, Sec. 5.3), but the actual footprint computation in the implementation replaces the closed form in Eq. (38) with the polynomial approximation ρ_i ≈ 0.03279 min{f_i,4.28}^{3.4} in Eq. (41). Figure 10 shows qualitative closeness but no error bound, and the polynomial is also used to clamp opacity at 0.99. Since the 'almost exact rendering' claim is a central contribution, this approximation should be listed alongside global sorting, or the implementation should use the exact form, or the error should be quantified and shown to be negligible. The statement in Sec. 5.3 that per-ray sorting is 'the only approximation' is factually incomplete.
minor comments (6)
  1. [Eq. (8) and surrounding text] The symbol rendered as 'NM' in Eq. (8) is presumably a custom boxed-plus or diamond operator; it should be typeset properly and defined consistently with the operator L introduced earlier in Sec. 4.1.
  2. [Sec. 4.1, Eq. (12)] The piecewise definition of F(x(t)) with 'ϵ → 0' is not a well-defined function; consider presenting it as a limit or as a distribution carefully.
  3. [Sec. 4.2, proof of Theorem 2] The proof sketch is only two sentences; a more explicit argument showing that Eq. (16) with Eq. (20) is continuous at ordering swaps (not just that bc_i = bc_j at ties) would strengthen the paper.
  4. [Sec. 2 and Sec. 5.3] The work of StopThePop [38] on view-consistent sorting is mentioned only in the supplementary; it should be cited in the main text where global sorting is discussed.
  5. [Sec. 5.3, Table 4] The ablation table reports only averaged Chamfer distances; pointing to the full per-scene results in Table 5 of the supplementary is helpful, but a brief statement on variance or statistical significance would make the comparisons more robust.
  6. [Sec. 4.3, Eq. (22)] The MLP takes both ω and ω_o as inputs; the notation ω is reused for the ray direction and later for the surfel normal-related weight in Eq. (6). This overloaded notation should be disambiguated.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the footprint derivation is a self-contained mathematical consequence of the defined geometry-field model; the polynomial and k-NN approximations are fit to analytic functions or chosen for efficiency, not to the evaluation target.

full rationale

The paper's central derivation, Eq. (15), follows from the explicitly stated local extrusion model (Eq. 13) via the logarithmic-derivative identity in Sec. A.2 of the supplementary, and no fitted quantity enters the closed form. The geometry-field density conversion (Eq. 6) is taken from the external prior work of Miller et al. [34], which is a cited starting point, not a self-citation. The constants c=3, tau=100, and k=10 are hyperparameters, and the polynomial approximation in Eq. (41) is fit to the already-defined analytic footprint function, not to the Chamfer-distance evaluation metric or to the rendered images being predicted. The continuity theorem (Theorem 2) is verified for the per-ray color blend of Eq. (20), whose defining property bci=bcj whenever ti=tj is constructed precisely to make the proof go through; this is a designed condition, not a circular import. The implementation replaces Eq. (20) with the 3D-space approximation of Eq. (21) and uses global sorting, so the formal continuity guarantee does not automatically transfer to the deployed system, and the paper's claim that global sorting is the 'only approximation' is an overstatement; however, this is a proof-to-implementation gap and a correctness risk, not a circular derivation. The extrusion ansatz in Sec. 4.1 is a modeling choice that makes the integral tractable, but the footprint limit depends only on the boundary values (-c outside, fi-c at the peak) for the single-peaked profile, so Eq. (15) is not secretly an input to its own derivation. No step in the claimed derivation chain reduces to its own output, and no load-bearing premise rests on a self-citation. Therefore the paper is self-contained against external benchmarks and merits a circularity score of 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 1 invented entities

The central derivation rests on the geometry-field density model from Miller et al. and on a new ad hoc linear-decay extrusion profile. The model uses several hand-chosen hyperparameters (c, tau, k, lambda1) and a polynomial fit, none fitted to the benchmark target. No new physical entities beyond the extrusion profile are introduced.

free parameters (5)
  • c = 3
    Manually chosen constant in the geometry-field-to-density conversion, g(u)=-2 ln Ψ(c-u). Chosen because Ψ(3) is approximately 0.999. Not fitted to data, but changes the opacity mapping and affects optimization.
  • tau (τ) = 100
    Bandwidth of the color-blending kernel in Eqs. 20 and 21. Chosen by hand; replaces the exact per-ray blend with a distance-weighted blend and is central to the continuity claim.
  • k (closest points) = 10
    Number of nearest Gaussian surfels used to approximate the color blend, chosen to fit GPU memory around 10 GB. This is an approximation to the exact per-ray blend and is not part of the theoretical continuity guarantee.
  • lambda1 (depth distortion weight) = 1000 (DTU), 10 (BlendedMVS), 0 (Mip-NeRF 360)
    Dataset-specific regularization weight in the loss, following the 2DGS convention. Affects how strongly surfels cluster and the final geometry quality.
  • polynomial coefficients for footprint approximation = 0.03279, exponent 3.4
    Numerical fit to approximate ρi = -2 ln(0.5 + 0.5 erf((3 - min(fi,4.28))/√2)) over [0,4.28]. Used to avoid vanishing gradients; approximates the exact mapping and is an additional implementation approximation.
assumptions (4)
  • domain assumption The stochastic geometry field F(x)=μ(x)s(x) with Gaussian-distributed G(x), occupancy o(x)=Ψ(F(x)), and the density conversion σ(x)=ψ(-F)/Ψ(-F) ||∇F|| (α|ω·n|+(1-α)/2) correctly models opaque solids.
    Inherited from Miller et al. [34]; the paper sets α=1 near the surface and builds the entire splatting on this density model (Eqs. 6 and 9).
  • ad hoc to paper The linear-decay extrusion profile of the geometry field near each surfel (Eq. 13, Fig. 4d) is the correct regularization of the discontinuous geometry field.
    Introduced in this paper to make the footprint integral analytically tractable; no external evidence that the geometry field actually decays this way. The closed-form footprint ρi=-2 ln Ψ(c-fi) depends on this profile.
  • domain assumption 2D Gaussian surfels intersect a ray in points, so the non-overlapping or fully-overlapping condition of Theorem 1 is satisfied in practice, and the training losses make clustered surfels coplanar within floating-point accuracy.
    Sec. 4.1 and supplementary A.2 discussion; this justifies the absence of partial-overlap sorting issues and supports the color-continuity analysis.
  • standard math Standard properties of the normal CDF and PDF, and the exponential integration used in the supplementary derivations.
    Basic calculus and properties of Ψ and ψ used in supplementary A.1-A.2; accepted as mathematical background.
invented entities (1)
  • Linear-decay extrusion profile (the expanded intersected line giving the surfel a 3D width h)
    purpose: Makes the footprint integral of the geometry-field density well-defined and yields the closed form ρi=-2 ln Ψ(c-fi).
    This is a modeling device introduced in this paper (Fig. 4d, Eq. 13). It has no independent falsifiable handle: the paper does not predict a measurable consequence of the profile, and the rendering is insensitive to h only for this linear profile. If the profile were different, the footprint formula would change.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Geometry Field Splatting with Gaussian Surfels." pith.science (2026). https://pith.science/paper/5IZNVDZK

@misc{pith2026241117067,
  author       = {Pith},
  title        = {Pith review of: Geometry Field Splatting with Gaussian Surfels},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5IZNVDZK}},
  note         = {Machine review of arXiv:2411.17067}
}
read the original abstract

Geometric reconstruction of opaque surfaces from images is a longstanding challenge in computer vision, with renewed interest from volumetric view synthesis algorithms using radiance fields. We leverage the geometry field proposed in recent work for stochastic opaque surfaces, which can then be converted to volume densities. We adapt Gaussian kernels or surfels to splat the geometry field rather than the volume, enabling precise reconstruction of opaque solids. Our first contribution is to derive an efficient and almost exact differentiable rendering algorithm for geometry fields parameterized by Gaussian surfels, while removing current approximations involving Taylor series and no self-attenuation. Next, we address the discontinuous loss landscape when surfels cluster near geometry, showing how to guarantee that the rendered color is a continuous function of the colors of the kernels, irrespective of ordering. Finally, we use latent representations with spherical harmonics encoded reflection vectors rather than spherical harmonics encoded colors to better address specular surfaces. We demonstrate significant improvement in the quality of reconstructed 3D surfaces on widely-used datasets.

Figures

Figures reproduced from arXiv: 2411.17067 by the authors.

Figure 1
Figure 1. We introduce a geometry reconstruction method from a [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Illustration of a ray intersecting with non-overlapping ker [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Overview of our algorithm. (a) We first use 2D Gaussians to parameterize the geometry field, F. These kernels are expected to cluster around the surface. (b) We then convert the geometry field into the density field σ and lastly, (c) we leverage our refined volume splatting algorithm for differentiable rendering. no kernels, F(x) = −c, such that v(x) ≈ 1. Therefore, c is a manually chosen large positive value. At th… view at source ↗
Figures from the paper (12 more)
Figure 5
Figure 5. Figure 5: Illustration of opacity value on different Gaussian surfels. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 4
Figure 4. Figure 4: (a) An illustration of intersecting the i th Gaussian surfel, whose normal vector is denoted as ni that is perpendicular to the plane containing the surfel, with a ray, whose direction is denoted as ω. The depth of the intersected point is denoted as ti, and the inters…
Figure 7
Figure 7. Figure 7: Demonstration of continuously changing the colors of [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: Qualitative comparison on DTU and BlendedMVS datasets. Non-obvious differences are highlighted by insets and red circles. [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: We demonstrate the differences between SH and latent [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Plotting of function values of F(x) = −2 ln(0.5 + 0.5erf( 3√−x 2 )) and G(x) = 0.03279x 3.4 over the range [0, 4.28]. the parameter due to the numerical error, which hinders the optimization. To alleviate the numerical error, we approximate Eqn. (40) with a polynomial…
Figure 11
Figure 11. Figure 11: Comparisons of normal maps and chamfer distances (de [PITH_FULL_IMAGE:figures/full_fig_p015_11.png]
Figure 12
Figure 12. Figure 12: Comparison of rendering of meshes extracted by our method (at the top), 2DGS trained with mean depth in the regularization (in [PITH_FULL_IMAGE:figures/full_fig_p017_12.png]
Figure 13
Figure 13. Figure 13: Comparison of rendering of meshes extracted by our method (at the top) and 2DGS (at the bottom) on the object cases of the [PITH_FULL_IMAGE:figures/full_fig_p018_13.png]
Figure 14
Figure 14. Figure 14: Comparison of rendering of meshes extracted by our method (at the top) and 2DGS (at the bottom) on the scene cases of BMVS [PITH_FULL_IMAGE:figures/full_fig_p019_14.png]
Figure 15
Figure 15. Figure 15: Rendering of meshes extracted by our method on the Tanks&Temples dataset. [PITH_FULL_IMAGE:figures/full_fig_p019_15.png]
Figure 16
Figure 16. Figure 16: Rendering of meshes extracted by our method on the Treehill, Garden, Bicycle and Stump scenes of MipNeRF 360 dataset. [PITH_FULL_IMAGE:figures/full_fig_p020_16.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. HaloGS: Loose Coupling of Compact Geometry and Gaussian Splats for 3D Scenes

    cs.CV 2025-05 conditional novelty 6.0 of 10

    A dual representation using learnable triangles plus neural Gaussians achieves competitive rendering and more compact geometric abstractions on common 3D scene benchmarks.

Reference graph

Works this paper leans on

79 extracted references · 62 canonical work pages · cited by 1 Pith paper

  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. In Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition, pages 5470–5479, 2022. 15

    work page 2022

  2. [2]

    Patchmatch stereo - stereo matching with slanted support windows

    Michael Bleyer, Christoph Rhemann, and Carsten Rother. Patchmatch stereo - stereo matching with slanted support windows. In British Machine Vision Conference, 2011. 2

    work page 2011

  3. [3]

    High-quality surface splatting on today’s gpus

    Mario Botsch, Alexander Hornung, Matthias Zwicker, and Leif Kobbelt. High-quality surface splatting on today’s gpus. In Proceedings of the Second Eurographics / IEEE VGTC Conference on Point-Based Graphics, page 17–24, Goslar, DEU, 2005. Eurographics Association. 16

    work page 2005

  4. [4]

    Riav-mvs: Recurrent-indexing an asymmetric volume for multi-view stereo

    Changjiang Cai, Pan Ji, Qingan Yan, and Yi Xu. Riav-mvs: Recurrent-indexing an asymmetric volume for multi-view stereo. In Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition, pages 919–928, 2023. 2

    work page 2023

  5. [5]

    Pgsr: Planar-based gaussian splatting for ef- ficient 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 ef- ficient and high-fidelity surface reconstruction. arXiv preprint arXiv:2406.06521, 2024. 2, 16

    arXiv 2024

  6. [6]

    Adaptive and Temporally Consistent Gaussian Surfels for Multi-view Dynamic Reconstruction

    Decai Chen, Brianne Oberson, Ingo Feldmann, Oliver Schreer, Anna Hilsmann, and Peter Eisert. Adaptive and temporally consistent gaussian surfels for multi-view dynamic recon- struction. arXiv preprint arXiv:2411.06602, 2024. 1

    work page Pith review arXiv 2024

  7. [7]

    3D Reconstruction with Fast Dipole Sums

    Hanyu Chen, Bailey Miller, and Ioannis Gkioulekas. 3d reconstruction with fast dipole sums. arXiv preprint arXiv:2405.16788, 2024. 1, 6, 7, 8

    work page Pith review arXiv 2024

  8. [8]

    A Volumetric Method for Building Complex Models from Range Images

    Brian Curless and Marc Levoy. A Volumetric Method for Building Complex Models from Range Images. Association for Computing Machinery, New York, NY , USA, 1 edition,

    work page

Show all 79 references
  1. [9]

    High-quality surface reconstruction using gaussian surfels

    Pinxuan Dai, Jiamin Xu, Wenxiang Xie, Xinguo Liu, Huamin Wang, and Weiwei Xu. High-quality surface reconstruction using gaussian surfels. In ACM SIGGRAPH 2024 Conference Papers, pages 1–11, 2024. 1, 2, 4, 7, 8, 13, 16

    2024

  2. [10]

    Improving neural im- plicit surfaces geometry with patch warping

    Franc ¸ois Darmon, B´en´edicte Bascle, Jean-Cl´ement Devaux, Pascal Monasse, and Mathieu Aubry. Improving neural im- plicit surfaces geometry with patch warping. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6260–6269, 2022. 2

    2022

  3. [11]

    Transmvsnet: Global context-aware multi-view stereo network with trans- formers

    Yikang Ding, Wentao Yuan, Qingtian Zhu, Haotian Zhang, Xiangyue Liu, Yuanjiang Wang, and Xiao Liu. Transmvsnet: Global context-aware multi-view stereo network with trans- formers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8585–8594,

  4. [12]

    Kd-mvs: Knowledge distil- lation based self-supervised learning for multi-view stereo

    Yikang Ding, Qingtian Zhu, Xiangyue Liu, Wentao Yuan, Haotian Zhang, and Chi Zhang. Kd-mvs: Knowledge distil- lation based self-supervised learning for multi-view stereo. In European conference on computer vision, pages 630–646. Springer, 2022. 2

    2022

  5. [13]

    Zheng Dong, Ke Xu, Yaoan Gao, Hujun Bao, Weiwei Xu, and Rynson W. H. Lau. Gaussian surfel splatting for live human performance capture. ACM Trans. Graph., 43(6), 2024. 1

    2024

  6. [14]

    Geo-neus: Geometry-consistent neural implicit surfaces learning for multi-view reconstruction

    Qiancheng Fu, Qingshan Xu, Yew Soon Ong, and Wenbing Tao. Geo-neus: Geometry-consistent neural implicit surfaces learning for multi-view reconstruction. Advances in Neural Information Processing Systems, 35:3403–3416, 2022. 2

    2022

  7. [15]

    Accurate, dense, and robust multiview stereopsis

    Yasutaka Furukawa and Jean Ponce. Accurate, dense, and robust multiview stereopsis. IEEE Transactions on Pattern Analysis and Machine Intelligence, 32(8):1362–1376, 2010. 2

    2010

  8. [16]

    Ref-neus: Ambiguity-reduced neural implicit surface learning for multi-view reconstruction with reflection

    Wenhang Ge, Tao Hu, Haoyu Zhao, Shu Liu, and Ying-Cong Chen. Ref-neus: Ambiguity-reduced neural implicit surface learning for multi-view reconstruction with reflection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4251–4260, 2023. 2

    2023

  9. [17]

    Michael Goesele, Noah Snavely, Brian Curless, Hugues Hoppe, and Steven M. Seitz. Multi-view stereo for com- munity photo collections. In 2007 IEEE 11th International Conference on Computer Vision, pages 1–8, 2007. 2

    2007

  10. [18]

    Implicit geometric regularization for learning shapes

    Amos Gropp, Lior Yariv, Niv Haim, Matan Atzmon, and Yaron Lipman. Implicit geometric regularization for learning shapes. arXiv preprint arXiv:2002.10099, 2020. 1

    2002 arXiv

  11. [19]

    Cascade cost volume for high-resolution multi-view stereo and stereo matching

    Xiaodong Gu, Zhiwen Fan, Siyu Zhu, Zuozhuo Dai, Feitong Tan, and Ping Tan. Cascade cost volume for high-resolution multi-view stereo and stereo matching. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2495–2504, 2020. 2

    2020

  12. [20]

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

    Antoine Gu´edon and Vincent Lepetit. Sugar: Surface-aligned gaussian splatting for efficient 3d mesh reconstruction and high-quality mesh rendering. CVPR, 2024. 1, 2, 7, 8, 15

    2024

  13. [21]

    Stereo processing by semiglobal match- ing and mutual information

    Heiko Hirschmuller. Stereo processing by semiglobal match- ing and mutual information. IEEE Transactions on Pattern Analysis and Machine Intelligence, 30(2):328–341, 2008. 2

    2008

  14. [22]

    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. In ACM SIGGRAPH 2024 Conference Papers, pages 1–11, 2024. 1, 2, 4, 6, 7, 8, 13, 14, 15, 16

    2024

  15. [23]

    Large scale multi-view stereopsis eval- uation

    Rasmus Jensen, Anders Dahl, George V ogiatzis, Engil Tola, and Henrik Aanæs. Large scale multi-view stereopsis eval- uation. In 2014 IEEE Conference on Computer Vision and Pattern Recognition, pages 406–413, 2014. 7, 8, 14, 15

    2014

  16. [24]

    Gaussianshader: 3d gaussian splatting with shading functions for reflective surfaces

    Yingwenqi Jiang, Jiadong Tu, Yuan Liu, Xifeng Gao, Xiaox- iao Long, Wenping Wang, and Yuexin Ma. Gaussianshader: 3d gaussian splatting with shading functions for reflective surfaces. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5322–5332,

  17. [25]

    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 Trans. Graph., 42(4):139–1, 2023. 1, 2, 3, 4, 6, 7, 12, 13, 14, 15

    2023

  18. [26]

    Tanks and temples: Benchmarking large-scale scene reconstruction

    Arno Knapitsch, Jaesik Park, Qian-Yi Zhou, and Vladlen Koltun. Tanks and temples: Benchmarking large-scale scene reconstruction. ACM Transactions on Graphics, 36(4), 2017. 15

    2017

  19. [27]

    A survey on deep learning tech- niques for stereo-based depth estimation

    Hamid Laga, Laurent Valentin Jospin, Farid Boussaid, and Mohammed Bennamoun. A survey on deep learning tech- niques for stereo-based depth estimation. IEEE transactions on pattern analysis and machine intelligence , 44(4):1738– 1764, 2020. 2 9

    2020

  20. [28]

    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 IEEE Conference on Computer Vision and Pattern Recogni- tion (CVPR), 2023. 1, 2, 7, 8

    2023

  21. [29]

    Nero: Neural geometry and brdf reconstruction of reflective objects from multiview images

    Yuan Liu, Peng Wang, Cheng Lin, Xiaoxiao Long, Jiepeng Wang, Lingjie Liu, Taku Komura, and Wenping Wang. Nero: Neural geometry and brdf reconstruction of reflective objects from multiview images. ACM Transactions on Graphics (TOG), 42(4):1–22, 2023. 6

    2023

  22. [30]

    Neudf: Leaning neural unsigned distance fields with volume rendering

    Yu-Tao Liu, Li Wang, Jie Yang, Weikai Chen, Xiaoxu Meng, Bo Yang, and Lin Gao. Neudf: Leaning neural unsigned distance fields with volume rendering. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 237–247, 2023. 1, 2, 7, 8

    2023

  23. [31]

    Sparseneus: Fast generalizable neural surface reconstruction from sparse views

    Xiaoxiao Long, Cheng Lin, Peng Wang, Taku Komura, and Wenping Wang. Sparseneus: Fast generalizable neural surface reconstruction from sparse views. In European Conference on Computer Vision, pages 210–227. Springer, 2022. 2, 15

    2022

  24. [32]

    Neuraludf: Learning unsigned distance fields for multi-view reconstruction of surfaces with arbitrary topologies

    Xiaoxiao Long, Cheng Lin, Lingjie Liu, Yuan Liu, Peng Wang, Christian Theobalt, Taku Komura, and Wenping Wang. Neuraludf: Learning unsigned distance fields for multi-view reconstruction of surfaces with arbitrary topologies. In Pro- ceedings of the IEEE/CVF Conference on Compu...

    2023

  25. [33]

    Srinivasan, Matthew Tancik, Jonathan T

    Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthe- sis. In ECCV, 2020. 1, 2, 15

    2020

  26. [34]

    Objects as volumes: A stochastic geometry view of opaque solids

    Bailey Miller, Hanyu Chen, Alice Lai, and Ioannis Gkioulekas. Objects as volumes: A stochastic geometry view of opaque solids. In Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR), pages 87–97, 2024. 1, 2, 3, 4, 7, 8

    2024

  27. [35]

    tiny-cuda-nn, 2021

    Thomas M ¨uller. tiny-cuda-nn, 2021. 7

    2021

  28. [36]

    Instant neural graphics primitives with a multiresolu- tion hash encoding

    Thomas M¨uller, Alex Evans, Christoph Schied, and Alexander Keller. Instant neural graphics primitives with a multiresolu- tion hash encoding. ACM transactions on graphics (TOG), 41(4):1–15, 2022. 2, 15

    2022

  29. [37]

    Surfels: surface elements as rendering primi- tives

    Hanspeter Pfister, Matthias Zwicker, Jeroen van Baar, and Markus Gross. Surfels: surface elements as rendering primi- tives. In Proceedings of the 27th Annual Conference on Com- puter Graphics and Interactive Techniques, page 335–342, USA, 2000. ACM Press/Addison-Wesley Publis...

    2000

  30. [38]

    Stopthe- pop: Sorted gaussian splatting for view-consistent real-time rendering

    Lukas Radl, Michael Steiner, Mathias Parger, Alexander Weinrauch, Bernhard Kerbl, and Markus Steinberger. Stopthe- pop: Sorted gaussian splatting for view-consistent real-time rendering. ACM Transactions on Graphics (TOG) , 43(4): 1–17, 2024. 12, 16

    2024

  31. [39]

    An efficient repre- sentation for irradiance environment maps

    Ravi Ramamoorthi and Pat Hanrahan. An efficient repre- sentation for irradiance environment maps. In Proceedings of the 28th Annual Conference on Computer Graphics and Interactive Techniques, page 497–500, New York, NY , USA,

  32. [40]

    Accelerating 3d deep learning with pytorch3d

    Nikhila Ravi, Jeremy Reizenstein, David Novotny, Tay- lor Gordon, Wan-Yen Lo, Justin Johnson, and Georgia Gkioxari. Accelerating 3d deep learning with pytorch3d. arXiv:2007.08501, 2020. 7, 14

    2007 arXiv

  33. [41]

    Merf: Memory-efficient radiance fields for real-time view synthesis in unbounded scenes

    Christian Reiser, Rick Szeliski, Dor Verbin, Pratul Srinivasan, Ben Mildenhall, Andreas Geiger, Jon Barron, and Peter Hed- man. Merf: Memory-efficient radiance fields for real-time view synthesis in unbounded scenes. ACM Transactions on Graphics (TOG), 42(4):1–12, 2023. 15

    2023

  34. [42]

    Srinivasan, Dor Verbin, Richard Szeliski, Ben Mildenhall, Jonathan T

    Christian Reiser, Stephan Garbin, Pratul P. Srinivasan, Dor Verbin, Richard Szeliski, Ben Mildenhall, Jonathan T. Barron, Peter Hedman, and Andreas Geiger. Binary opacity grids: Capturing fine geometric detail for mesh-based view synthe- sis. SIGGRAPH, 2024. 15

    2024

  35. [43]

    Improving neural surface reconstruction with feature priors from multi-view image

    Xinlin Ren, Chenjie Cao, Yanwei Fu, and Xiangyang Xue. Improving neural surface reconstruction with feature priors from multi-view image. arXiv preprint arXiv:2408.02079,

    arXiv

  36. [44]

    Sch¨onberger, Enliang Zheng, Jan-Michael Frahm, and Marc Pollefeys

    Johannes L. Sch¨onberger, Enliang Zheng, Jan-Michael Frahm, and Marc Pollefeys. Pixelwise view selection for unstructured multi-view stereo. In Computer Vision – ECCV 2016, pages 501–518, Cham, 2016. Springer International Publishing. 2

    2016

  37. [45]

    Stereo matching using belief propagation

    Jian Sun, Nan-Ning Zheng, and Heung-Yeung Shum. Stereo matching using belief propagation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 25(7):787–800,

  38. [46]

    Torch-ngp: a pytorch implementation of instant-ngp, 2022

    Jiaxiang Tang. Torch-ngp: a pytorch implementation of instant-ngp, 2022. https://github.com/ashawkey/torch-ngp. 14

    2022

  39. [47]

    Ref-nerf: Struc- tured view-dependent appearance for neural radiance fields

    Dor Verbin, Peter Hedman, Ben Mildenhall, Todd Zickler, Jonathan T Barron, and Pratul P Srinivasan. Ref-nerf: Struc- tured view-dependent appearance for neural radiance fields. In 2022 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 5481–5490. IE...

    2022

  40. [48]

    Nerf-casting: Improved view-dependent appearance with con- sistent reflections

    Dor Verbin, Pratul P Srinivasan, Peter Hedman, Ben Milden- hall, Benjamin Attal, Richard Szeliski, and Jonathan T Barron. Nerf-casting: Improved view-dependent appearance with con- sistent reflections. In SIGGRAPH Asia 2024 Conference Papers, pages 1–10, 2024. 2, 6

    2024

  41. [49]

    High accuracy and visibility-consistent dense multiview stereo

    Hoang-Hiep Vu, Patrick Labatut, Jean-Philippe Pons, and Re- naud Keriven. High accuracy and visibility-consistent dense multiview stereo. IEEE Transactions on Pattern Analysis and Machine Intelligence, 34(5):889–901, 2012. 2

    2012

  42. [50]

    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

  43. [51]

    Unisdf: Unifying neural representations for high- fidelity 3d reconstruction of complex scenes with reflections

    Fangjinhua Wang, Marie-Julie Rakotosaona, Michael Niemeyer, Richard Szeliski, Marc Pollefeys, and Federico Tombari. Unisdf: Unifying neural representations for high- fidelity 3d reconstruction of complex scenes with reflections. Advances in Neural Information Processing System...

    2025

  44. [52]

    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. NeurIPS, 2021. 1, 2, 3

    2021

  45. [53]

    Efficient multi-view stereo by iterative dynamic cost volume

    Shaoqian Wang, Bo Li, and Yuchao Dai. Efficient multi-view stereo by iterative dynamic cost volume. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8655–8664, 2022. 2 10

    2022

  46. [54]

    Multi-view stereo in the deep learning era: A comprehensive review

    Xiang Wang, Chen Wang, Bing Liu, Xiaoqing Zhou, Liang Zhang, Jin Zheng, and Xiao Bai. Multi-view stereo in the deep learning era: A comprehensive review. Displays, 70: 102102, 2021. 2

    2021

  47. [55]

    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. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3295–3306, 2023. 1, 2, 7, 8

    2023

  48. [56]

    GS2Mesh: Sur- face reconstruction from Gaussian splatting via novel stereo views

    Yaniv Wolf, Amit Bracha, and Ron Kimmel. GS2Mesh: Sur- face reconstruction from Gaussian splatting via novel stereo views. In European Conference on Computer Vision, pages 207–224. Springer, 2025. 1

    2025

  49. [57]

    V oxurf: V oxel-based efficient and accurate neural surface reconstruction

    Tong Wu, Jiaqi Wang, Xingang Pan, Xudong Xu, Christian Theobalt, Ziwei Liu, and Dahua Lin. V oxurf: V oxel-based efficient and accurate neural surface reconstruction. arXiv preprint arXiv:2208.12697, 2022. 2

    2022 arXiv

  50. [58]

    Cl-mvsnet: Unsuper- vised multi-view stereo with dual-level contrastive learning

    Kaiqiang Xiong, Rui Peng, Zhe Zhang, Tianxing Feng, Jianbo Jiao, Feng Gao, and Ronggang Wang. Cl-mvsnet: Unsuper- vised multi-view stereo with dual-level contrastive learning. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3769–3780, 2023. 2

    2023

  51. [59]

    Mvsnet: Depth inference for unstructured multi-view stereo

    Yao Yao, Zixin Luo, Shiwei Li, Tian Fang, and Long Quan. Mvsnet: Depth inference for unstructured multi-view stereo. In Proceedings of the European conference on computer vi- sion (ECCV), pages 767–783, 2018. 2

    2018

  52. [60]

    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. Computer Vision and Pattern Recognition (CVPR), 2020. 7, 8, 15

    2020

  53. [61]

    Multiview neural surface reconstruction by disentangling geometry and appear- ance

    Lior Yariv, Yoni Kasten, Dror Moran, Meirav Galun, Matan Atzmon, Basri Ronen, and Yaron Lipman. Multiview neural surface reconstruction by disentangling geometry and appear- ance. Advances in Neural Information Processing Systems, 33:2492–2502, 2020. 1, 2

    2020

  54. [62]

    V ol- ume rendering of neural implicit surfaces.Advances in Neural Information Processing Systems, 34:4805–4815, 2021

    Lior Yariv, Jiatao Gu, Yoni Kasten, and Yaron Lipman. V ol- ume rendering of neural implicit surfaces.Advances in Neural Information Processing Systems, 34:4805–4815, 2021. 1, 2

    2021

  55. [63]

    Srinivasan, Richard Szeliski, Jonathan T

    Lior Yariv, Peter Hedman, Christian Reiser, Dor Verbin, Pratul P. Srinivasan, Richard Szeliski, Jonathan T. Barron, and Ben Mildenhall. Bakedsdf: Meshing neural sdfs for real- time view synthesis. In ACM SIGGRAPH 2023 Conference Proceedings, New York, NY , USA, 2023. Associati...

    2023

  56. [64]

    3d gaussian splatting with deferred reflection

    Keyang Ye, Qiming Hou, and Kun Zhou. 3d gaussian splatting with deferred reflection. In ACM SIGGRAPH 2024 Confer- ence Papers, pages 1–10, 2024. 6

    2024

  57. [65]

    Absgs: Recovering fine details in 3d gaussian splatting

    Zongxin Ye, Wenyu Li, Sidun Liu, Peng Qiao, and Yong Dou. Absgs: Recovering fine details in 3d gaussian splatting. In ACM Multimedia 2024, 2024. 7, 16

    2024

  58. [66]

    Fast-mvsnet: Sparse-to-dense multi-view stereo with learned propagation and gauss-newton refinement

    Zehao Yu and Shenghua Gao. Fast-mvsnet: Sparse-to-dense multi-view stereo with learned propagation and gauss-newton refinement. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1949–1958,

    1949

  59. [67]

    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 (CVPR), pages 19447– 19456, 2024. 15, 16

    2024

  60. [68]

    Gaussian opacity fields: Efficient high-quality compact surface recon- struction in unbounded scenes

    Zehao Yu, Torsten Sattler, and Andreas Geiger. Gaussian opacity fields: Efficient high-quality compact surface recon- struction in unbounded scenes. arXiv:2404.10772, 2024. 1, 2, 7, 8, 15, 16

    2024 arXiv

  61. [69]

    Rgb ↔x: Image decomposition and synthesis using material-and lighting-aware diffusion models

    Zheng Zeng, Valentin Deschaintre, Iliyan Georgiev, Yannick Hold-Geoffroy, Yiwei Hu, Fujun Luan, Ling-Qi Yan, and Miloˇs Haˇsan. Rgb ↔x: Image decomposition and synthesis using material-and lighting-aware diffusion models. In ACM SIGGRAPH 2024 Conference Papers, pages 1–11, 2024. 16

    2024

  62. [70]

    Rade-gs: Rasterizing depth in gaussian splatting

    Baowen Zhang, Chuan Fang, Rakesh Shrestha, Yixun Liang, Xiaoxiao Long, and Ping Tan. Rade-gs: Rasterizing depth in gaussian splatting. arXiv preprint arXiv:2406.01467, 2024. 1, 2, 7, 8, 15, 16

    2024 arXiv

  63. [71]

    Surfel-based gaussian inverse rendering for fast and relightable dynamic human reconstruction from monocular video

    Yiqun Zhao, Chenming Wu, Binbin Huang, Yihao Zhi, Chen Zhao, Jingdong Wang, and Shenghua Gao. Surfel-based gaussian inverse rendering for fast and relightable dynamic human reconstruction from monocular video. arXiv preprint arXiv:2407.15212, 2024. 1

    2024

  64. [72]

    Open3d: A modern library for 3d data processing

    Qian-Yi Zhou, Jaesik Park, and Vladlen Koltun. Open3d: A modern library for 3d data processing. arXiv preprint arXiv:1801.09847, 2018. 2, 7

    2018 arXiv

  65. [73]

    Deep learning for multi-view stereo via plane sweep: A survey

    Qingtian Zhu, Chen Min, Zizhuang Wei, Yisong Chen, and Guoping Wang. Deep learning for multi-view stereo via plane sweep: A survey. arXiv preprint arXiv:2106.15328, 2021. 2

    2021 arXiv

  66. [74]

    Zwicker, H

    M. Zwicker, H. Pfister, J. van Baar, and M. Gross. Ewa volume splatting. In Proceedings Visualization, 2001. VIS ’01., pages 29–538, 2001. 1, 2, 12, 13, 16 11 Geometry Field Splatting with Gaussian Surfels Supplementary Material A. Derivations A.1. Revisit Volume Splatting A.1...

    2001

  67. [76]

    and BlendedMVS dataset [60] between our method and 2DGS [22] in Fig. 12, Fig. 13, and Fig. 14. In the 2DGS algorithm, using the median depth in the regularization leads to better quantitative results, while using the mean depth in the regularization produces smoother geometry ...

  68. [77]

    but this dataset does not provide ground-truth camera pa- rameters, which requires an iterative-closest-point procedure to align the extracted mesh and ground-truth mesh. After manual inspection, we find that the metrics heavily depend on how well this off-the-shelf alignment ...

  69. [78]

    It would be useful to apply the multi-view constraint with our clearly defined geometry, which may further improve the reconstruction

    shows that the multi-view stereo constraint could be helpful for the geometry reconstruction quality using the volume splatting representation. It would be useful to apply the multi-view constraint with our clearly defined geometry, which may further improve the reconstruction...

  70. [79]

    However, a different strategy which may fully solve this problem is welcome

    leads to smaller Gaussian surfels in general without losing the quality, which alleviates such a problem. However, a different strategy which may fully solve this problem is welcome. Mesh Extraction. We follow [22] to use the TSDF fusion to extract the mesh. Even though it giv...

  71. [2001]

    Association for Computing Machinery. 2, 6

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.