Pith. sign in

REVIEW 3 major objections 5 minor 59 references

DMesh++: An Efficient Differentiable Mesh for Complex Shapes

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read The paper claims that replacing weighted Delaunay triangulation with a Minimum-Ball condition cuts differentiable mesh tessellation from linear to logarithmic time, enabling reconstruction of complex 2D and 3D shapes.

desk verdict A genuinely faster differentiable mesh tessellation, but the O(log N) headline is not what the implementation delivers. read the letter →

arxiv 2412.16776 v2 pith:LAW3VEJ6 submitted 2024-12-21 cs.CV cs.GRcs.LG

Sanghyun Son , Matheus Gadelha , Yang Zhou , Matthew Fisher , Zexiang Xu , Yi-Ling Qiao , Ming C. Lin , Yi Zhou This is my paper · ORCID
classification cs.CVcs.GRcs.LG
keywords differentiablemeshMinimum-BallconditionDelaunaytriangulationnearestneighborsearchpointcloudreconstructionmulti-viewtopologyprobabilistic
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

DMesh++ sets out to remove the computational bottleneck that kept differentiable meshes—surfaces whose connectivity is learned by gradient descent rather than fixed in advance—from representing highly detailed shapes. The earlier DMesh formulation decided which faces exist by computing a weighted Delaunay triangulation, a sequential construction whose practical cost grows linearly with the number of points. DMesh++ replaces that global construction with a Minimum-Ball condition: a candidate face survives only if the smallest ball through its vertices contains no other point, which is decided by a single nearest-neighbor query per face. The paper argues this brings tessellation cost to $O(\log N)$ under GPU parallelization, and demonstrates the payoff on intricate 2D drawings, dense 3D point clouds, and multi-view images, with better reconstruction accuracy than prior differentiable mesh methods and a fraction of the time.

What carries the argument

The Minimum-Ball algorithm is the paper's central object. For each query face $F$, it computes the smallest ball whose circumference or surface passes through the face's vertices—via closed-form center formulas in 2D and 3D—and then finds the nearest point of the point set to the ball's center. The signed distance $d(B_F,P)$ between that point and the ball's boundary decides whether $F$ qualifies, and is passed through a scaled sigmoid $\sigma(d \cdot \alpha_{\min})$ to produce a face-existence probability. This single nearest-neighbor query replaces the whole weighted Delaunay triangulation of DMesh, converting a sequential $O(N)$ global construction into per-face queries that can be parallelized across the GPU; between optimization steps the algorithm caches nearest neighbors and periodically refreshes the candidate face list, so gradients flow through positions and real values without rebuilding the full structure each step.

What would settle it

Measure the wall-clock time of a single tessellation step on random point sets of size $N = 10^3$ through $10^6$, with the brute-force neighbor scan replaced by an exact accelerated search; if per-face query time does not stay near $O(\log N)$, or if total time scales as $O(N^2)$, the complexity claim fails. Separately, count the candidate faces $|F|$ during a running reconstruction: if $|F|$ grows linearly with $N$, the effective tessellation cost is $O(N \log N)$, not $O(\log N)$.

Watch

Extended reading notes

Core claim

DMesh++ claims that mesh connectivity can be made differentiable and scalable at the same time by substituting the Minimum-Ball condition for weighted Delaunay triangulation. For a candidate face $F$ with $d$ vertices in $d$-dimensional space, it computes the unique minimum bounding ball $B_F$ whose boundary passes through those vertices, then measures the signed distance from the ball's surface to the nearest other point. The face is real when that distance is positive (no other point lies strictly inside the ball) and every vertex carries a real value $\psi > 0.5$; the signed distance is mapped through a sigmoid to give a differentiable existence probability $\Lambda_{\min}(F)$. Because every face passing the Minimum-Ball condition also belongs to the ordinary Delaunay triangulation (Lemma 3.2), the resulting mesh inherits Delaunay's guarantees of no self-intersections and few thin triangles. Points carry only position, a real value, and optional extra features such as color, so optimizing these continuous quantities alone drives the discrete topology changes observed during reconstruction.

Load-bearing premise

The claimed logarithmic speedup rests on two premises: that the nearest point to a face's ball center can be found in logarithmic time by an accelerated search structure, and that the number of candidate faces does not grow as the point count grows—yet the reported implementation uses a brute-force neighbor scan and periodically rebuilds candidate faces with a full Delaunay triangulation.

Editorial extensions

If this is right

  • Reconstruction of complex shapes becomes practical at high resolution: DMesh++ reports handling 2D drawings with nearly a million edges and 3D point clouds with hundreds of thousands of points, where DMesh exhausts memory or time.
  • Tessellation is up to 32 times faster than DMesh in 3D while using up to 75% less GPU memory, so finer meshes fit in the same computational budget.
  • Because the Minimum-Ball condition selects a subset of Delaunay faces, reconstructed meshes avoid self-intersections and thin triangles without a separate post-processing pass.
  • Multi-view reconstruction recovers open and closed surfaces, including colored meshes, that can be used directly for downstream applications such as physics simulation.
  • Discrete topology changes emerge from optimizing continuous per-point features alone, so no explicit connectivity prediction or remeshing operator is needed inside the optimization loop.

Reading between the lines

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

  • The Minimum-Ball condition is a local version of Delaunay's empty-circumsphere property, so the expressible meshes form a subset of ordinary Delaunay meshes; shapes that demand strongly anisotropic or non-Delaunay connectivity may be under-represented no matter how the points are arranged.
  • The $O(\log N)$ asymptotics presuppose a true spatial-index query and a bounded candidate count; the paper's timings use a brute-force neighbor scan and a periodically rebuilt triangulation, so the measured speedups and the stated complexity class describe different quantities until the implementation changes.
  • The Reinforce-Ball procedure in the appendix—stochastic optimization of per-point existence probabilities with a log-derivative gradient—looks like a general template for differentiable mesh simplification that could extend to 3D and to learning connectivity priors.
  • The reported failure on real-world multi-view images is diagnosed as a rendering-model limitation, so joining the Minimum-Ball tessellation with a photorealistic renderer is the most direct route to practical use.
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

3 major / 5 minor

Summary. The paper introduces DMesh++, a differentiable mesh representation that replaces DMesh's Weighted Delaunay Triangulation (WDT) with a Minimum-Ball condition for face existence. The Minimum-Ball condition (Definition 3.1) declares a face present when its minimum bounding ball contains no other point, and Lemma 3.2 shows these faces are a subset of Delaunay faces. The paper claims an O(log N) tessellation complexity, lower memory use, and demonstrates 2D and 3D point-cloud reconstruction plus 3D multi-view reconstruction, comparing against PSR, VoroMesh, PoNQ, DMTet, FlexiCubes, GShell, and DMesh. The supplementary material adds implementation details, nearest-neighbor caching, a periodic query-face refresh that runs a full Delaunay triangulation, and an experimental Reinforce-Ball algorithm.

Significance. If the efficiency and reconstruction claims hold, the Minimum-Ball formulation is a useful step toward scalable differentiable meshes: the geometric lemma is correct, the method is benchmarked against external methods rather than only the authors' prior DMesh, the code and project page are promised, and the reported reconstruction metrics are often better than the baselines, especially on open surfaces. The measured wall-clock speedups over DMesh are plausible and the paper is transparent about remaining limitations such as non-manifoldness and real-image reconstruction. However, the central asymptotic claim in the abstract and Section 3.2 is not supported by the paper's own analysis or its implementation, and the actual pipeline still invokes full Delaunay triangulations. The practical contribution therefore needs to be separated from the unsupported O(log N) claim.

major comments (3)
  1. [Abstract and Sec. 3.2] The O(log N) complexity claim is not supported. Section 3.2 derives only O(|F| log |P|), and the reduction to O(log |P|) rests on parallelization and on the parenthetical assumption that |F| does not grow with |P|. Footnote 2 states that the implementation uses PyTorch3D's knn_points, which is a brute-force linear scan over all points for each query, so the per-face nearest-neighbor cost is O(|P|), not O(log |P|). Section 5.1 then chooses |F| = N query faces, making the total tessellation work O(N^2) even with an idealized logarithmic index. Parallelizing a fixed total amount of work changes wall-clock time on a given GPU but does not change the asymptotic complexity. The measured 'sub-linear' behavior up to 50K points and the sharper increase beyond that are consistent with brute-force GPU scans, not with O(log N) scaling. This claim must be removed or supported with an actual spatial-index implementation and an explicit accounting of |F|.
  2. [Algorithm 2, line 9 and Appendix 8.2.2] The statement that DMesh++ 'eliminates WDT' is contradicted by the reconstruction pipeline. Algorithm 2's Update-Query-Faces function performs a full Delaunay triangulation of the entire point set every n1 steps, and Step 3 (Appendix 8.2.3) explicitly computes the DT of the points and then checks which DT faces satisfy the Minimum-Ball condition. These operations have essentially the same cost class as the WDT that the paper claims to remove, and their cost is excluded from the O(log N) claim in Section 3.2. The paper should include these Delaunay computations in the complexity analysis, or reframe the contribution as reducing the cost of per-face probability evaluation given an externally supplied candidate-face set.
  3. [Definition 3.1 and Appendix 8.2.2] The theoretical tessellation function is not shown to be a complete tessellation. The paper concedes in Section 3.2 that Fmin is only a subset of Delaunay faces and may omit faces such as AB in Figure 5, so Fmin alone does not tessellate the convex domain. The actual reconstruction relies on candidate faces extracted from a full Delaunay triangulation (Appendix 8.2.2), not on Fmin alone. The relationship between the theoretical Minimum-Ball tessellation and the mesh generated in the experiments should be clarified, because the claim that DMesh++ retains DMesh's core advantage of a self-contained tessellation function depends on this point.
minor comments (5)
  1. [Sec. 5.1, Fig. 7] Figure 7 reports averages over 5 trials without error bars or per-trial values; given the small number of trials and the large speedup claims, the variability should be reported.
  2. [Footnote 4 and Sec. 5.1] The memory comparison is partly explained by the note that the 2D DMesh implementation is not CUDA-optimized; the '96% less GPU memory' claim should be stated with this caveat in the main text.
  3. [Sec. 3.2, footnote 3] The assumption that |F| does not grow exponentially with |P| is essential to the complexity claim but appears only in a footnote; it should be moved to the main text and justified with the actual query-face generation procedure.
  4. [Tables 3 and 5] The captions state that the best results for closed and open surfaces are highlighted in red and blue, but the printed tables do not show these colors unambiguously; a legend or explicit formatting marker would help.
  5. [Appendix 7.4] The nearest-neighbor caching introduces hyperparameters n0, n1, and K, but no ablation is provided for their effect on reconstruction quality or speed; a brief sensitivity study would strengthen the practical claims.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the Minimum-Ball derivation is self-contained, and the contested O(log N) claim is a correctness/implementation concern rather than a circular reduction.

full rationale

The paper's central derivation is Definition 3.1 (Minimum-Ball condition) together with the signed-distance formulation in Eqs. (3)-(5). The face probability is computed directly from point geometry and a sigmoid with a fixed coefficient derived from the initial grid density (Appendix 7.3), not from any fitted parameter that is later reported as a prediction. Lemma 3.2 is proven from the standard Delaunay characterization cited to Cheng et al., and the self-intersection and triangle-quality claims follow from the subset relation Fmin ⊆ Fdt; those are mathematical consequences, not imports of the authors' own prior results. The reconstruction losses (expected Chamfer distance, rendering L1, triangle quality, real-value regularization) are standard and are evaluated against external baselines (PSR, VoroMesh, PoNQ, DMTet, FlexiCubes, GShell, Remeshing) as well as the authors' prior DMesh; no benchmark result is a renamed fit. Self-citations to DMesh are used as a baseline and as a source of standard components, but the novel Minimum-Ball algorithm does not reduce to DMesh by construction. The strongest potential concern is the O(log N) complexity assertion in Sec. 3.2: the implementation uses PyTorch3D's brute-force kNN, the tessellation benchmark fixes |F| = N, and Algorithm 2 periodically performs a full Delaunay triangulation. These points undermine the asymptotic claim and are legitimate correctness risks, but they are not circularity: no equation in the paper is equal to its input by definition, and no fitted value is relabeled as a prediction. Therefore the appropriate circularity score is 0.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The central efficiency claim rests on the assumed NN oracle and bounded query-face count, not on fitted numerical values. The reconstruction pipeline, however, depends on many hand-set hyperparameters, some of which materially affect reported results.

free parameters (7)
  • lambda_real = 1e-4
    Weight for the real-value regularization loss in Eq. (11); set by hand in Step 1 of reconstruction.
  • epsilon_card = 1e-6 to 1e-5
    Cardinality loss weight in the Reinforce-Ball algorithm (Table 6); tuned per experiment.
  • cache refresh interval n1 = 50
    Steps between nearest-neighbor cache updates in 3D multi-view reconstruction (Appendix 7.4).
  • cache size K = 10
    Number of nearest neighbors cached per query face in multi-view experiments.
  • initial grid edge length = 3x point density (point cloud) / 0.05 (multi-view)
    Initial resolution of the regular grid; set per task.
  • learning rates = 0.3 / 0.001 / 0.01
    Optimization rates for real value and position updates, differing by task.
  • point retention thresholds = 0.01 and 0.5
    Thresholds for collecting points after Step 1 and for discarding points in Reinforce-Ball.
assumptions (4)
  • standard math A face is in Delaunay triangulation iff there exists an empty circumsphere; Minimum-Ball faces are a subset.
    Used in Lemma 3.2 to guarantee Fmin is a subset of Delaunay faces, giving self-intersection freedom and triangle quality.
  • domain assumption An O(log N) nearest-neighbor query oracle exists for the point set.
    The O(log N) complexity claim in Sec. 3.2 depends on logarithmic-time per-face NN queries; the paper does not supply such a data structure and the PyTorch3D kNN used is a linear scan.
  • ad hoc to paper The number of query faces |F| does not grow exponentially with N.
    Stated in footnote 3 to support the effective O(log N) complexity; if |F| scales with N, total tessellation work is larger.
  • domain assumption For multi-view reconstruction, lighting and camera parameters are fully known.
    Assumed in Sec. 5.2.3 and acknowledged as a limitation for real-world images in Appendix 9.4.2.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DMesh++: An Efficient Differentiable Mesh for Complex Shapes." pith.science (2026). https://pith.science/paper/LAW3VEJ6

@misc{pith2026241216776,
  author       = {Pith},
  title        = {Pith review of: DMesh++: An Efficient Differentiable Mesh for Complex Shapes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LAW3VEJ6}},
  note         = {Machine review of arXiv:2412.16776}
}
read the original abstract

Recent probabilistic methods for 3D triangular meshes capture diverse shapes by differentiable mesh connectivity, but face high computational costs with increased shape details. We introduce a new differentiable mesh processing method that addresses this challenge and efficiently handles meshes with intricate structures. Our method reduces time complexity from O(N) to O(log N) and requires significantly less memory than previous approaches. Building on this innovation, we present a reconstruction algorithm capable of generating complex 2D and 3D shapes from point clouds or multi-view images. Visit our project page (https://sonsang.github.io/dmesh2-project) for source code and supplementary material.

Figures

Figures reproduced from arXiv: 2412.16776 by the authors.

Figure 1
Figure 1. DMesh++ for complex 2D and 3D shapes. DMesh++ encodes all geometric and topological information into continuous point features. (a) By optimizing these point features, DMesh++ is able to reconstruct complex 2D drawings from sample points. (b) This approach is also applicable to 3D, where it reconstructs the complex geometric structure of DNA from a point cloud. (c) By incorporating additional color features, DMesh++… view at source ↗
Figure 2
Figure 2. Conceptual comparison of traditional mesh and variants of DMesh [42]. Traditional meshes employ a non￾differentiable, discrete data structure, F, to store vertex indices that define connectivity, whereas DMesh++ encodes connectiv￾ity and additional information into continuous point-wise fea￾tures, P. Mesh generated from DMesh++ avoids several degenera￾cies—such as self-intersections and thin triangles—that can com￾p… view at source ↗
Figure 5
Figure 5. Minimum-Ball condition in 2D. In the left, 2D Delau￾nay Triangulation (DT) of 6 points is given. In middle and right figure, we render BF for two faces (AB, DF) in blue. determine if a face F exists on the mesh: TDMesh(P, F) = (F ∈ Fwdt) ∧ (min p∈F Ψ(p) > 0.5). (1) DMesh++ introduces an alternative tessellation function for faster processing. By removing the need for WDT, we eliminate the WDT weight and represent ea… view at source ↗
Figures from the paper (17 more)
Figure 6
Figure 6. Figure 6: Reconstruction process for 3D multi-view colored images of a sculpture. In each stage, we optimize different per-point features: the position and the real (ψ), while the per-point color is refined at every stage. (Left) We display the meshes at each stage during the fi…
Figure 8
Figure 8. Figure 8: Qualitative comparison of 2D point cloud reconstruc￾tion results. The outputs of DMesh [42] and DMesh++ are ren￾dered in red and blue, respectively. Method CD(×10−6 )↓ # Verts. # Edges. Time (sec) DMesh [42] 1.97 2506 2245 30.39 DMesh++ 1.82 2862 2793 11.33 [PITH_FULL…
Figure 7
Figure 7. Figure 7: Comparison of tessellation cost. Our method computes face probabilities up to 16 times faster in 2D and 32 times faster in 3D than DMesh [42], while using up to 96% less GPU memory in 2D and 75% less in 3D. 5. Experiments This section presents our experimental results.…
Figure 9
Figure 9. Figure 9: Qualitative comparison of 3D point cloud reconstruc￾tion results for a toad sitting on a leaf. For each result, we render its diffuse image on the left, and view-point normals on the right. 5.2.1. 2D Point Cloud Reconstruction In this task, we aim to reconstruct a 2D m…
Figure 10
Figure 10. Figure 10: Qualitative comparison of 3D multi-view reconstruction results for open surface. Here we illustrate from back of an open surface model (the front view is rendered at the left top of (a)). Colors represent inside and outside facing surfaces. Geometric Accuacy Mesh Qual…
Figure 12
Figure 12. Figure 12: Physics simulation on a staircase reconstructed from multi-view images. We simulate the motion of bouncing balls directly on the mesh generated by DMesh++. 6. Conclusion We presented DMesh++, a probabilistic approach for ef￾ficient, differentiable mesh connectivity ha…
Figure 13
Figure 13. Figure 13: Common signed distance for a 2D (left) and 3D (right) face in (initial) regular grid. We compute the signed dis￾tance by subtracting the radius of the minimum bounding ball from the length of the red line. The red dot represents the center of the minimum bounding ball…
Figure 14
Figure 14. Figure 14: Role of visibility gradient in geometric optimization. In this experiment, we optimize the translation vector of the object by comparing its rendered image and the ground truth image on the left. Since the differentiable renderer of DMesh [42] does not implement visib…
Figure 15
Figure 15. Figure 15: Implementation of anti-aliasing in our differentiable renderer. On the left, we show the process of anti-aliasing: 1) Find pixels that overlap with the given triangle, 2) Find the area that each pixel overlaps with the given triangle, and 3) Determine the color of eac…
Figure 16
Figure 16. Figure 16: Grid structure to initialize real values in 2D (left) and 3D (right). Every face in the grid structure satisfies Minimum￾Ball condition (Definition 3.1). the Minimum-Ball condition (Definition 3.1), and initialize the point-wise real values (ψ) with additional feature…
Figure 17
Figure 17. Figure 17: Point insertion for removing undesirable face. (Left) To reconstruct the ground truth shape, we need to set the real value (ψ) of points A-E to 1. The point rendered with dotted line has real value of 0. Then, we observe unnecessary face BD exists. (Right) To remove t…
Figure 18
Figure 18. Figure 18: 2D point cloud reconstruction result for complex drawings. For each drawing, we report both the number of edges and the reconstruction time. For the Chinese drawing, we additionally render the “imaginary” part on the right to clearly illustrate its complexity. 9.3. 3D…
Figure 19
Figure 19. Figure 19: Qualitative comparison of 3D point cloud reconstruction results for a closed surface (vase). For each image, we render the view-point normal on the right, and the diffuse image on the left. Among the baseline methods that reconstruct watertight mesh from point clouds,…
Figure 20
Figure 20. Figure 20: Qualitative comparison of 3D multi-view reconstruction results for a closed surface (sculpture). We render the input diffuse and depth images alongside the ground truth image. For each image, we render the view-point normal on the left, and the diffuse image on the ri…
Figure 21
Figure 21. Figure 21: 3D reconstruction from real-world images in DTU dataset [15]. The input images are shown on left, and the recon￾structed mesh is shown on right. 10. Reinforce-Ball algorithm Here we introduce an experimental algorithm that further enhances DMesh++’s capability. As dis…
Figure 23
Figure 23. Figure 23: Overview of Reinforce-Ball Algorithm. Based on per-point existence probability (Φ(P)), we sample points for B number of batches (P i ). Here we use B = 4, and assume we are reconstructing shape “A”. The points with ψ = 1 are rendered in black, while those with ψ = 0 a…
Figure 24
Figure 24. Figure 24: Qualitative ablation studies on Reinforce-Ball al￾gorithm (for letter ‘Q’). We render “imaginary” (black) part and “real part” (red, blue) together. ometry. As described in Sec. 5.2.1, we conducted 2D point cloud reconstruction experiments on the font dataset. In Tab.…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 43 canonical work pages

  1. [1]

    Power diagrams: properties, algo- rithms and applications

    Franz Aurenhammer. Power diagrams: properties, algo- rithms and applications. SIAM Journal on Computing , 16 (1):78–96, 1987. 3

    work page 1987

  2. [2]

    Continuous skele- ton computation by voronoi diagram

    Jonathan W Brandt and V Ralph Algazi. Continuous skele- ton computation by voronoi diagram. CVGIP: Image under- standing, 55(3):329–338, 1992. 7

    work page 1992

  3. [3]

    Fast poisson disk sampling in arbitrary di- mensions

    Robert Bridson. Fast poisson disk sampling in arbitrary di- mensions. SIGGRAPH sketches, 10(1):1, 2007. 6

    work page 2007

  4. [4]

    Learn- ing to predict 3d objects with an interpolation-based differ- entiable renderer

    Wenzheng Chen, Huan Ling, Jun Gao, Edward Smith, Jaakko Lehtinen, Alec Jacobson, and Sanja Fidler. Learn- ing to predict 3d objects with an interpolation-based differ- entiable renderer. Advances in neural information processing systems, 32, 2019. 2

    work page 2019

  5. [5]

    Meshanything: Artist-created mesh generation with au- toregressive transformers

    Yiwen Chen, Tong He, Di Huang, Weicai Ye, Sijin Chen, Ji- axiang Tang, Xin Chen, Zhongang Cai, Lei Yang, Gang Yu, et al. Meshanything: Artist-created mesh generation with au- toregressive transformers. arXiv preprint arXiv:2406.10163,

    arXiv

  6. [6]

    Meshany- thing v2: Artist-created mesh generation with adjacent mesh tokenization

    Yiwen Chen, Yikai Wang, Yihao Luo, Zhengyi Wang, Zilong Chen, Jun Zhu, Chi Zhang, and Guosheng Lin. Meshany- thing v2: Artist-created mesh generation with adjacent mesh tokenization. arXiv preprint arXiv:2408.02555, 2024. 1, 2

    arXiv 2024

  7. [7]

    Neural dual contouring

    Zhiqin Chen, Andrea Tagliasacchi, Thomas Funkhouser, and Hao Zhang. Neural dual contouring. ACM Transactions on Graphics (TOG), 41(4):1–13, 2022. 5

    work page 2022

  8. [8]

    Delaunay mesh generation

    Siu-Wing Cheng, Tamal Krishna Dey, Jonathan Shewchuk, and Sartaj Sahni. Delaunay mesh generation . CRC Press Boca Raton, 2013. 4

    work page 2013

Show all 59 references
  1. [9]

    Meshlab: an open-source mesh processing tool

    Paolo Cignoni, Marco Callieri, Massimiliano Corsini, Mat- teo Dellepiane, Fabio Ganovelli, Guido Ranzuglia, et al. Meshlab: an open-source mesh processing tool. In Eurographics Italian chapter conference , pages 129–136. Salerno, Italy, 2008. 6

    2008

  2. [10]

    Objaverse: A universe of annotated 3d objects

    Matt Deitke, Dustin Schwenk, Jordi Salvador, Luca Weihs, Oscar Michel, Eli VanderBilt, Ludwig Schmidt, Kiana Ehsani, Aniruddha Kembhavi, and Ali Farhadi. Objaverse: A universe of annotated 3d objects. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Rec...

    2023

  3. [11]

    Variance reduction techniques for gradient estimates in rein- forcement learning

    Evan Greensmith, Peter L Bartlett, and Jonathan Baxter. Variance reduction techniques for gradient estimates in rein- forcement learning. Journal of Machine Learning Research, 5(9), 2004. 10

    2004

  4. [12]

    Meshudf: Fast and differentiable meshing of unsigned distance field networks

    Benoit Guillard, Federico Stella, and Pascal Fua. Meshudf: Fast and differentiable meshing of unsigned distance field networks. In European Conference on Computer Vision , pages 576–592. Springer, 2022. 2

    2022

  5. [13]

    2d gaussian splatting for geometrically ac- curate radiance fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically ac- curate radiance fields. In ACM SIGGRAPH 2024 conference papers, pages 1–11, 2024. 8

    2024

  6. [14]

    3D triangulations

    Cl ´ement Jamin, Sylvain Pion, and Monique Teillaud. 3D triangulations. In CGAL User and Reference Manual. CGAL Editorial Board, 5.6 edition, 2023. 3

    2023

  7. [15]

    Large scale multi-view stereopsis eval- uation

    Rasmus Jensen, Anders Dahl, George V ogiatzis, Engin Tola, and Henrik Aanæs. Large scale multi-view stereopsis eval- uation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 406–413, 2014. 7, 8

    2014

  8. [16]

    Dual contouring of hermite data

    Tao Ju, Frank Losasso, Scott Schaefer, and Joe Warren. Dual contouring of hermite data. In Proceedings of the 29th an- nual conference on Computer graphics and interactive tech- niques, pages 339–346, 2002. 2

    2002

  9. [17]

    Screened poisson sur- face reconstruction

    Michael Kazhdan and Hugues Hoppe. Screened poisson sur- face reconstruction. ACM Transactions on Graphics (ToG), 32(3):1–13, 2013. 6

    2013

  10. [18]

    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), 2023. 2, 8

    2023

  11. [19]

    Modular primitives for high-performance differentiable rendering

    Samuli Laine, Janne Hellsten, Tero Karras, Yeongho Seol, Jaakko Lehtinen, and Timo Aila. Modular primitives for high-performance differentiable rendering. ACM Transac- tions on Graphics (TOG), 39(6):1–14, 2020. 2, 3

    2020

  12. [20]

    Deep march- ing cubes: Learning explicit surface representations

    Yiyi Liao, Simon Donne, and Andreas Geiger. Deep march- ing cubes: Learning explicit surface representations. In Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 2916–2925, 2018. 2

    2018

  13. [21]

    Soft ras- terizer: A differentiable renderer for image-based 3d reason- ing

    Shichen Liu, Tianye Li, Weikai Chen, and Hao Li. Soft ras- terizer: A differentiable renderer for image-based 3d reason- ing. In Proceedings of the IEEE/CVF International Confer- ence on Computer Vision, pages 7708–7717, 2019. 2

    2019

  14. [22]

    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. 2

    2023

  15. [23]

    Ghost on the shell: An expressive representation of general 3d shapes

    Zhen Liu, Yao Feng, Yuliang Xiu, Weiyang Liu, Liam Paull, Michael J Black, and Bernhard Sch ¨olkopf. Ghost on the shell: An expressive representation of general 3d shapes. arXiv preprint arXiv:2310.15168, 2023. 2, 7, 6

    2023 arXiv

  16. [24]

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

    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 topolo- gies. In Proceedings of the IEEE/CVF Conference on Com- ...

  17. [25]

    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. InSem- inal graphics: pioneering efforts that shaped the field, pages 347–353. 1998. 2

    1998

  18. [26]

    V oromesh: Learning water- tight surface meshes with voronoi diagrams

    Nissim Maruani, Roman Klokov, Maks Ovsjanikov, Pierre Alliez, and Mathieu Desbrun. V oromesh: Learning water- tight surface meshes with voronoi diagrams. In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 14565–14574, 2023. 6

    2023

  19. [27]

    Ponq: a neural qem-based mesh representation

    Nissim Maruani, Maks Ovsjanikov, Pierre Alliez, and Math- ieu Desbrun. Ponq: a neural qem-based mesh representation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3647–3657, 2024. 6, 7

    2024

  20. [28]

    A level set theory for neural implicit evolution under 9 explicit flows

    Ishit Mehta, Manmohan Chandraker, and Ravi Ramamoor- thi. A level set theory for neural implicit evolution under 9 explicit flows. In European Conference on Computer Vision, pages 711–729. Springer, 2022. 2

    2022

  21. [29]

    Extracting triangular 3d models, materials, and lighting from images

    Jacob Munkberg, Jon Hasselgren, Tianchang Shen, Jun Gao, Wenzheng Chen, Alex Evans, Thomas M¨uller, and Sanja Fi- dler. Extracting triangular 3d models, materials, and lighting from images. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , ...

    2022

  22. [30]

    Scalable parallel programming with cuda: Is cuda the parallel programming model that application developers have been waiting for? Queue, 6(2):40–53, 2008

    John Nickolls, Ian Buck, Michael Garland, and Kevin Skadron. Scalable parallel programming with cuda: Is cuda the parallel programming model that application developers have been waiting for? Queue, 6(2):40–53, 2008. 5

    2008

  23. [31]

    Large steps in inverse rendering of geometry

    Baptiste Nicolet, Alec Jacobson, and Wenzel Jakob. Large steps in inverse rendering of geometry. ACM Transactions on Graphics (TOG), 40(6):1–13, 2021. 2

    2021

  24. [32]

    Unisurf: Unifying neural implicit surfaces and radiance fields for multi-view reconstruction

    Michael Oechsle, Songyou Peng, and Andreas Geiger. Unisurf: Unifying neural implicit surfaces and radiance fields for multi-view reconstruction. In International Con- ference on Computer Vision (ICCV), 2021. 2

    2021

  25. [33]

    Continuous remeshing for inverse render- ing

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

  26. [34]

    Deepsdf: Learning con- tinuous signed distance functions for shape representation

    Jeong Joon Park, Peter Florence, Julian Straub, Richard Newcombe, and Steven Lovegrove. Deepsdf: Learning con- tinuous signed distance functions for shape representation. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 165–174, 2019. 2

    2019

  27. [35]

    Automatic differentiation in pytorch

    Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Al- ban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in pytorch. 2017. 5

    2017

  28. [36]

    Illumination for computer generated pic- tures

    Bui Tuong Phong. Illumination for computer generated pic- tures. In Seminal graphics: pioneering efforts that shaped the field, pages 95–101. 1998. 7

    1998

  29. [37]

    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 preprint arXiv:2007.08501, 2020. 4

    2007 arXiv

  30. [38]

    Deep marching tetrahedra: a hybrid repre- sentation for high-resolution 3d shape synthesis

    Tianchang Shen, Jun Gao, Kangxue Yin, Ming-Yu Liu, and Sanja Fidler. Deep marching tetrahedra: a hybrid repre- sentation for high-resolution 3d shape synthesis. Advances in Neural Information Processing Systems , 34:6087–6101,

  31. [39]

    Flexible isosurface extraction for gradient-based mesh optimization

    Tianchang Shen, Jacob Munkberg, Jon Hasselgren, Kangxue Yin, Zian Wang, Wenzheng Chen, Zan Gojcic, Sanja Fidler, Nicholas Sharp, and Jun Gao. Flexible isosurface extraction for gradient-based mesh optimization. ACM Transactions on Graphics (TOG), 42(4):1–16, 2023. 2, 7, 6

    2023

  32. [40]

    Spacemesh: A continuous representation for learning man- ifold surface meshes

    Tianchang Shen, Zhaoshuo Li, Marc Law, Matan Atzmon, Sanja Fidler, James Lucas, Jun Gao, and Nicholas Sharp. Spacemesh: A continuous representation for learning man- ifold surface meshes. arXiv preprint arXiv:2409.20562 ,

    arXiv

  33. [41]

    Meshgpt: Generating triangle meshes with decoder-only transformers

    Yawar Siddiqui, Antonio Alliegro, Alexey Artemov, Ta- tiana Tommasi, Daniele Sirigatti, Vladislav Rosov, Angela Dai, and Matthias Nießner. Meshgpt: Generating triangle meshes with decoder-only transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Patte...

    2024

  34. [42]

    Dmesh: A differentiable represen- tation for general meshes

    Sanghyun Son, Matheus Gadelha, Yang Zhou, Zexiang Xu, Ming C Lin, and Yi Zhou. Dmesh: A differentiable represen- tation for general meshes. arXiv preprint arXiv:2404.13445,

    arXiv

  35. [43]

    Visualizing data using t-sne

    Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of machine learning research , 9 (11), 2008. 2

    2008

  36. [44]

    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. 2

    2021 arXiv

  37. [45]

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

    Yiming Wang, Qin Han, Marc Habermann, Kostas Dani- ilidis, Christian Theobalt, and Lingjie Liu. Neus2: Fast learning of neural implicit surfaces for multi-view recon- struction. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023

    2023

  38. [46]

    Neumanifold: Neural watertight manifold reconstruction with efficient and high- quality rendering support

    Xinyue Wei, Fanbo Xiang, Sai Bi, Anpei Chen, Kalyan Sunkavalli, Zexiang Xu, and Hao Su. Neumanifold: Neural watertight manifold reconstruction with efficient and high- quality rendering support. arXiv preprint arXiv:2305.17134,

    arXiv

  39. [47]

    Meshlrm: Large reconstruction model for high- quality mesh

    Xinyue Wei, Kai Zhang, Sai Bi, Hao Tan, Fujun Luan, Valentin Deschaintre, Kalyan Sunkavalli, Hao Su, and Zex- iang Xu. Meshlrm: Large reconstruction model for high- quality mesh. arXiv preprint arXiv:2404.12385, 2024. 2

    2024 arXiv

  40. [48]

    Simple statistical gradient-following al- gorithms for connectionist reinforcement learning

    Ronald J Williams. Simple statistical gradient-following al- gorithms for connectionist reinforcement learning. Machine learning, 8:229–256, 1992. 9, 10

    1992

  41. [49]

    V olume rendering of neural implicit surfaces

    Lior Yariv, Jiatao Gu, Yoni Kasten, and Yaron Lipman. V olume rendering of neural implicit surfaces. In Thirty- Fifth Conference on Neural Information Processing Systems,

  42. [50]

    Surf-d: High-quality surface generation for arbitrary topologies using diffusion models

    Zhengming Yu, Zhiyang Dou, Xiaoxiao Long, Cheng Lin, Zekun Li, Yuan Liu, Norman M ¨uller, Taku Komura, Marc Habermann, Christian Theobalt, et al. Surf-d: High-quality surface generation for arbitrary topologies using diffusion models. arXiv preprint arXiv:2311.17050, 2023. 2

    2023 arXiv

  43. [51]

    Clay: A controllable large-scale generative model for creat- ing high-quality 3d assets

    Longwen Zhang, Ziyu Wang, Qixuan Zhang, Qiwei Qiu, Anqi Pang, Haoran Jiang, Wei Yang, Lan Xu, and Jingyi Yu. Clay: A controllable large-scale generative model for creat- ing high-quality 3d assets. ACM Transactions on Graphics (TOG), 43(4):1–20, 2024. 2

    2024

  44. [52]

    Thingi10k: A dataset of 10,000 3d-printing models

    Qingnan Zhou and Alec Jacobson. Thingi10k: A dataset of 10,000 3d-printing models. arXiv preprint arXiv:1605.04797, 2016. 2, 5, 6, 7

    2016 arXiv

  45. [53]

    Fully convolutional mesh autoencoder using efficient spatially varying kernels

    Yi Zhou, Chenglei Wu, Zimo Li, Chen Cao, Yuting Ye, Jason Saragih, Hao Li, and Yaser Sheikh. Fully convolutional mesh autoencoder using efficient spatially varying kernels. Ad- vances in neural information processing systems , 33:9251– 9262, 2020. 2 10 DMesh++: An Efficient Di...

    2020

  46. [55]

    Details about Minimum-Ball algorithm 7.1. Algorithm Algorithm 1 Minimum-Ball 1: P, F ← Set of points and query faces 2: αmin ← Coefficient for sigmoid function 3: Bc F, Br F ← Compute-Minimum-Ball(P, F) 4: P nearest F ← Find-Nearest-Neighbor(Bc F, P) 5: d(BF, P) ← Br F − ||P n...

    2000

  47. [56]

    Details about Reconstruction Process In this section, we provide implementation details about our reconstruction process described in Sec. 4. Before delving into these details, we introduce the loss formulations for re- construction problems. 8.1. Loss Formulation Our final lo...

  48. [57]

    imaginary

    Experimental Details and Additional Results In this section, we outline the experimental settings used for the results in Sec. 5 and present additional results to support our claims. 9.1. Dataset Here, we provide details on the datasets described in Sec. 5.2. 9.1.1. Font We us...

    2000

  49. [58]

    As discussed in Sec

    Reinforce-Ball algorithm Here we introduce an experimental algorithm that further enhances DMesh++’s capability. As discussed in Sec. 3.1, DMesh++ no longer uses the per-point weights found in DMesh [42]. In DMesh, optimizing per-point weights helps control mesh complexity: st...

  50. [59]

    cardinality

    We optimize Φ(P) to do that. be: 0.2 + (1.0 − 0.2) · 1.0 = 1.0. (17) However, the alpha blending technique used here does not account for such dependencies, leading to a reduction in ac- cumulated opacity. This reduction artificially increases the reconstruction loss. To minim...

  51. [2024]

    1, 2, 3, 4, 5, 6, 7, 8, 10

Pith tools

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