Pith. sign in

REVIEW 2 major objections 5 minor 46 references

Towards Fully Dynamic Omnitrees: Moment-Conserving Anisotropic Compression With Wavelets

T0 review · 2 major / 5 minor · reviewed 2026-07-11 · grok-4.5

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

Pith's one-line read Omnitrees with wavelet-guided coarsening and downsplit store anisotropic 3D data in up to 28 imes less space than OpenVDB and beat isotropic rate limits.

desk verdict Solid algorithmic completion of omnitrees: coarsening + downsplit + Haar algebra, with real storage wins vs OpenVDB on public 3-D data. read the letter →

arxiv 2607.04881 v1 pith:GE64QKSQ submitted 2026-07-06 cs.DS cs.CGcs.GRcs.NAmath.NA

Theresa Pollinger , Masado Ishii , Jens Domke (RIKEN Center for Computational Science , Kobe , Japan) This is my paper
classification cs.DScs.CGcs.GRcs.NAmath.NA
keywords omnitreesanisotropicrefinementHaarwaveletsadaptivemeshlossycompressionOpenVDBspace-partitioningtreesmomentconservation
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

Omnitrees are space-partitioning trees that can refine different dimensions independently and at different depths. Earlier work only showed how to refine them; this paper supplies the missing coarsening and “downsplit” operations so the tree can be made fully adaptive. Haar wavelets sit naturally on the same hierarchy: their detail coefficients measure how much the stored function varies along each dimension, so they both decide where coarsening is safe and keep the integral (mass) exact. A simple heuristic that repeatedly downsplits the least-varying dimension then coarsens produces encodings that, on thousands of solid 3D shapes, need far fewer coefficients than the industry-standard OpenVDB structure and that grow more slowly than the classic surface-complexity bound. On a continuous cloud density the same process yields roughly five-fold compression with almost no visible change. The result is a practical route to moment-conserving anisotropic compression for graphics, simulation, and high-dimensional data.

What carries the argument

Wavelet-guided downsplit-coarsening: a node is coarsened wherever its detail coefficients fall below a threshold; multi-dimensional nodes are first “downsplit” along the dimension whose pure one-dimensional detail is smallest, exposing new sibling leaves that can then be fused.

What would settle it

On a fresh collection of watertight solid meshes or volumetric densities, count leaf coefficients after the published downsplit-coarsening loop versus after an exhaustive or dynamic-programming search for the globally sparsest omnitree; if the heuristic’s savings vanish or fall below OpenVDB, the central practical claim fails.

Watch

Extended reading notes

Core claim

Once coarsening and downsplit are added and driven by Haar wavelet coefficients, omnitrees become a fully dynamic, moment-preserving compressor that stores moderately anisotropic three-dimensional data in substantially less space than OpenVDB and that asymptotically outpaces the isotropic rate predicted by the quadtree/octree complexity theorem.

Load-bearing premise

Choosing only the single least-varying dimension at each multi-dimensional parent is enough to expose most of the remaining redundancy; the paper itself notes that many cross-level opportunities are still missed.

Editorial extensions

If this is right

  • Lossless storage of solid 3-D shapes can drop by more than an order of magnitude relative to OpenVDB while remaining exact.
  • Lossy volumetric compression can reach roughly 5 imes with mass conservation and little visual change.
  • Fully adaptive mesh refinement becomes feasible in 4–6 spatial dimensions where isotropic trees explode.
  • Higher-order multiwavelets can be substituted for Haar to conserve momentum or energy without redesigning the tree operations.
  • The same hierarchy can serve as a mixed-precision layout, keeping coarse scales in high precision and fine details in low precision.

Reading between the lines

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

  • The same downsplit-coarsening loop should compound with learned sparse representations (neural hierarchical codecs) rather than compete with them.
  • A fixed-depth forest-of-omnitrees layout analogous to existing high-performance octree forests would make the structure competitive for real-time and distributed simulation.
  • Space-frequency tiling algorithms that currently rely on isotropic or dyadic trees can be re-cast on omnitrees to obtain still sparser adaptive bases in higher dimensions.
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 / 5 minor

Summary. The paper extends omnitrees from refinement-only to fully adaptive compression by defining coarsening (negative markers that fuse children along selected dimensions) and downsplit (factoring multi-dimensional splits into cascades of lower-dimensional splits). It integrates Haar wavelets so that detail coefficients serve both as coarsening indicators and as a heuristic for choosing which dimension to downsplit (Eq. 18). Moment conservation follows by construction from the vanishing moments of the discarded wavelets. Algorithms are derived for coefficient remapping under both operations (§3–4). Empirical evaluation on 4 166 filtered Thingi10K solids (lossless, binary) and the Disney cloud volume (lossy, continuous) reports coefficient and blosc2 file-size reductions versus OpenVDB of up to 28 imes (Thingi) and ≈5× (cloud at ε=10^{-1} with acceptable visual quality), together with an asymptotic approximation rate that undercuts the classical quadtree/octree bound.

Significance. If the reported storage factors and the improved asymptotic rate hold under independent re-implementation, the work supplies a concrete, moment-conserving anisotropic alternative to isotropic hierarchical volumes for moderately anisotropic 3-D data and a clear path toward higher-dimensional AMR. Strengths that raise confidence include: (i) explicit, reproducible comparison against a widely used baseline (OpenVDB) under identical voxelizations, (ii) public code and intermediate data, (iii) a clean derivation of the wavelet coefficient transformations under coarsening and downsplit, and (iv) an a-posteriori L1 error bound that is tight enough to be useful. The single-dimension downsplit heuristic is acknowledged as incomplete; the paper does not claim optimality, only that the heuristic already yields the stated empirical gains.

major comments (2)
  1. §5.1 and Fig. 4: the asymptotic claim that downsplit omnitrees “exceed theoretical expectations” rests on a single measured rate r_a(7)=0.51 versus the quadtree prediction ≈0.60. Because only levels ℓ=2….7 are shown and the sampling artifact at ℓ=2,3 is large, it is unclear whether the slope has stabilized. A short table of successive rates r_a(ℓ) for ℓ=5,6,7 (and ideally one higher resolution if feasible) would make the asymptotic statement falsifiable rather than visual.
  2. §4.4.2 / Eq. (18): the dimension-selection heuristic always downsplits the pure one-dimensional detail of smallest magnitude. The paper itself notes that this misses cross-level opportunities and that optimal selection is infeasible. While the empirical claim does not require optimality, a brief ablation (e.g., random dimension choice, or greedy multi-dimension downsplit on a subset of Thingi models) would quantify how much of the reported 2–4 imes extra saving is attributable to the particular heuristic versus the mere availability of the downsplit primitive.
minor comments (5)
  1. §3.2: the restriction to last-level coarsening markers is stated clearly, yet the text later speaks of “fully dynamic” omnitrees. A single clarifying sentence that multi-level coarsening remains future work would avoid over-claiming.
  2. Fig. 1 and Fig. 2: the 2-D worked example is helpful, but the mapping of wavelet coefficients onto the linearized descriptor (Fig. 2c) is dense; a short legend or color key would improve readability.
  3. §5.2: the cloud experiments use a fixed zero-pad to (128,128,256). Reporting the fraction of padded empty space would help readers judge how much of the OpenVDB baseline cost is structural overhead versus true data.
  4. Eq. (17): the a-priori L1 bound is correctly derived but is known to be loose; the a-posteriori aggregation mentioned in the text is never plotted. A single curve of realized versus predicted error would strengthen the error-analysis section.
  5. Typographical: “omnittree” (Fig. 1 caption), “coarseninng” (§6), and occasional missing spaces around × symbols.

Circularity Check

1 steps flagged · score 1.0 of 10

No significant circularity: empirical storage gains are measured against external OpenVDB baselines on public data; wavelet coarsening/downsplit follow from standard Haar algebra and the authors' prior refinement primitive without reducing the claims to inputs by construction.

  1. self citation load bearing [§3 (opening) and citations to [29]]
    "Before introducing the novel transformations coarsening (Section 4.4.1) and downsplit (Section 4.4.2), we briefly recapitulate the core concepts of omnitrees [29]: ... Refinement proceeds in four steps: ... The algorithm guarantees a valid omnitree ..."

    The base omnitree encoding, normalization, and refinement algorithm are taken from the authors' own prior paper [29]. This is ordinary self-citation of a data-structure definition and is not load-bearing for the new coarsening/downsplit results or the empirical compression claims, which are derived and measured independently; hence only a minor (score-1) contribution.

full rationale

The paper's strongest claims (up to 28× space savings vs OpenVDB on Thingi10K, ≈5× lossy on the Disney cloud, and asymptotic rate r_a(7)=0.51 undercutting the isotropic quadtree prediction ≈0.60) are empirical measurements of coefficient counts and file sizes after applying the newly defined coarsening (Def. 3.1) and downsplit (Def. 3.2) operations guided by Haar wavelet detail coefficients (Eqs. 16 and 18). These operations are derived from the tensor-product Haar transform (Eqs. 10–15) and the information mapping M; moment conservation follows immediately from orthogonality of the discarded wavelets (Eq. 8) and is not fitted. The single self-citation chain is to the authors' prior work [29] that supplies only the base omnitree encoding and the refinement algorithm; the present paper explicitly extends that encoding with new coarsening/downsplit and the wavelet integration, then validates against an independent external baseline (OpenVDB) on public datasets. No parameter is fitted to a subset and then re-presented as a prediction; no uniqueness theorem is imported to forbid alternatives; the downsplit heuristic (Eq. 18) is openly acknowledged as incomplete (§4.4, §6) rather than claimed optimal. The derivation is therefore self-contained against external benchmarks, with only the ordinary (non-load-bearing) self-citation of the authors' own prior definition of the data structure.

Assumptions & free parameters 2 free parameters · 3 assumptions · 2 invented entities

The central claims rest on standard multiresolution analysis, the authors’ earlier omnitree encoding, and a small set of algorithmic heuristics and thresholds chosen for the experiments. No new physical entities are postulated.

free parameters (2)
  • wavelet threshold ε = swept 1e-6 … 1
    User-chosen error density that decides which detail coefficients may be discarded; varied from 0 (lossless) to 1 in the cloud experiments.
  • downsplit dimension heuristic = arg min |w_τj|
    Always select the pure 1-D detail of smallest magnitude (Eq. 18); a free algorithmic choice not derived from optimality.
assumptions (3)
  • standard math Haar wavelets on dyadic rectangles are orthogonal and conserve the integral of the stored function.
    Invoked throughout §4 to justify moment conservation and the L1 error bound.
  • domain assumption An omnitree admits a unique normalized representation obtained by gathering splits as high as possible.
    Taken from the authors’ prior work [29] and used to guarantee that coarsening/downsplit produce well-defined trees.
  • ad hoc to paper Last-level coarsening markers with identical subtrees are sufficient for practical compression.
    Explicit restriction in §3.2 that simplifies the mapping construction; deeper markers are left for future work.
invented entities (2)
  • downsplit operation
    purpose: Reorders a multi-dimensional split into a cascade so that previously hidden sibling leaves become fusable.
    Defined in Def. 3.2; no independent existence outside the paper’s algorithmic framework.
  • coarsening markers (negative markers)
    purpose: Encode fusion requests that reverse refinement while preserving the information mapping needed for wavelet updates.
    Introduced in §3.2 as the dual of the earlier positive refinement markers.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Fully Dynamic Omnitrees: Moment-Conserving Anisotropic Compression With Wavelets." pith.science (2026). https://pith.science/paper/GE64QKSQ

@misc{pith2026260704881,
  author       = {Pith},
  title        = {Pith review of: Towards Fully Dynamic Omnitrees: Moment-Conserving Anisotropic Compression With Wavelets},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GE64QKSQ}},
  note         = {Machine review of arXiv:2607.04881}
}
abstract

Recently, omnitrees were introduced as a flexible space partitioning tree that improves upon the benefits of both octrees and k-d trees: Omnitrees' efficient encoding of anisotropic refinements holds particular interest for applications with anisotropic features and high dimensionality. These include, but are not limited to, computer graphics, databases, machine learning, and physics simulations. The present paper defines new operations on the omnitree encoding that extend its capabilities from the existing refinement to also include coarsening and therefore fully adaptive compression. It demonstrates natural integration of omnitrees with wavelets, which conserves moments of the stored function by design. For omnitrees, the wavelet coefficients can be interpreted as local refinement priorities, which can be used to guide the adaptation process. We derive algorithms for coarsening and downsplit that are guided by wavelet coefficients, and show their application to a large dataset of 3D shapes, as well as the continuous-valued density field of a cloud. The comparison to OpenVDB, a widely-used data structure for sparse volumetric data in computer graphics, enables a demonstration of the practical benefits of omnitrees even for moderately anisotropic three-dimensional data. Compared to OpenVDB, objects can be stored using up to 28x less space, and asymptotically show savings that exceed theoretical expectations. Using lossy compression, the cloud dataset can be compressed by $\approx5\times$ compared to OpenVDB, with negligible loss of visual quality. This demonstrates the potential of omnitrees for efficient storage and processing, and motivates further research into their applications in various domains.

Figures

Figures reproduced from arXiv: 2607.04881 by the authors.

Figure 1
Figure 1. Two-dimensional example for coarsening and downsplit + coarsen [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Illustrations of the equivalence of nodal and wavelet representations [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Cat object (Thingi 34965) at various resolution levels ( [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Boxplots and lineplots showing the number of stored binary coeffi [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]
Figure 5
Figure 5. Figure 5: Boxplots and lineplots showing the the file size required to store the [PITH_FULL_IMAGE:figures/full_fig_p019_5.png]
Figure 6
Figure 6. Figure 6: Disney cloud (one sixteenth) in its original and lossily compressed [PITH_FULL_IMAGE:figures/full_fig_p020_6.png]
Figure 7
Figure 7. Figure 7: Lossy compression errors, coefficients, and storage for different wavelet [PITH_FULL_IMAGE:figures/full_fig_p022_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

46 extracted references · 18 canonical work pages

  1. [1]

    A Class of Bases in L2 for the Sparse Representation of Integral Operators

    Bradley K. Alpert. “A Class of Bases in L2 for the Sparse Representation of Integral Operators”. In: SIAM Journal on Mathematical Analysis 24.1 (Jan. 1, 1993), pp. 246–262. issn: 0036-1410. doi: 10.1137/0524016. url: https://epubs.siam.org/doi/10.1137/0524016

    doi:10.1137/0524016 1993

  2. [2]

    AMM: Adaptive Multilinear Meshes

    Harsh Bhatia et al. “AMM: Adaptive Multilinear Meshes”. In: IEEE Transactions on Visualization and Computer Graphics 28.6 (June 2022), pp. 2350–2363. issn: 1941-0506. doi: 10.1109/TVCG.2022.3165392

    work page doi:10.1109/tvcg.2022.3165392 2022

  3. [3]

    Blender - The Free and Open Source 3D Creation Software

    Blender Online Community. Blender - The Free and Open Source 3D Creation Software. Version 5.1.2. Blender Institute, Amsterdam: Blender Foundation, 2026. url: http://www.blender.org

    2026

  4. [4]

    A fast, compressed and persistent data store library

    Blosc Development Team. A fast, compressed and persistent data store library. https://blosc.org. 2009-2026. 4will be published after review; preview link: https:// zenodo.org/records/19659695?preview=1&token=eyJhbGciOiJIUzUxMiJ9. eyJpZCI6Ijc1NzAwNGQ0LWQwOWQtNDMzOS05NDlkLTE3YWFhYzA0NDQ4NiIsImRhdGEiOnt9LCJyYW5kb20iOiI2ZmY2ZjcwOTYwY2RmNDdiZWQ5MjMwZGMxODAyZjl...

    arXiv 2009

  5. [5]

    Sparse Grids

    Hans-Joachim Bungartz and Michael Griebel. “Sparse Grids”. In: Acta Numerica 13 (May 2004), pp. 147–269. issn: 1474-0508, 0962-4929. doi: 10 . 1017 / S0962492904000182. url: https : / / www . cambridge . org / core/journals/acta-numerica/article/sparse-grids/47EA2993DB84C9D231BB96ECB26F615C

    2004

  6. [6]

    P4est: Scalable Algorithms for Parallel Adaptive Mesh Refinement on Forests of Octrees

    Carsten Burstedde, Lucas C. Wilcox, and Omar Ghattas. “P4est: Scalable Algorithms for Parallel Adaptive Mesh Refinement on Forests of Octrees”. In: SIAM Journal on Scientific Computing 33.3 (Jan. 2011), pp. 1103–

    2011

  7. [7]

    doi: 10.1137/100791634

    issn: 1064-8275. doi: 10.1137/100791634

    doi:10.1137/100791634

  8. [8]

    Nonconforming Mesh Refinement for High-Order Finite Elements

    Jakub ˇCerven´ y, Veselin Dobrev, and Tzanio Kolev. “Nonconforming Mesh Refinement for High-Order Finite Elements”. In: SIAM Journal on Scien- tific Computing (2019). url: https://epubs.siam.org/doi/10.1137/ 18M1193992

    2019

Show all 46 references
  1. [9]

    An Anisotropic hp-Adaptation Framework for Ultraweak Discontinuous Petrov–Galerkin Formulations

    Ankit Chakraborty, Stefan Henneking, and Leszek Demkowicz. “An Anisotropic hp-Adaptation Framework for Ultraweak Discontinuous Petrov–Galerkin Formulations”. In:Computers & Mathematics with Applications 167 (2024), pp. 315–327. issn: 0898-1221. doi: 10 . 1016 / j . camwa . 202...

    2024

  2. [10]

    Tree Approximation and Optimal Encoding

    Albert Cohen, Wolfgang Dahmen, Ingrid Daubechies, and Ronald DeVore. “Tree Approximation and Optimal Encoding”. In: Applied and Computa- tional Harmonic Analysis 11.2 (Sept. 1, 2001), pp. 192–226. issn: 1063-

    2001

  3. [11]

    url: https://www.sciencedirect

    doi: 10.1006/acha.2001.0336. url: https://www.sciencedirect. com/science/article/pii/S1063520301903367

    2001 doi

  4. [12]

    Biorthog- onal Bases of Compactly Supported Wavelets

    Albert Cohen, Ingrid Daubechies, and Jean-Christophe Feauveau. “Biorthog- onal Bases of Compactly Supported Wavelets”. In: Communications on Pure and Applied Mathematics 45.5 (1992), pp. 485–560. issn: 1097-0312. doi: 10.1002/cpa.3160450502 . url: https://onlinelibrary.wiley. ...

    1992 doi

  5. [13]

    Ten Lectures on Wavelets

    Ingrid Daubechies. Ten Lectures on Wavelets. SIAM, 1992

    1992

  6. [14]

    Splitflow - Progress in 3D CFD with Cartesian Omni-Tree Grids for Complex Geometries

    Neal Domel and Steve Karman. “Splitflow - Progress in 3D CFD with Cartesian Omni-Tree Grids for Complex Geometries”. In: 38th Aerospace Sciences Meeting and Exhibit . American Institute of Aeronautics and As- tronautics, 2000. doi: 10.2514/6.2000-1006

    doi

  7. [15]

    Sparse-Grid Finite-Volume Multigrid for 3D-problems

    Pieter W. Hemker. “Sparse-Grid Finite-Volume Multigrid for 3D-problems”. In: Advances in Computational Mathematics 4.1 (Dec. 1, 1995), pp. 83–

    1995

  8. [16]

    issn: 1572-9044. doi: 10 . 1007 / BF02123474. url: https : / / doi . org/10.1007/BF02123474

    doi

  9. [17]

    hp3D: A Scalable MPI/OpenMP hp-Adaptive Finite Element Software Library for Complex Multiphysics Applications

    Stefan Henneking et al. “ hp3D: A Scalable MPI/OpenMP hp-Adaptive Finite Element Software Library for Complex Multiphysics Applications”. In: Journal of Open Source Software 9.95 (2024), p. 5946. issn: 2475-9066. doi: 10.21105/joss.05946 . url: https://joss.theoj.org/papers/ 1...

    2024 doi

  10. [18]

    Mixed Element Trees: A Generalization of Modified Octrees for the Generation of Meshes for the Simulation of Complex 3-D Semiconductor Device Structures

    Nancy Hitschfeld, Paolo Conti, and Wolfgang Fichtner. “Mixed Element Trees: A Generalization of Modified Octrees for the Generation of Meshes for the Simulation of Complex 3-D Semiconductor Device Structures”. In: IEEE Transactions on Computer-Aided Design of Integrated Circui...

    1993 doi

  11. [19]

    RAPTOR: Practical Numerical Profiling of Scientific Applications

    Faveo Hoerold et al. “RAPTOR: Practical Numerical Profiling of Scientific Applications”. In: Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis . SC ’25. New York, NY, USA: Association for Computing Machinery, Nov. 15, ...

    2025 doi

  12. [20]

    T8code - Modular Adaptive Mesh Refinement in the Exascale Era

    Johannes Holke et al. “T8code - Modular Adaptive Mesh Refinement in the Exascale Era”. In: Journal of Open Source Software 10.106 (Feb. 6, 2025), p. 6887. issn: 2475-9066. doi: 10.21105/joss.06887

    2025 doi

  13. [21]

    Walt Disney Animation Studios Cloud Data Set

    Disney Enterprises Inc. Walt Disney Animation Studios Cloud Data Set

  14. [22]

    url: https://disneyanimation.com/resources/clouds/

  15. [23]

    NeuralVDB: High-resolution Sparse Volume Representation Using Hierarchical Neural Networks

    Doyub Kim, Minjae Lee, and Ken Museth. “NeuralVDB: High-resolution Sparse Volume Representation Using Hierarchical Neural Networks”. In: ACM Trans. Graph.43.2 (Feb. 28, 2024), 20:1–20:21.issn: 0730-0301. doi: 10.1145/3641817. url: https://dl.acm.org/doi/10.1145/3641817

    2024 doi

  16. [24]

    Multiskalen-basierte Finite-Differenzen-Verfahren auf adap- tiven d¨ unnen Gittern

    Frank Koster. “Multiskalen-basierte Finite-Differenzen-Verfahren auf adap- tiven d¨ unnen Gittern”. Thesis. Universit¨ ats- und Landesbibliothek Bonn,

  17. [25]

    500.11811/1696

    url: https://bonndoc.ulb.uni- bonn.de/xmlui/handle/20. 500.11811/1696

  18. [26]

    SZ3: A Modular Framework for Composing Prediction- Based Error-Bounded Lossy Compressors

    Xin Liang et al. “SZ3: A Modular Framework for Composing Prediction- Based Error-Bounded Lossy Compressors”. In: IEEE Transactions on Big Data 9.2 (Apr. 2023), pp. 485–498. issn: 2332-7790. doi: 10.1109/ TBDATA.2022.3201176. url: https://ieeexplore.ieee.org/document/ 9866018

    2023

  19. [27]

    Image Compression with Adap- tive Haar-Walsh Tilings

    Maj Lindberg and Lars F. Villemoes. “Image Compression with Adap- tive Haar-Walsh Tilings”. In: Wavelet Applications in Signal and Image Processing VIII. Vol. 4119. SPIE, 2000, pp. 911–921. url: https://www. spiedigitallibrary.org/conference- proceedings- of- spie/4119/ 0000/I...

  20. [28]

    A Theory for Multiresolution Signal Decompo- sition: The Wavelet Representation

    St´ ephane Georges Mallat. “A Theory for Multiresolution Signal Decompo- sition: The Wavelet Representation”. In: IEEE Transactions on Pattern Analysis and Machine Intelligence 11.7 (July 1989), pp. 674–693. issn: 1939-3539. doi: 10.1109/34.192463. url: https://ieeexplore.ieee...

    1989 doi

  21. [29]

    NanoVDB: A GPU-Friendly and Portable VDB Data Struc- ture For Real-Time Rendering And Simulation

    Ken Museth. “NanoVDB: A GPU-Friendly and Portable VDB Data Struc- ture For Real-Time Rendering And Simulation”. In: ACM SIGGRAPH 2021 Talks. SIGGRAPH ’21. New York, NY, USA: Association for Com- puting Machinery, Aug. 6, 2021, pp. 1–2. isbn: 978-1-4503-8373-8. doi: 10.1145/345...

    2021 doi

  22. [30]

    VDB: High-resolution Sparse Volumes with Dynamic Topol- ogy

    Ken Museth. “VDB: High-resolution Sparse Volumes with Dynamic Topol- ogy”. In: ACM Trans. Graph. 32.3 (July 4, 2013), 27:1–27:22. issn: 0730-

    2013

  23. [31]

    url: https://dl.acm.org/doi/ 10.1145/2487228.2487235

    doi: 10.1145/2487228.2487235. url: https://dl.acm.org/doi/ 10.1145/2487228.2487235

    doi

  24. [32]

    An Adaptive Cartesian Mesh Flow Solver Based on the Tree-data with Anisotropic Mesh Refinement

    Takanobu Ogawa. “An Adaptive Cartesian Mesh Flow Solver Based on the Tree-data with Anisotropic Mesh Refinement”. In: Computational Fluid Dynamics 2002. Ed. by Steve W. Armfield, Patrick Morgan, and Karke- nahalli Srinivas. Berlin, Heidelberg: Springer, 2003, pp. 453–458. isbn...

    2002 doi

  25. [33]

    freifrauvonbleifrei/DyAda

    Theresa Pollinger. freifrauvonbleifrei/DyAda. version 0.0.12. 2026. url: https://github.com/freifrauvonbleifrei/DyAda

    2026

  26. [34]

    freifrauvonbleifrei/wavelets with omnitrees

    Theresa Pollinger. freifrauvonbleifrei/wavelets with omnitrees. Will be pub- lished alongside paper. 2026.url: https://github.com/freifrauvonbleifrei/ wavelets_with_omnitrees

    2026

  27. [35]

    The Beauty of Anisotropic Mesh Refinement: Omnitrees for Efficient Dyadic Discretizations

    Theresa Pollinger, Masado Ishii, and Jens Domke. The Beauty of Anisotropic Mesh Refinement: Omnitrees for Efficient Dyadic Discretizations. Aug. 8,

  28. [36]

    arXiv: 2508.06316 [cs]

    doi: 10.48550/arXiv.2508.06316. arXiv: 2508.06316 [cs] . url: http://arxiv.org/abs/2508.06316. Pre-published

    doi arXiv

  29. [37]

    A Stable and Mass-Conserving Sparse Grid Combination Tech- nique with Biorthogonal Hierarchical Basis Functions for Kinetic Sim- ulations

    Theresa Pollinger, Johannes Rentrop, Dirk Pfl¨ uger, and Katharina Kor- mann. “A Stable and Mass-Conserving Sparse Grid Combination Tech- nique with Biorthogonal Hierarchical Basis Functions for Kinetic Sim- ulations”. In: Journal of Computational Physics 491 (Oct. 15, 2023), ...

    2023

  30. [38]

    Hierarchical Data Structures and Algorithms for Computer Graphics. I. Fundamentals

    Hanan Samet and Robert E. Webber. “Hierarchical Data Structures and Algorithms for Computer Graphics. I. Fundamentals”. In:IEEE Computer Graphics and Applications 8.3 (May 1988), pp. 48–68. issn: 1558-1756. doi: 10.1109/38.513. url: https://ieeexplore.ieee.org/abstract/ document/513

    1988 doi

  31. [39]

    Hyperbolic Wavelet Analysis of Classical Isotropic and Anisotropic Besov–Sobolev Spaces

    Martin Sch¨ afer, Tino Ullrich, and B´ eatrice Vedel. “Hyperbolic Wavelet Analysis of Classical Isotropic and Anisotropic Besov–Sobolev Spaces”. In: Journal of Fourier Analysis and Applications 27.3 (May 18, 2021), p. 51. issn: 1531-5851. doi: 10.1007/s00041-021-09844-z . url:...

    2021 doi

  32. [40]

    A Fast Algorithm for Adapted Time–Frequency Tilings

    Christoph M. Thiele and Lars F. Villemoes. “A Fast Algorithm for Adapted Time–Frequency Tilings”. In:Applied and Computational Harmonic Anal- ysis 3.2 (Apr. 1, 1996), pp. 91–99. issn: 1063-5203. doi: 10.1006/acha. 1996.0009. url: https://www.sciencedirect.com/science/article/ ...

    1996 doi

  33. [41]

    NanoMap: A GPU-Accelerated OpenVDB-Based Mapping and Simulation Package for Robotic Agents

    Violet Walker, Fernando Vanegas, and Felipe Gonzalez. “NanoMap: A GPU-Accelerated OpenVDB-Based Mapping and Simulation Package for Robotic Agents”. In: Remote Sensing 14.21 (Jan. 2022), p. 5463. issn: 2072-4292. doi: 10.3390/rs14215463 . url: https://www.mdpi.com/ 2072-4292/14/21/5463

    2022 doi

  34. [42]

    Sparse Grid Dis- continuous Galerkin Methods for High-Dimensional Elliptic Equations

    Zixuan Wang, Qi Tang, Wei Guo, and Yingda Cheng. “Sparse Grid Dis- continuous Galerkin Methods for High-Dimensional Elliptic Equations”. In: Journal of Computational Physics 314 (June 1, 2016), pp. 244–263. issn: 0021-9991. doi: 10.1016/j.jcp.2016.03.005 . url: http://www. sci...

    2016 doi

  35. [43]

    Peano—A Traversal and Storage Scheme for Octree-Like Adaptive Cartesian Multiscale Grids

    Tobias Weinzierl and Miriam Mehl. “Peano—A Traversal and Storage Scheme for Octree-Like Adaptive Cartesian Multiscale Grids”. In: SIAM Journal on Scientific Computing (Oct. 27, 2011). doi: 10.1137/100799071. url: https://epubs.siam.org/doi/10.1137/100799071

    2011 doi

  36. [44]

    fVDB: A Deep-Learning Framework for Sparse, Large-Scale, and High-Performance Spatial Intelligence

    Francis Williams et al. “fVDB: A Deep-Learning Framework for Sparse, Large-Scale, and High-Performance Spatial Intelligence”. In: ACM Trans- actions on Graphics 43.4 (July 19, 2024). doi: 10.1145/3658226. arXiv: 2407.01781 [cs.CV]. url: http://arxiv.org/abs/2407.01781

    2024 doi

  37. [45]

    GPU Volume Rendering with Hierarchical Compression Using VDB

    Stefan Zellmann, Milan Jaros, Jefferson Amstutz, and Ingo Wald. GPU Volume Rendering with Hierarchical Compression Using VDB . Version 2. Apr. 10, 2025. doi: 10.48550/arXiv.2504.04564 . arXiv: 2504.04564 [cs]. url: http://arxiv.org/abs/2504.04564. Pre-published

    2025 doi arXiv

  38. [46]

    Thingi10K: A Dataset of 10,000 3D- Printing Models

    Qingnan Zhou and Alec Jacobson. Thingi10K: A Dataset of 10,000 3D- Printing Models. July 2, 2016. doi: 10.48550/arXiv.1605.04797. arXiv: 1605.04797 [cs]. Pre-published. 29

    2016 doi arXiv

Pith tools

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