Pith. sign in

REVIEW 4 major objections 4 minor 36 references

Adaptive Voxelization for Transform coding of 3D Gaussian splatting data

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

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

Pith's one-line read Adaptive voxelization cuts 3D Gaussian bitrates by 13%.

desk verdict Genuinely new adaptive voxelization for 3DGS compression, but the evidence is thin and the thresholds are undisclosed; worth a serious referee. read the letter →

arxiv 2506.00271 v1 pith:ODMHU62O submitted 2025-05-30 eess.IV

Chenjunjie Wang , Shashank N. Sridhara , Eduardo Pavez , Antonio Ortega , Cheng Chang This is my paper
classification eess.IV
keywords 3DGaussianSplattingAdaptiveVoxelizationPointCloudCompressionGPCCTransformCoding3DGSRate-DistortionFine-tuning
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 proposes a hybrid compression framework for 3D Gaussian Splatting (3DGS) that combines adaptive voxelization, lightweight attribute fine-tuning, and the GPCC point-cloud transform codec. The goal is to compress a pre-trained 3DGS model at multiple bitrates in one encoding pass, something post-training point-cloud codecs cannot do efficiently because uniform voxelization wastes bits on dense low-volume Gaussian regions and displaces large-volume Gaussians. The adaptive voxelization keeps high resolution where large-volume Gaussians live and lowers resolution in dense regions, cutting the number of Gaussians and the bitrate. A fine-tuning initialization that averages SH coefficients and opacities within each voxel and takes the largest Gaussian's covariance recovers rendering quality in about 8K iterations instead of 30K. If the claimed savings hold, 3DGS streaming can use a single encoded model to serve many rate points.

What carries the argument

The load-bearing mechanism is the set of adaptive voxelization criteria, which replace the expensive iterative 2-Wasserstein barycenter computation (costly to solve for Σ∗) with three cheap local tests: Gaussian volume, per-voxel Gaussian count τ1, and distance to voxel center τ2, with octree depths ranging from Jlow to Jhigh. These tests decide where the octree splits so that high-volume Gaussians keep fine position resolution and dense clusters of small Gaussians are merged. The second mechanism is the attribute initialization: after voxelization, SH/opacity are initialized by voxel-wise averaging and covariance by the largest Gaussian in the voxel, which lets constrained fine-tuning with fixed means recover ground-truth PSNR in about 8K iterations, compared with 30K without initialization.

What would settle it

Run the codec's adaptive voxelization with the same fixed thresholds on a diverse set of at least ten 3DGS scenes (bounded and unbounded) and measure bitrate savings against uniform voxelization; if the savings fall near zero or reverse on a substantial fraction of scenes, the claimed general advantage is not real. A simpler check is to inspect the released code and see whether the thresholds are constants or need manual adjustment per input.

Watch

Extended reading notes

Core claim

The paper's central claim is that 3DGS compression can be made both multi-rate and close to generative-codec quality by treating Gaussian means as a point cloud but voxelizing it adaptively rather than uniformly. The adaptive criterion is a proxy for minimizing the 2-Wasserstein distance between the original Gaussians and their merged representatives: the largest v% of Gaussians by volume go to the finest octree depth Jhigh, while other voxels split only if they contain more than τ1 Gaussians or if points sit farther than half a voxel from the center (τ2). This reduces the number of Gaussians and the position bitrate, and the paper reports 13.42% and 12.64% bitrate savings over uniform voxelization on the Playroom and Truck scenes, respectively, with small average PSNR gains. Compared with post-training codecs GGSC and SPZ, the proposed method wins by large margins and is competitive with the hybrid MesonGS without requiring decoder-side fine-tuning.

Load-bearing premise

The whole scheme depends on the undisclosed threshold values (v, τ1, τ2, Jlow, Jhigh) acting as a universal proxy for the optimal Wasserstein-barycenter grouping across scenes, without per-scene tuning.

Editorial extensions

If this is right

  • A pre-trained 3DGS model can be encoded once and decoded at several bitrates by adjusting GPCC quantization parameters, enabling rate-adaptive streaming.
  • Adaptive voxelization reduces the number of Gaussians enough that GPCC octree coding of positions becomes cheaper than uniform-voxel post-training codecs at equal rendering quality.
  • The 8K-iteration fine-tuning with proposed initialization makes post-training retraining practical on standard hardware, avoiding per-rate model training.
  • The hybrid pipeline closes part of the quality gap to generative 3DGS compression while preserving the encoder-side rate control that generative methods lack.

Reading between the lines

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

  • The same adaptive-voxelization criteria could be applied to other primitive-based scene representations (e.g., 3D point clouds with per-point size attributes), where the volume term of the Wasserstein distance naturally reweights positional distortion.
  • The undisclosed thresholds (v, τ1, τ2, Jlow, Jhigh) are likely scene-dependent; a sensitivity study over more scenes would reveal whether the 13% savings generalize or require per-scene tuning.
  • Because the paper measures gains on only two scenes, a natural test is to run the identical pipeline on a broader benchmark (e.g., Mip-NeRF 360 scenes) and report the distribution of bitrate savings.
  • Combining adaptive voxelization with the generative compression methods' entropy models could yield a single codec that is both multi-rate and closer to generative rate-distortion performance.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes a hybrid compression framework for pre-trained 3D Gaussian splatting models. It first adaptively voxelizes Gaussian means, giving higher resolution to large-volume Gaussians and to voxels with many or widely spread Gaussians, and then fine-tunes the remaining attributes with an initialization based on intra-voxel averaging and largest-volume Gaussian covariance. The voxelized positions are losslessly compressed with GPCC octree coding, while SH coefficients, opacity, and optionally covariance are compressed with transform coding or VQ. Experiments on Playroom and Truck report 13.42% and 12.64% BD-rate savings over uniform voxelization, and the method is compared against GGSC, SPZ, MesonGS, and several generative codecs. The paper claims that this enables multi-rate, computationally efficient compression of pre-trained 3DGS models.

Significance. If the reported results hold, the paper makes a useful contribution by combining a 3DGS-specific adaptive voxelization with transform coding from point cloud compression, and by showing that a short fine-tuning stage with a sensible initialization can restore fidelity after voxelization. The framework is genuinely hybrid: it produces multiple rate points per encoding, unlike typical generative codecs, and it avoids decoder-side fine-tuning, unlike MesonGS. The source code is released, which strengthens reproducibility. The main limitation is that the evidence base is narrow: two scenes, no error bars, no sensitivity analysis, and several key hyperparameters are undisclosed. The adaptive-voxelization component, which is the paper's central novelty, is therefore not yet convincingly established as a general method rather than a per-scene tuned procedure.

major comments (4)
  1. [Section 3.2.2] The adaptive voxelization algorithm depends on four parameters that are never disclosed: the largest-volume Gaussian percentage v, the Gaussian count threshold tau1, the distance threshold tau2, and the octree depth range Jlow to Jhigh. Only Jlow and Jhigh appear in the figure captions (e.g., Fig. 8: Jlow=8, Jhigh=16/18), while v, tau1, and tau2 are absent from the text entirely. Since Table 1 attributes 13.42% and 12.64% bitrate savings to this algorithm, the lack of these values prevents reproduction and makes it impossible to distinguish a robust codec from one that has been tuned to these two scenes. Please disclose all parameter values and provide a sensitivity analysis, or a principled rule for setting them that does not require per-scene optimization.
  2. [Section 4.1.3, Table 1] The central quantitative claim is based on only two scenes (Playroom and Truck), with no repeated runs and no error bars. Fine-tuning and transform coding involve stochastic and rate-dependent choices, so single-run BD-rate numbers should be treated cautiously. Please either evaluate on additional scenes (e.g., more Deep Blending, Mip-NeRF 360, or Tanks and Temples scenes) or provide multiple trials and error bars, and state whether the reported savings are typical over the chosen operating points.
  3. [Abstract and Section 4.2] The abstract states that the proposed framework 'outperforms existing methods', but Section 4.2 explicitly says that generative compression methods such as CompGS, Compact 3D, RDO-GS, and Compressed 3DGS achieve superior compression performance. The results in Fig. 9 support only the narrower claim that the method outperforms the post-training baselines GGSC and SPZ and is comparable to MesonGS. Please revise the abstract and any other claims to match the evidence.
  4. [Section 3.2.1 and Section 3.2.2] The adaptive voxelization criteria are motivated by the claim that they approximately minimize the 2-Wasserstein barycenter objective in Eq. (4), but no experiment validates this connection. The paper does not compare the proposed heuristic partitions against the exact or iteratively computed barycenter, nor against a baseline that computes d in Eq. (4) for uniform versus adaptive groupings. Since this approximation is the stated rationale for the new voxelization method, please include such a validation, even a small-scale one, or weaken the claim to an empirical observation.
minor comments (4)
  1. [Fig. 5 caption] The caption defines tau2 = W/2^{J+1}, but this quantity is not defined in the main text and the notation J is ambiguous; please define it explicitly in Section 3.2.2.
  2. [Fig. 1 and Conclusion] There are typographical inconsistencies: 'V oxelization' appears in the Fig. 1 caption and 'V oxelized' appears in the Conclusion. Please fix these spacing errors.
  3. [Table 2] The table uses inconsistent abbreviations (CGS vs CompGS, C3D vs Compact 3D, C-3DGS vs Compressed 3DGS) and the column 'r-d pts/retrain' is not explained in the text or caption; please unify the method names and describe the column.
  4. [Section 2.4] The method for generating multiple R-D points is described only through the quantization parameters Qdc, Qac, and Qop. It would be clearer to state explicitly whether one encoding pass suffices to produce all points or whether the quantization parameters are applied in separate passes.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: central claims are benchmarked externally and the adaptive-voxelization criteria are heuristics, not re-statements of the outcome.

full rationale

The paper's main contribution—adaptive voxelization plus fine-tuning plus GPCC/RAHT transform coding—is evaluated against external post-training codecs (GGSC, SPZ, MesonGS) and generative codecs, so the central comparison is not self-referential. The adaptive-voxelization rules (volume percentile, tau1, tau2, Jlow-Jhigh) are introduced in Section 3.2.2 as heuristics chosen to approximate the W2-barycenter objective of Eq. (4); they are not derived from the reported BD-rate savings, and Table 1 compares adaptive vs uniform voxelization within the same pipeline, which is an internal ablation rather than a fitted prediction. The fine-tuning initialization averages attributes and uses the largest-volume Gaussian for covariance; this is an input to the pipeline, not a quantity that the paper then claims to predict. Two self-citations exist: [20] for earlier adaptive voxelization in LiDAR point clouds and [22] for YUV decorrelation of SH coefficients. Neither is load-bearing: [20] is context for the general idea, and [22] supports a standard linear color-channel decorrelation used inside the codec; the paper's central claim does not reduce to either. Missing hyperparameters (v, tau1, tau2) and the two-scene evaluation are reproducibility and generalization concerns, not circularity.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central method rests on three assumptions: that fixed-position fine-tuning recovers quality, that the heuristic voxelization criteria approach the Wasserstein-optimal grouping, and that GPCC is an appropriate and efficient codec for the resulting voxelized data. The thresholds that control the adaptive voxelization are free parameters that are not disclosed, and the method's reported gains may depend on tuning them to the test scenes.

free parameters (4)
  • v_percent_largest_gaussians
    Fraction of largest-volume Gaussians assigned to the highest octree depth J_high; value is not disclosed in the paper.
  • tau1_gaussian_count_threshold
    Number of Gaussians in a voxel that triggers further partitioning; not reported in the paper.
  • tau2_distance_threshold = W/2^(J+1)
    Distance from Gaussian means to voxel center must be below half the voxel size to avoid partitioning; defined in the Fig. 5 caption but no numeric value is given.
  • Jlow_Jhigh_octree_depths = e.g., Jlow=8, Jhigh=16 or 18
    Minimum and maximum octree depths for adaptive voxelization; values vary per scene and are experimental choices.
assumptions (3)
  • domain assumption Fine-tuning 3DGS attributes with fixed Gaussian means recovers rendering quality close to the original model.
    Section 3.3 and Section 4.1.2 report that 8K iterations of constrained fine-tuning approach the ground-truth PSNR, relying on this assumption.
  • ad hoc to paper The heuristic criteria (volume, count, distance) approximate the minimizer of the 2-Wasserstein barycenter distance without computing it.
    Section 3.2.2 claims these criteria are selected to minimize equation (4), but no experiment validates the approximation.
  • domain assumption MPEG GPCC provides correct and efficient octree-based lossless position coding and RAHT/RLGR transform coding of the voxelized attributes.
    Sections 2.3 and 2.4 adopt GPCC as the coding engine based on the point cloud compression literature.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adaptive Voxelization for Transform coding of 3D Gaussian splatting data." pith.science (2026). https://pith.science/paper/ODMHU62O

@misc{pith2026250600271,
  author       = {Pith},
  title        = {Pith review of: Adaptive Voxelization for Transform coding of 3D Gaussian splatting data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ODMHU62O}},
  note         = {Machine review of arXiv:2506.00271}
}
read the original abstract

We present a novel compression framework for 3D Gaussian splatting (3DGS) data that leverages transform coding tools originally developed for point clouds. Contrary to existing 3DGS compression methods, our approach can produce compressed 3DGS models at multiple bitrates in a computationally efficient way. Point cloud voxelization is a discretization technique that point cloud codecs use to improve coding efficiency while enabling the use of fast transform coding algorithms. We propose an adaptive voxelization algorithm tailored to 3DGS data, to avoid the inefficiencies introduced by uniform voxelization used in point cloud codecs. We ensure the positions of larger volume Gaussians are represented at high resolution, as these significantly impact rendering quality. Meanwhile, a low-resolution representation is used for dense regions with smaller Gaussians, which have a relatively lower impact on rendering quality. This adaptive voxelization approach significantly reduces the number of Gaussians and the bitrate required to encode the 3DGS data. After voxelization, many Gaussians are moved or eliminated. Thus, we propose to fine-tune/recolor the remaining 3DGS attributes with an initialization that can reduce the amount of retraining required. Experimental results on pre-trained datasets show that our proposed compression framework outperforms existing methods.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

36 extracted references · 33 canonical work pages

  1. [1]

    3DGS can produce high-quality models with a fraction of the training time while allowing faster ren- dering than its precursors—implicit radiance fields, such as NeRF’s

    INTRODUCTION 3D Gaussian Splatting (3DGS) is a state-of-the-art image-based 3D scene representation method [1]. 3DGS can produce high-quality models with a fraction of the training time while allowing faster ren- dering than its precursors—implicit radiance fields, such as NeRF’s

    work page

  2. [2]

    Adaptive Voxelization for Transform coding of 3D Gaussian splatting data

    and Instant-NGP [3], and Plenoxels [4]. 3DGS provides an ex- plicit representation of 3D scenes [1], consisting of a list of Gaus- sians to describe the scene geometry, each represented by (i) a mean vector (position), (ii) a covariance matrix, and (iii) appearance pa- rameters to represent view-dependent colors, i.e., opacity and spher- ical harmonic (SH...

    work page Pith review arXiv 2025

  3. [3]

    3D Gaussian Splatting data 3DGS are an explicit point-based 3D scene representation [1]

    COMPRESSION PIPELINE 2.1. 3D Gaussian Splatting data 3DGS are an explicit point-based 3D scene representation [1]. The ith Gaussian has a mean or positionµ i ∈R 3, a covariance ma- trixΣ i ∈R 3×3,Dspherical harmonic coefficientsC i ∈R D×3 to represent view-dependent color and opacitiesα i ∈[0,1]. Thus, a 3DGS model withNGaussians can be described as a lis...

    work page

  4. [4]

    geodesic distance

    VOXELIZATION 3.1. Uniform voxelization In uniform voxelization, the 3D space is partitioned into volume el- ements calledvoxels(cubes). The positions of all points assigned to a voxel are quantized to the voxel center, and their corresponding attributes are averaged [23]. If the bounding box of the 3D points has a volume ofW×W×W, the first level of the oc...

    work page

  5. [5]

    6: Rate-distortion plots of proposed codec with uniform vox- elization and fine-tuning without initialization on bounded (Fig

    EXPERIMENTS We evaluate the proposed compression pipeline on two scenes: a bounded indoor scene,Playroom, from the Deep blending dataset (a) Playroom (Deep blending [26]) (b) Truck (Tanks and temples [27]) Fig. 6: Rate-distortion plots of proposed codec with uniform vox- elization and fine-tuning without initialization on bounded (Fig. 6a) and unbounded (...

    work page

  6. [6]

    This framework integrates adaptive vox- elization and attribute refinement to improve compression efficiency while preserving rendering quality

    CONCLUSION We introduced a hybrid compression framework for 3DGS data based on the GPCC codec. This framework integrates adaptive vox- elization and attribute refinement to improve compression efficiency while preserving rendering quality. Adaptive voxelization dynam- ically adjusts resolution based on Gaussian volume and spatial redundancy, reducing the ...

    work page

  7. [7]

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

    B. Kerbl, G. Kopanas, T. Leimk ¨uhler, and G. Drettakis, “3d gaussian splatting for real-time radiance field rendering.,”ACM Trans. Graph., vol. 42, no. 4, pp. 139–1, 2023

    2023

  8. [8]

    Nerf: Representing scenes as neu- ral radiance fields for view synthesis,

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ra- mamoorthi, and R. Ng, “Nerf: Representing scenes as neu- ral radiance fields for view synthesis,”Communications of the ACM, vol. 65, no. 1, pp. 99–106, 2021

    work page 2021

Show all 36 references
  1. [9]

    Instant neu- ral graphics primitives with a multiresolution hash encoding,

    T. M ¨uller, A. Evans, C. Schied, and A. Keller, “Instant neu- ral graphics primitives with a multiresolution hash encoding,” ACM Transactions on Graphics, vol. 41, p. 1–15, July 2022

    2022

  2. [10]

    Plenoxels: Radiance fields without neural net- works,

    S. Fridovich-Keil, A. Yu, M. Tancik, Q. Chen, B. Recht, and A. Kanazawa, “Plenoxels: Radiance fields without neural net- works,” inProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pp. 5501–5510, June 2022

    2022

  3. [11]

    3d gaussian splatting as new era: A survey,

    B. Fei, J. Xu, R. Zhang, Q. Zhou, W. Yang, and Y . He, “3d gaussian splatting as new era: A survey,”IEEE Transactions on Visualization and Computer Graphics, pp. 1–20, 2024

    2024

  4. [12]

    3dgs. zip: A survey on 3d gaussian splatting compression methods,

    M. T. Bagdasarian, P. Knoll, Y .-H. Li, F. Barthel, A. Hils- mann, P. Eisert, and W. Morgenstern, “3dgs. zip: A survey on 3d gaussian splatting compression methods,”arXiv preprint arXiv:2407.09510, 2024

    2024 arXiv

  5. [13]

    A hierarchical compression technique for 3d gaussian splatting compression,

    H. Huang, W. Huang, Q. Yang, Y . Xu,et al., “A hierarchical compression technique for 3d gaussian splatting compression,” arXiv preprint arXiv:2411.06976, 2024

    2024 arXiv

  6. [14]

    Light- gaussian: Unbounded 3d gaussian compression with 15x re- duction and 200+ FPS,

    Z. Fan, K. Wang, K. Wen, Z. Zhu, D. Xu, and Z. Wang, “Light- gaussian: Unbounded 3d gaussian compression with 15x re- duction and 200+ FPS,” inThe Thirty-eighth Annual Confer- ence on Neural Information Processing Systems, 2024

    2024

  7. [15]

    Com- pressed 3d gaussian splatting for accelerated novel view syn- thesis,

    S. Niedermayr, J. Stumpfegger, and R. Westermann, “Com- pressed 3d gaussian splatting for accelerated novel view syn- thesis,” inProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pp. 10349–10358, 2024

    2024

  8. [16]

    Hac: Hash- grid assisted context for 3d gaussian splatting compression,

    Y . Chen, Q. Wu, W. Lin, M. Harandi, and J. Cai, “Hac: Hash- grid assisted context for 3d gaussian splatting compression,” inEuropean Conference on Computer Vision, pp. 422–438, Springer, 2024

    2024

  9. [17]

    Compact3d: Compressing gaussian splat ra- diance field models with vector quantization,

    K. Navaneet, K. P. Meibodi, S. A. Koohpayegani, and H. Pirsiavash, “Compact3d: Compressing gaussian splat ra- diance field models with vector quantization,”arXiv preprint arXiv:2311.18159, 2023

    2023 arXiv

  10. [18]

    A benchmark for gaussian splatting compression and quality assessment study,

    Q. Yang, K. Yang, Y . Xing, Y . Xu, and Z. Li, “A benchmark for gaussian splatting compression and quality assessment study,” (New York, NY , USA), Association for Computing Machinery, 2024

    2024

  11. [19]

    Open-sourcing .spz: it’s .jpg for 3d gaussian splats

    “Open-sourcing .spz: it’s .jpg for 3d gaussian splats.” https://scaniverse.com/news/spz-gaussian-splat-open-source- file-format. Accessed: 10-24-2024

    2024

  12. [20]

    An overview of ongoing point cloud com- pression standardization activities: video-based (v-pcc) and geometry-based (g-pcc),

    D. Graziosi, O. Nakagami, S. Kuma, A. Zaghetto, T. Suzuki, and A. Tabatabai, “An overview of ongoing point cloud com- pression standardization activities: video-based (v-pcc) and geometry-based (g-pcc),”APSIPA Transactions on Signal and Information Processing, vol. 9, p. e13, 2020

    2020

  13. [21]

    A comprehen- sive study and comparison of core technologies for mpeg 3-d point cloud compression,

    H. Liu, H. Yuan, Q. Liu, J. Hou, and J. Liu, “A comprehen- sive study and comparison of core technologies for mpeg 3-d point cloud compression,”IEEE Transactions on Broadcast- ing, vol. 66, no. 3, pp. 701–717, 2020

    2020

  14. [22]

    Geometric modeling using octree encoding,

    D. Meagher, “Geometric modeling using octree encoding,” Computer Graphics and Image Processing, vol. 19, no. 2, pp. 129 – 147, 1982

    1982

  15. [23]

    Compression of 3d point clouds using a region-adaptive hierarchical transform,

    R. L. De Queiroz and P. A. Chou, “Compression of 3d point clouds using a region-adaptive hierarchical transform,”IEEE Transactions on Image Processing, vol. 25, no. 8, pp. 3947– 3956, 2016

    2016

  16. [24]

    Mesongs: Post-training compression of 3d gaussians via ef- ficient attribute transformation,

    S. Xie, W. Zhang, C. Tang, Y . Bai, R. Lu, S. Ge, and Z. Wang, “Mesongs: Post-training compression of 3d gaussians via ef- ficient attribute transformation,” inEuropean Conference on Computer Vision, pp. 434–452, Springer, 2024

    2024

  17. [25]

    Polarnet: An improved grid representation for on- line lidar point clouds semantic segmentation,

    Y . Zhang, Z. Zhou, P. David, X. Yue, Z. Xi, B. Gong, and H. Foroosh, “Polarnet: An improved grid representation for on- line lidar point clouds semantic segmentation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 9601–9610, 2020

    2020

  18. [26]

    Cylindrical coordi- nates for lidar point cloud compression,

    S. N. Sridhara, E. Pavez, and A. Ortega, “Cylindrical coordi- nates for lidar point cloud compression,” in2021 IEEE Inter- national Conference on Image Processing (ICIP), pp. 3083– 3087, IEEE, 2021

    2021

  19. [27]

    Emerg- ing mpeg standards for point cloud compression,

    S. Schwarz, M. Preda, V . Baroncini, M. Budagavi, P. Cesar, P. A. Chou, R. A. Cohen, M. Krivoku ´ca, S. Lasserre, Z. Li, J. Llach, K. Mammou, R. Mekuria, O. Nakagami, E. Siahaan, A. Tabatabai, A. M. Tourapis, and V . Zakharchenko, “Emerg- ing mpeg standards for point cloud com...

    2019

  20. [28]

    Decorrelation of spherical harmonic coefficients towards ef- ficient compression of 3dgs

    S. N. Sridhara, A. Zaghetto, D. Graziosi, and A. Tabatabai, “Decorrelation of spherical harmonic coefficients towards ef- ficient compression of 3dgs.” ISO/IEC JTC 1/SC 29/WG 7 MPEG coding for 3D graphics and haptics, Doc m69039, Sap- poro, Japan, July, 2024

    2024

  21. [29]

    Dy- namic polygon clouds: representation and compression for vr/ar,

    E. Pavez, P. A. Chou, R. L. de Queiroz, and A. Ortega, “Dy- namic polygon clouds: representation and compression for vr/ar,”APSIPA Transactions on Signal and Information Pro- cessing, vol. 7, p. e15, 2018

    2018

  22. [30]

    A wasserstein-type distance in the space of gaussian mixture models,

    J. Delon and A. Desolneux, “A wasserstein-type distance in the space of gaussian mixture models,”SIAM Journal on Imaging Sciences, vol. 13, no. 2, pp. 936–970, 2020

    2020

  23. [31]

    A fixed-point approach to barycenters in wasser- stein space,

    P. C. ´Alvarez-Esteban, E. Del Barrio, J. Cuesta-Albertos, and C. Matr ´an, “A fixed-point approach to barycenters in wasser- stein space,”Journal of Mathematical Analysis and Applica- tions, vol. 441, no. 2, pp. 744–762, 2016

    2016

  24. [32]

    Deep blending for free-viewpoint image-based rendering,

    P. Hedman, J. Philip, T. Price, J.-M. Frahm, G. Drettakis, and G. Brostow, “Deep blending for free-viewpoint image-based rendering,” vol. 37, no. 6, pp. 257:1–257:15, 2018

    2018

  25. [33]

    Tanks and temples: Benchmarking large-scale scene reconstruction,

    A. Knapitsch, J. Park, Q.-Y . Zhou, and V . Koltun, “Tanks and temples: Benchmarking large-scale scene reconstruction,” ACM Transactions on Graphics, vol. 36, no. 4, 2017

    2017

  26. [34]

    Calculation of average psnr differences be- tween RD-curves,

    G. Bjontegaard, “Calculation of average psnr differences be- tween RD-curves,”VCEG-M33, 2001

    2001

  27. [35]

    Compact 3d gaussian representation for radiance field,

    J. C. Lee, D. Rho, X. Sun, J. H. Ko, and E. Park, “Compact 3d gaussian representation for radiance field,” 2024

    2024

  28. [36]

    End-to-end rate-distortion optimized 3d gaussian representa- tion,

    H. Wang, H. Zhu, T. He, R. Feng, J. Deng, J. Bian, and Z. Chen, “End-to-end rate-distortion optimized 3d gaussian representa- tion,” inEuropean Conference on Computer Vision, pp. 76–92, Springer, 2024

    2024

Pith tools

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