Pith. sign in

REVIEW 4 major objections 5 minor 51 references

Steepest Descent Density Control for Compact 3D Gaussian Splatting

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read The paper proves that a Gaussian should split only when its splitting matrix has a negative eigenvalue, and that the optimal split is two offspring with half opacity each, displaced in opposite directions.

desk verdict Useful theory and plausible 50% point reduction, but the implemented position-only splitting matrix may not actually instantiate the theorem, and key algorithm details are missing. read the letter →

arxiv 2505.05587 v1 pith:SDFGQTMN submitted 2025-05-08 cs.CV

Peihao Wang , Yuehao Wang , Dilin Wang , Sreyas Mohan , Zhiwen Fan , Lemeng Wu , Ruisi Cai , Yu-Ying Yeh
show 3 more authors
Zhangyang Wang Qiang Liu Rakesh Ranjan
This is my paper · ORCID
classification cs.CV
keywords 3DGaussianSplattingdensitycontrolsplittingmatrixsteepestdescentsaddlepointescapecompactscenerepresentationnovelviewsynthesisoptimizationtheory
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to replace the heuristic densification step in 3D Gaussian Splatting with a rule that comes out of the loss landscape rather than from ad hoc thresholds. It defines a per-Gaussian splitting matrix and shows that splitting a primitive lowers the rendering error only when that matrix has a negative smallest eigenvalue; in that case the optimal move is to create two offspring, each with half the parent's opacity, displaced symmetrically along the eigenvector of the smallest eigenvalue. The resulting method, SteepGS, reports about a 50% reduction in the number of Gaussian points on standard real-world scenes while keeping rendering quality close to the original 3DGS, with lower training time and memory use. If the claim is right, density control is no longer an artisanal collection of heuristics but a solved optimization step with an analytic answer.

What carries the argument

The load-bearing object is the splitting matrix $S^{(i)}(\theta) = \mathbb{E}_{(\Pi,x)\sim D}\left[\frac{\partial \ell}{\partial \sigma_{\Pi}(x;\theta^{(i)})}\nabla^2_{\theta^{(i)}}\sigma_{\Pi}(x;\theta^{(i)})\right]$, the expectation over cameras and pixels of the scalar loss-gradient with respect to one Gaussian's projected output times the Hessian of that Gaussian's projected kernel with respect to its own parameters. It is the part of the loss curvature that a split can actually change. Its smallest eigenvalue and corresponding eigenvector decide the split threshold, the number of offspring, their displacement directions, and the opacity normalization, all through a second-order Taylor expansion of the post-split loss.

What would settle it

Split the same trained Gaussian twice, once along $v_{\min}(S^{(i)}(\theta))$ and once along an arbitrary unit vector, both with half opacity and the same small step size; whenever $\lambda_{\min}(S^{(i)}(\theta))<0$, the eigenvector split must give the smaller photometric loss, otherwise Theorem 2 is wrong.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the effect of splitting any single Gaussian on the total photometric loss is governed entirely by one pointwise matrix, the splitting matrix $S^{(i)}(\theta)$. Theorem 1 expands the post-split loss into mean-shift terms, which ordinary gradient updates can already achieve, plus a quadratic form in $S^{(i)}(\theta)$; Theorem 2 then states that the optimal split uses exactly two offspring, each reweighted by $1/2$, displaced by $\pm v_{\min}(S^{(i)}(\theta))$, and that this split strictly decreases the loss if and only if $\lambda_{\min}(S^{(i)}(\theta))<0$. The paper reports that the resulting SteepGS algorithm roughly halves the Gaussian count on standard real-world novel-view benchmarks while keeping PSNR, SSIM, and LPIPS close to the original 3DGS, because it refuses splits that cannot reduce the loss.

Load-bearing premise

The empirical gains rest on the unverified assumption that computing the splitting matrix from position parameters alone, while ignoring how projection and color depend on position, preserves the steepest-descent behavior proven for the full parameter vector, and the algorithm does not specify the small displacement step size that the theorem's expansion requires.

Editorial extensions

If this is right

  • A Gaussian should be densified only when $\lambda_{\min}(S^{(i)}(\theta))<0$; splitting any other Gaussian cannot lower the rendering error.
  • Two offspring are sufficient for the steepest possible local descent; generating more than two adds points without further reducing the loss.
  • The two offspring should carry exactly half the parent's opacity each, which preserves the local density and replaces approximate opacity normalizations with an exact one.
  • The new positions are fixed by the splitting matrix: the offspring move by $\pm v_{\min}(S^{(i)}(\theta))$, rather than along the gradient or by sampling from the parent's covariance.
  • On the datasets tested, this rule yields roughly a 50% reduction in Gaussian point count at comparable rendering quality, with correspondingly lower memory use.

Reading between the lines

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

  • A natural testable extension is to compute the full-parameter splitting matrix, including covariance, opacity, and spherical-harmonic coefficients, on small scenes and compare its smallest eigenvalue and eigenvector with the position-only version used in SteepGS; frequent sign flips would reveal how much the cheap approximation costs.
  • The same saddle-point logic should transfer to any scene representation built from a sum of local basis kernels, such as other point-based or volumetric radiance fields, because the splitting matrix needs only a per-particle Hessian of the kernel and the gradient of the loss with respect to that kernel's output.
  • The 'compactest splitting' variant described in the supplementary material, which splits only when the gradient is small and the splitting matrix is indefinite, suggests a path to a fully automatic densification schedule with no remaining hyper-parameter thresholds; running that variant as the main algorithm would show whether the criterion alone is enough.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. This paper studies densification in 3D Gaussian Splatting from a second-order optimization perspective. The authors define a per-Gaussian splitting matrix S(i)(θ) = E[∂ℓ/∂σ ∇²θσ] and prove a Taylor expansion (Theorem 1) of the loss after replacing a Gaussian by a weighted mixture of offspring. From this expansion they derive a necessary condition for splitting to reduce loss (λmin(S(i)) < 0), prove that two offspring with half opacity displaced along ±vmin(S(i)) are optimal (Theorem 2), and implement this as SteepGS, a CUDA-based density control scheme that computes a position-only version of the splitting matrix. Experiments on Mip-NeRF 360, Tanks & Temples, and Deep Blending report roughly a 50% reduction in Gaussian count with modest PSNR changes but larger SSIM and LPIPS degradations on several scenes. The paper claims that SteepGS is a principled replacement for the heuristic Adaptive Density Control.

Significance. If the theoretical claim holds, replacing the heuristic ADC with a principled second-order splitting rule would be a meaningful contribution to compact 3D Gaussian splatting: it would justify both when to split and where to place the offspring, and the reported point reduction is practically valuable for memory and rendering speed. The paper's strengths include the self-contained Taylor-expansion derivations in Appendix C, the closed-form Hessian approximation in Appendix C.4, and the per-scene breakdown tables in the supplementary material. The contribution is not circular: the splitting rule is derived from the same loss being optimized, which is standard practice. However, the current manuscript does not fully establish that the implemented SteepGS is an instantiation of the proven SDC rule, because the implementation uses only a position block of the splitting matrix and the gating condition on G(i) is left undefined. The empirical quality claims are also stronger than the tables support. Overall the work is promising and the theory-implementation gap appears fixable, but the central claim as written is not yet fully supported.

major comments (4)
  1. [Sec. 4.3, Theorem 2] Theorem 2 is stated as the optimal solution to Eq. 7, but the proof only minimizes the splitting characteristic function Δ(i) subject to ||δ_j||≤1 and ignores the mean-shift terms in Theorem 1. Unless the gradient ∇θL(θ) vanishes, a configuration with a nonzero average displacement μ can reduce L by O(ε), which dominates the O(ε²) splitting term. The theorem therefore establishes the optimal split only at stationary points or under an additional small-gradient condition, but the theorem statement and the surrounding text do not state such a condition. This makes the theoretical claim stronger than what is proved.
  2. [Sec. 4.4 and App. C.4] The implementation computes the splitting matrix only for the position parameters, dropping the covariance and color blocks and explicitly breaking the dependency of the projection matrix and view-dependent colors on the mean (Appendix C.4). Theorem 2 concerns the full-parameter S(i). No argument, ablation, or experiment shows that the position-only 3×3 block preserves either the sign of λmin or the direction vmin of the full matrix. If the negative curvature of the full matrix lies in the covariance block, the implemented test would never split that Gaussian; conversely, the position-only direction need not be the steepest direction claimed by the theorem. This gap means that the empirical ≈50% point reduction in Table 1 has not been tied to the theoretical SDC mechanism.
  3. [Algorithm 1, Sec. 4.4, App. A.2] Algorithm 1 gates splitting on a 'condition on G(i)' that is never defined in the main text. If the intended condition is the original ADC large-gradient test, then splits occur at non-stationary points where the first-order mean-shift term in Theorem 1 is O(ε) and dominates the O(ε²) splitting term, so the saddle-point justification of Theorem 2 does not apply. If the intended condition is the small-gradient 'compactest' condition of Appendix A.2 (||G(i)||≤ε_grad), that fact is essential and must be stated together with the threshold used. The current ambiguity prevents the reader from verifying that the reported experiments actually exercise the theoretical split rule.
  4. [Table 1; Supp. Tables 3 and 4] The abstract and Sec. 5.2 claim 'no quality compromise', but the reported metrics show noticeable SSIM and LPIPS degradation on Mip-NeRF 360 (SSIM 0.872→0.857, LPIPS 0.183→0.211). The per-scene supplement shows a large SSIM drop on Stump (0.908→0.742) and Garden (0.867→0.851). Table 6 reports improved PSNR with more densification iterations but provides no SSIM or LPIPS for that setting. The quality claim should be qualified, or the extended-training protocol should be made the standard one with full metrics reported.
minor comments (5)
  1. [Sec. 4.2 and App. C.2] The definitions of μ(i) and δ(i) in Sec. 4.2 are actual parameter displacements, while the proof in Appendix C.2 defines them divided by ε; please reconcile these definitions so that Theorem 1, its proof, and Algorithm 1 use the same scaling.
  2. [Algorithm 1 and Sec. 5.1] The offspring displacement step size ε is an input to Algorithm 1 but is never reported in the experiments; only the eigenvalue threshold ε_split = -1e-6 is given. Please report ε and its sensitivity.
  3. [Sec. 5.1] The '3DGS + Thres.' baseline is mentioned in Sec. 5.1 but not described as a separate method in the caption or table header; please clarify what threshold is used and how the point count is matched.
  4. [Table 7] The FPS footnote states that measurements can be inconsistent and the values 'should be considered as a reference'; as presented, this column is not a reliable efficiency metric and should either be removed or supported by multiple runs with error bars.
  5. [Throughout] There are several typographical and formatting issues, including 'Gaussion' in Sec. 5.1 and the misplaced backtick in the bibliography entry for Bulò et al.; these should be cleaned up.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: SDC is derived from a self-contained Taylor expansion; the position-only implementation is an acknowledged approximation, not a reduction to inputs.

full rationale

The paper's core derivation is self-contained. Theorem 1 expands the photometric loss after splitting into mean-shift terms and a quadratic form governed by the splitting matrix S(i)(θ), which is defined directly from the loss and the Gaussian parameterization. Theorem 2 then solves the resulting constrained quadratic minimization via the Rayleigh–Ritz variational characterization; the split condition λmin(S(i))<0 and the displacement along ±vmin(S(i)) follow from the same expansion, not from an assumed conclusion. This is ordinary optimization-theoretic reasoning, not circularity. No parameter in the theory is fitted to the evaluation benchmarks: the splitting threshold is a fixed hyperparameter (−1e−6), and the reported point counts are produced by the algorithm rather than used to tune it. The cited S2D/Firefly works by the same authors are contextual and analogical; Theorem 2 is proved in the appendix rather than imported as an external uniqueness theorem, so no load-bearing self-citation occurs. The acknowledged simplification in Appendix C.4—computing the splitting matrix for position parameters only and breaking the dependency of the projection matrix and colors on the mean—is a genuine approximation gap between the theorem and the implementation, and it is a correctness risk rather than a circular reduction: the implemented criterion is still computed from the same loss via a closed-form Hessian, not from the outcome it is used to predict. The undefined 'condition on G(i)' in Algorithm 1 is an implementation ambiguity, not a circular step. Therefore no circularity is found.

Assumptions & free parameters 3 free parameters · 4 assumptions · 1 invented entities

The central derivation relies on a second-order Taylor expansion of the photometric loss with respect to the Gaussian parameters (bounded third derivatives), on a local opacity-conservation constraint Σw=1, and on the assumption that offspring remain within a small ε-neighborhood. The implementation adds a position-only approximation of the splitting matrix and a hand-set eigenvalue threshold, neither of which follows from the theory.

free parameters (3)
  • eigenvalue threshold ε_split = -1e-6
    Hand-set threshold on λmin(S(i)) below which a Gaussian is split; not derived from theory and no sensitivity analysis is reported (Sec. 4.4).
  • offspring displacement step size ε = unspecified
    Algorithm 1 displaces offspring by θ(i) ± ε vmin, but the value of ε used in experiments is not stated in the paper; it controls how far offspring move and therefore affects both quality and point count.
  • gradient condition on G(i) = unspecified
    Algorithm 1 splits a point only if 'condition on G(i)' holds, but the condition is never defined; this determines which points are eligible for splitting.
assumptions (4)
  • domain assumption The photometric loss L(θ) has bounded third-order derivatives with respect to the Gaussian parameters (Theorem 1 assumption).
    Invoked at the start of Sec. 4.2 and used in all Taylor expansions; the ℓ1+SSIM rendering loss is not proven to satisfy this uniformly.
  • ad hoc to paper Total opacity is conserved after splitting: Σ_j w_j = 1 for each Gaussian (Sec. 4.2).
    Introduced as a practical constraint; it shapes the optimality result and rules out other opacity normalizations that might perform differently in alpha-blended rendering.
  • domain assumption Offspring parameters remain within an ε-neighborhood of the parent and the second-order Taylor expansion is accurate on that scale.
    Theorem 1 assumes ∥ϑ_j - θ∥ ≤ ε and drops O(ε³) terms; the actual displacement in Algorithm 1 is chosen by an unspecified ε.
  • ad hoc to paper The projection matrix and view-dependent colors are treated as independent of the mean position when computing the Hessian (Appendix C.4).
    This simplification makes the Hessian computation tractable but means the implemented splitting matrix is not the full-parameter S(i) of the theorems.
invented entities (1)
  • splitting matrix S(i)(θ) independent evidence
    purpose: Encodes the local second-order effect of splitting a Gaussian into offspring; its smallest eigenvalue and eigenvector decide whether and where to split.
    It is a well-defined closed-form quantity computed from the loss gradient and the Gaussian Hessian; it is not a hidden variable, and its predictions are checked against rendering quality on standard benchmarks.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Steepest Descent Density Control for Compact 3D Gaussian Splatting." pith.science (2026). https://pith.science/paper/SDFGQTMN

@misc{pith2026250505587,
  author       = {Pith},
  title        = {Pith review of: Steepest Descent Density Control for Compact 3D Gaussian Splatting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SDFGQTMN}},
  note         = {Machine review of arXiv:2505.05587}
}
read the original abstract

3D Gaussian Splatting (3DGS) has emerged as a powerful technique for real-time, high-resolution novel view synthesis. By representing scenes as a mixture of Gaussian primitives, 3DGS leverages GPU rasterization pipelines for efficient rendering and reconstruction. To optimize scene coverage and capture fine details, 3DGS employs a densification algorithm to generate additional points. However, this process often leads to redundant point clouds, resulting in excessive memory usage, slower performance, and substantial storage demands - posing significant challenges for deployment on resource-constrained devices. To address this limitation, we propose a theoretical framework that demystifies and improves density control in 3DGS. Our analysis reveals that splitting is crucial for escaping saddle points. Through an optimization-theoretic approach, we establish the necessary conditions for densification, determine the minimal number of offspring Gaussians, identify the optimal parameter update direction, and provide an analytical solution for normalizing off-spring opacity. Building on these insights, we introduce SteepGS, incorporating steepest density control, a principled strategy that minimizes loss while maintaining a compact point cloud. SteepGS achieves a ~50% reduction in Gaussian points without compromising rendering quality, significantly enhancing both efficiency and scalability.

Figures

Figures reproduced from arXiv: 2505.05587 by the authors.

Figure 1
Figure 1. We theoretically investigate density control in 3DGS. As training via gradient descent progresses, many Gaussian primitives are [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Illustrative notation for the splitting process. The updates ϑ (i) j − θ (i) can be decomposed as first taking a mean￾field shift µ (i) and then applying individual updates δ (i) j . By this decomposition, Pw (i) j δ (i) j = 0. for shorthand. A densification algorithm determines values for {mi} n i=1, w, and ϑ at each step. After specifying values of w, these coefficients will be absorbed into the opacity values of … view at source ↗
Figure 3
Figure 3. Illustration of Steepest Density Control. SDC, as the optimal solution to Eq. 7, takes the steepest descent on the loss after splitting. Geometrically, it moves two off-spring Gaussians to opposite directions along the smallest eigenvector of the splitting matrix and shrinks the opacity of each Gaussian by 0.5. Eq. 4 is a highly non-convex objective, exhibiting numerous superfluous saddle points in its loss landscap… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Qualitative Results. We compare our SteepGS with other densification baselines. For each scene, the first row shows the rendered view, while the second row visualizes the error with respect to the ground truth. Key details are highlighted in the blue box. [2] including…
Figure 5
Figure 5. Figure 5: Visualization of splitting points. The rendered views are present on the left and the corresponding points to be split are visualized on the right. ter the entire training process, our method achieves these savings without the need for post-training steps. 5.3. Visuali…
Figure 6
Figure 6. Figure 6: Improved visual quality of our method after more [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: More visualizations of splitting points. We compare the number of points split by our proposed method and the original ADC. [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 27 canonical work pages

  1. [1]

    Neural point-based graphics

    Kara-Ali Aliev, Artem Sevastopolsky, Maria Kolos, Dmitry Ulyanov, and Victor Lempitsky. Neural point-based graphics. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXII 16, pages 696–712. Springer, 2020. 2

    work page 2020

  2. [2]

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

    Jonathan T Barron, Ben Mildenhall, Dor Verbin, Pratul P Srinivasan, and Peter Hedman. Mip-nerf 360: Unbounded anti-aliased neural radiance fields. In Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition, pages 5470–5479, 2022. 7

    work page 2022

  3. [3]

    Revising densification in gaussian splatting

    Samuel Rota Bul `o, Lorenzo Porzi, and Peter Kontschieder. Revising densification in gaussian splatting. arXiv preprint arXiv:2404.06109, 2024. 2, 3, 4, 6, 7, 8, 1

    arXiv 2024

  4. [4]

    Mvsnerf: Fast general- izable radiance field reconstruction from multi-view stereo

    Anpei Chen, Zexiang Xu, Fuqiang Zhao, Xiaoshuai Zhang, Fanbo Xiang, Jingyi Yu, and Hao Su. Mvsnerf: Fast general- izable radiance field reconstruction from multi-view stereo. In Proceedings of the IEEE/CVF international conference on computer vision, pages 14124–14133, 2021. 2

    2021

  5. [5]

    Tensorf: Tensorial radiance fields

    Anpei Chen, Zexiang Xu, Andreas Geiger, Jingyi Yu, and Hao Su. Tensorf: Tensorial radiance fields. In European conference on computer vision , pages 333–350. Springer,

  6. [6]

    bert2bert: Towards reusable pretrained language models

    Cheng Chen, Yichun Yin, Lifeng Shang, Xin Jiang, Yujia Qin, Fengyu Wang, Zhi Wang, Xiao Chen, Zhiyuan Liu, and Qun Liu. bert2bert: Towards reusable pretrained language models. arXiv preprint arXiv:2110.07143, 2021. 3

    arXiv 2021

  7. [7]

    Net2net: Accelerating learning via knowledge transfer

    Tianqi Chen, Ian Goodfellow, and Jonathon Shlens. Net2net: Accelerating learning via knowledge transfer. arXiv preprint arXiv:1511.05641, 2015. 3

    arXiv 2015

  8. [8]

    Gradmax: Growing neural networks using gradient information

    Utku Evci, Bart van Merrienboer, Thomas Unterthiner, Max Vladymyrov, and Fabian Pedregosa. Gradmax: Growing neural networks using gradient information. arXiv preprint arXiv:2201.05125, 2022. 3

    arXiv 2022

Show all 51 references
  1. [9]

    Lightgaussian: Unbounded 3d gaussian compression with 15x reduction and 200+ fps.arXiv preprint arXiv:2311.17245, 2023

    Zhiwen Fan, Kevin Wang, Kairun Wen, Zehao Zhu, Dejia Xu, and Zhangyang Wang. Lightgaussian: Unbounded 3d gaussian compression with 15x reduction and 200+ fps.arXiv preprint arXiv:2311.17245, 2023. 2, 3, 8

    2023 arXiv

  2. [10]

    Plenoxels: Radiance fields without neural networks

    Sara Fridovich-Keil, Alex Yu, Matthew Tancik, Qinhong Chen, Benjamin Recht, and Angjoo Kanazawa. Plenoxels: Radiance fields without neural networks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5501–5510, 2022. 2

    2022

  3. [11]

    Pup 3d-gs: Principled uncertainty pruning for 3d gaussian splatting

    Alex Hanson, Allen Tu, Vasu Singla, Mayuka Jayawardhana, Matthias Zwicker, and Tom Goldstein. Pup 3d-gs: Principled uncertainty pruning for 3d gaussian splatting. arXiv preprint arXiv:2406.10219, 2024. 2, 6

    2024

  4. [12]

    Deep blending for free-viewpoint image-based rendering

    Peter Hedman, Julien Philip, True Price, Jan-Michael Frahm, George Drettakis, and Gabriel Brostow. Deep blending for free-viewpoint image-based rendering. ACM Transactions on Graphics (ToG), 37(6):1–15, 2018. 7

    2018

  5. [13]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 42(4):139–1, 2023. 2, 3, 4, 6, 7, 8

    2023

  6. [14]

    3d gaussian splatting as markov chain monte carlo

    Shakiba Kheradmand, Daniel Rebain, Gopal Sharma, Wei- wei Sun, Jeff Tseng, Hossam Isack, Abhishek Kar, Andrea Tagliasacchi, and Kwang Moo Yi. 3d gaussian splatting as markov chain monte carlo. arXiv preprint arXiv:2404.09591,

    arXiv

  7. [15]

    Tanks and temples: Benchmarking large-scale scene reconstruction

    Arno Knapitsch, Jaesik Park, Qian-Yi Zhou, and Vladlen Koltun. Tanks and temples: Benchmarking large-scale scene reconstruction. ACM Transactions on Graphics (ToG), 36(4): 1–13, 2017. 7

    2017

  8. [16]

    Compact 3d gaussian representation for radiance field

    Joo Chan Lee, Daniel Rho, Xiangyu Sun, Jong Hwan Ko, and Eunbyung Park. Compact 3d gaussian representation for radiance field. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21719– 21728, 2024. 2, 3, 5

    2024

  9. [17]

    Soft raster- izer: A differentiable renderer for image-based 3d reasoning

    Shichen Liu, Tianye Li, Weikai Chen, and Hao Li. Soft raster- izer: A differentiable renderer for image-based 3d reasoning. In Proceedings of the IEEE/CVF international conference on computer vision, pages 7708–7717, 2019. 2

    2019

  10. [18]

    Taming 3dgs: High-quality radiance fields with limited resources

    Saswat Subhajyoti Mallick, Rahul Goel, Bernhard Kerbl, Francisco Vicente Carrasco, Markus Steinberger, and Fer- nando De La Torre. Taming 3dgs: High-quality radiance fields with limited resources. arXiv preprint arXiv:2406.15643 ,

    arXiv

  11. [19]

    Optical models for direct volume rendering

    Nelson Max. Optical models for direct volume rendering. IEEE Transactions on Visualization and Computer Graphics, 1(2):99–108, 1995. 3

    1995

  12. [20]

    Nerf: Representing scenes as neural radiance fields for view syn- thesis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view syn- thesis. Communications of the ACM, 65(1):99–106, 2021. 2, 3

    2021

  13. [21]

    Compact 3d scene representation via self- organizing gaussian grids

    Wieland Morgenstern, Florian Barthel, Anna Hilsmann, and Peter Eisert. Compact 3d scene representation via self- organizing gaussian grids. arXiv preprint arXiv:2312.13299,

    arXiv

  14. [22]

    Instant neural graphics primitives with a multires- olution hash encoding

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

    2022

  15. [23]

    Compact3d: Com- pressing gaussian splat radiance field models with vector quantization

    KL Navaneet, Kossar Pourahmadi Meibodi, Soroush Abbasi Koohpayegani, and Hamed Pirsiavash. Compact3d: Com- pressing gaussian splat radiance field models with vector quantization. arXiv preprint arXiv:2311.18159, 2023. 2

    2023 arXiv

  16. [24]

    Compressed 3d gaussian splatting for accelerated novel view synthesis

    Simon Niedermayr, Josef Stumpfegger, and R ¨udiger West- ermann. Compressed 3d gaussian splatting for accelerated novel view synthesis. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 10349–10358, 2024

    2024

  17. [25]

    Reducing the mem- ory footprint of 3d gaussian splatting

    Panagiotis Papantonakis, Georgios Kopanas, Bernhard Kerbl, Alexandre Lanvin, and George Drettakis. Reducing the mem- ory footprint of 3d gaussian splatting. Proceedings of the ACM on Computer Graphics and Interactive Techniques, 7 (1):1–17, 2024. 2

    2024

  18. [26]

    Structure- from-motion revisited

    Johannes L Schonberger and Jan-Michael Frahm. Structure- from-motion revisited. In Proceedings of the IEEE confer- ence on computer vision and pattern recognition, pages 4104– 4113, 2016. 3

    2016

  19. [27]

    Deep marching tetrahedra: a hybrid represen- tation for high-resolution 3d shape synthesis

    Tianchang Shen, Jun Gao, Kangxue Yin, Ming-Yu Liu, and Sanja Fidler. Deep marching tetrahedra: a hybrid represen- tation for high-resolution 3d shape synthesis. Advances in Neural Information Processing Systems, 34:6087–6101, 2021. 2

    2021

  20. [28]

    Light field networks: Neural scene representations with single-evaluation rendering

    Vincent Sitzmann, Semon Rezchikov, Bill Freeman, Josh Tenenbaum, and Fredo Durand. Light field networks: Neural scene representations with single-evaluation rendering. Ad- vances in Neural Information Processing Systems, 34:19313– 19325, 2021. 2

    2021

  21. [29]

    Eigenvalues of a symmetric 3 × 3 matrix

    Oliver K Smith. Eigenvalues of a symmetric 3 × 3 matrix. Communications of the ACM, 4(4):168, 1961. 6, 2

    1961

  22. [30]

    Energy-aware neural architecture optimization with fast splitting steepest descent

    Dilin Wang, Meng Li, Lemeng Wu, Vikas Chandra, and Qiang Liu. Energy-aware neural architecture optimization with fast splitting steepest descent. arXiv preprint arXiv:1910.03103,

    1910 arXiv

  23. [31]

    End-to-end rate- distortion optimized 3d gaussian representation

    Henan Wang, Hanxin Zhu, Tianyu He, Runsen Feng, Jia- jun Deng, Jiang Bian, and Zhibo Chen. End-to-end rate- distortion optimized 3d gaussian representation. arXiv preprint arXiv:2406.01597, 2024. 2

    2024 arXiv

  24. [32]

    Is attention all that nerf needs? arXiv preprint arXiv:2207.13298, 2022

    Peihao Wang, Xuxi Chen, Tianlong Chen, Subhashini Venu- gopalan, Zhangyang Wang, et al. Is attention all that nerf needs? arXiv preprint arXiv:2207.13298, 2022. 2

    2022 arXiv

  25. [33]

    Learn- ing to grow pretrained models for efficient transformer train- ing

    Peihao Wang, Rameswar Panda, Lucas Torroba Henni- gen, Philip Greengard, Leonid Karlinsky, Rogerio Feris, David Daniel Cox, Zhangyang Wang, and Yoon Kim. Learn- ing to grow pretrained models for efficient transformer train- ing. arXiv preprint arXiv:2303.00980, 2023. 3

    2023 arXiv

  26. [34]

    Ibrnet: Learning multi-view image-based rendering

    Qianqian Wang, Zhicheng Wang, Kyle Genova, Pratul P Srini- vasan, Howard Zhou, Jonathan T Barron, Ricardo Martin- Brualla, Noah Snavely, and Thomas Funkhouser. Ibrnet: Learning multi-view image-based rendering. In Proceedings of the IEEE/CVF conference on computer vision and p...

    2021

  27. [35]

    Image quality assessment: from error visibility to structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4):600–612, 2004. 7

    2004

  28. [36]

    Splitting steepest descent for growing neural architectures

    Lemeng Wu, Dilin Wang, and Qiang Liu. Splitting steepest descent for growing neural architectures. Advances in neural information processing systems, 32, 2019. 3, 6

    2019

  29. [37]

    Firefly neural architecture descent: a general approach for growing neural networks

    Lemeng Wu, Bo Liu, Peter Stone, and Qiang Liu. Firefly neural architecture descent: a general approach for growing neural networks. Advances in neural information processing systems, 33:22373–22383, 2020. 3

    2020

  30. [38]

    Steepest descent neural architecture optimization: Escaping local optimum with signed neural splitting

    Lemeng Wu, Mao Ye, Qi Lei, Jason D Lee, and Qiang Liu. Steepest descent neural architecture optimization: Escaping local optimum with signed neural splitting. arXiv preprint arXiv:2003.10392, 2020. 3

    2003 arXiv

  31. [39]

    Residual mixture of experts, 2022

    Lemeng Wu, Mengchen Liu, Yinpeng Chen, Dongdong Chen, Xiyang Dai, and Lu Yuan. Residual mixture of experts, 2022. 3

    2022

  32. [40]

    Point-nerf: Point- based neural radiance fields

    Qiangeng Xu, Zexiang Xu, Julien Philip, Sai Bi, Zhixin Shu, Kalyan Sunkavalli, and Ulrich Neumann. Point-nerf: Point- based neural radiance fields. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5438–5448, 2022. 2

    2022

  33. [41]

    Multiview neural surface reconstruction by disentangling geometry and appear- ance

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

    2020

  34. [42]

    Differentiable surface splatting for point-based geometry processing

    Wang Yifan, Felice Serena, Shihao Wu, Cengiz ¨Oztireli, and Olga Sorkine-Hornung. Differentiable surface splatting for point-based geometry processing. ACM Transactions on Graphics (TOG), 38(6):1–14, 2019. 2

    2019

  35. [43]

    Plenoctrees for real-time rendering of neural radiance fields

    Alex Yu, Ruilong Li, Matthew Tancik, Hao Li, Ren Ng, and Angjoo Kanazawa. Plenoctrees for real-time rendering of neural radiance fields. In Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision, pages 5752–5761,

  36. [44]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In CVPR, 2018. 7

    2018

  37. [45]

    Lp-3dgs: Learning to prune 3d gaussian splatting

    Zhaoliang Zhang, Tianchen Song, Yongjae Lee, Li Yang, Cheng Peng, Rama Chellappa, and Deliang Fan. Lp-3dgs: Learning to prune 3d gaussian splatting. In The Thirty-eighth Annual Conference on Neural Information Processing Sys- tems. 3, 5

  38. [46]

    Ewa splatting

    Matthias Zwicker, Hanspeter Pfister, Jeroen Van Baar, and Markus Gross. Ewa splatting. IEEE Transactions on Visu- alization and Computer Graphics , 8(3):223–238, 2002. 2, 3 Steepest Descent Density Control for Compact 3D Gaussian Splatting Supplementary Material A. Implementat...

    2002

  39. [47]

    When i =i′ andj =j′,∂ϑ(i′) j′ ∇σΠ(x;ϑ(i) j ) =∇2σΠ(x;ϑ(i) j ), and then ∂2 ϑ(i) j ϑ(i) j L(ϑ,w) =w(i)2 j EΠ,x∼D(X) h ∂2 σ(i)σ(i)ℓ eσΠ(x;ϑ(1)),··· ,eσΠ(x;ϑ(n)) ∇σΠ(x;ϑ(i) j )∇σΠ(x;ϑ(i) j )⊤ i +w(i) j EΠ,x∼D(X) h ∂σ(i)ℓ eσΠ(x;ϑ(1)),··· ,eσΠ(x;ϑ(n)) ∇2σΠ(x;ϑ(i) j ) i (22)

  40. [48]

    When i =i′ andj̸=j′,∂ϑ(i′) j′ ∇σΠ(x;ϑ(i) j ) = 0, and thus ∂2 ϑ(i) j ϑ(i) j′ L(ϑ,w) =w(i) j w(i) j′ EΠ,x∼D(X) h ∂2 σ(i)σ(i)ℓ eσΠ(x;ϑ(1)),··· ,eσΠ(x;ϑ(n)) ∇σΠ(x;ϑ(i) j )∇σΠ(x;ϑ(i) j′ )⊤ i (23)

  41. [49]

    When i̸=i′,∂ϑ(i′ ) j′ ∇σΠ(x;ϑ(i) j ) = 0, and henceforth ∂2 ϑ(i) j ϑ(i′ ) j′ L(ϑ,w) =w(i) j w(i′) j′ EΠ,x∼D(X) h ∂2 σ(i)σ(i′)ℓ eσΠ(x;ϑ(1)),··· ,eσΠ(x;ϑ(n)) ∇σΠ(x;ϑ(i) j )∇σΠ(x;ϑ(i′) j′ )⊤ i (24) Using this fact again:ϑ(i) j =θ(i) andeσΠ(x;ϑ(i)) =Pmi j=1w(i) j σΠ(x;θ(i)) =σΠ(x;...

  42. [50]

    When j =j′,∂ϑ(i) j′ ∇σΠ(x;ϑ(i) j ) =∇2σΠ(x;ϑ(i) j ), and then ∂2 ϑ(i) j ϑ(i) j L(θ(\i),ϑ(i),w(i)) ϵ=0 =w(i)2 j EΠ,x∼D(X) h ∂2 σ(i)σ(i)ℓ ··· ,eσΠ(x;ϑ(i)),··· ∇σΠ(x;ϑ(i) j )∇σΠ(x;ϑ(i) j )⊤ i +w(i) j EΠ,x∼D(X) h ∂σ(i)ℓ ··· ,eσΠ(x;ϑ(i)),··· ∇2σΠ(x;ϑ(i) j ) i (30)

  43. [51]

    When j̸=j′,∂ϑ(i) j′ ∇σΠ(x;ϑ(i) j ) = 0, and thus ∂2 ϑ(i) j ϑ(i) j′ L(θ(\i),ϑ(i),w(i)) ϵ=0 =w(i) j w(i) j′ EΠ,x∼D(X) h ∂2 σ(i)σ(i)ℓ ··· ,eσΠ(x;ϑ(i)),··· ∇σΠ(x;ϑ(i) j )∇σΠ(x;ϑ(i) j′ )⊤ i (31) Using this fact again:ϑ(i) j =θ(i) andeσΠ(x;ϑ(i)) =Pmi j=1w(i) j σΠ(x;θ(i)) =σΠ(x;θ(i))...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.