Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

Improving Densification in 3D Gaussian Splatting for High-Fidelity Rendering

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

Pith's one-line read A reworked densification pipeline renders sharper 3D scenes from fewer Gaussians.

desk verdict Solid incremental 3DGS densification improvements, but unablated hyperparameter changes and a contradicted 'no overhead' claim need fixing before the headline claims are credible. read the letter →

arxiv 2508.12313 v1 pith:DUEWCOMZ submitted 2025-08-17 cs.CV

classification cs.CV
keywords 3DGaussianSplattingdensificationnovelviewsynthesisedge-awarescorelong-axissplitadaptivedensitycontroloverfittingreal-timerendering
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 argues that the bottleneck in 3D Gaussian Splatting is not the rendering primitive but the densification schedule that grows the Gaussian set. It proposes replacing the standard clone-and-split procedure with an edge-aware candidate score, a deterministic long-axis split, and a set of overfitting controls, and claims this yields better reconstruction quality while using fewer Gaussians and adding no training or inference overhead. A sympathetic reader would care because the fix is local to the density-control stage, so it can be dropped into existing 3DGS-based systems without changing their CUDA kernels, and it makes high-fidelity real-time rendering cheaper.

What carries the argument

The load-bearing device is Long-Axis Split: instead of probabilistically sampling child positions, each parent is replaced by two children centered on its longest axis at distance $d=0.45L_0$ from the parent center, with the child's longest semi-axis set to $L_0-d$ and its two minor semi-axes set by $R_s=R_0\sqrt{1-d^2/L_0^2}$, so the children are tangent to the parent surface and the before/after shape difference is minimized. Around it, the Edge-Aware Score $S_{i,j}=\sum_{p\in P}\omega^p_{i,j}\alpha^p_{i,j}$ selects which Gaussians split by weighting each Gaussian's per-pixel rendering weight by a Laplacian edge map, and Growth Control imposes a convex budget curve $N=N_{\max}\sqrt{(I-I_{\text{start}})/(I_{\text{end}}-I_{\text{start}})}$ so the Gaussian count peaks only at the end of densification. Together these determine when to densify, which Gaussians to split, and how to split them with minimal disruption.

What would settle it

Run the same method on the Mip-NeRF 360 scenes with the clone operation restored alongside Long-Axis Split while keeping Edge-Aware Score and all overfitting controls fixed; if restoring clones improves PSNR or SSIM on any scene, the clone-free premise is wrong. Alternatively, construct a scene with large uniformly textured under-reconstructed regions, where no Gaussian has high edge-aware score, and check whether splitting alone can densify them.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the geometric disturbance introduced when a Gaussian is cloned or split is itself a major source of reconstruction error, and that this disturbance can be largely removed. The paper claims that scoring split candidates by their summed edge-weighted rendering contribution (Edge-Aware Score), splitting deterministically along the parent's longest axis with child size $R_s=R_0\sqrt{1-d^2/L_0^2}$, opacity scaled to 60%, and pruning/update/growth controls to curb overfitting, together outperform prior densification methods on Mip-NeRF 360, Tanks and Temples, and Deep Blending while using the smallest Gaussian budget.

Load-bearing premise

The paper's strongest unproven premise is that cloning can be dropped entirely because, during optimization, Gaussian sizes converge to a value that already balances under- and over-reconstruction, so every needed densification is a split; if that balance does not hold in some scene, the method has no clone operation to fall back on.

Editorial extensions

If this is right

  • Because the changes are confined to adaptive density control, any 3DGS-based renderer that currently uses the original clone/split schedule can adopt this pipeline as a drop-in replacement.
  • Fewer Gaussians at equal or better quality directly reduces memory footprint and per-frame render cost, which matters for mobile and embedded deployment.
  • The clone-free densification claim, if correct, implies that under-reconstruction can always be repaired by splitting alone, simplifying the densification operator to a single deterministic rule.
  • The overfitting controls (pruning, multi-step updates, growth cap) are decoupled from the split mechanism, so each can be tuned independently when porting to new scene types.
  • The reported training-time reduction (roughly half that of the fastest baseline under the same CUDA kernel) follows from both the lower peak Gaussian count and the multi-step update schedule.

Reading between the lines

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

  • The paper does not isolate the contribution of dropping clone: its w/o LAS ablation removes both Long-Axis Split and the clone operation at once, so restoring cloning alongside LAS would be a direct test of whether the clone-free premise is necessary.
  • The Laplacian edge weighting is only one choice of edge operator, so the Edge-Aware Score could plausibly be combined with perceptual losses or saliency maps, though the paper does not test this.
  • The growth-control curve resembles a progressive-training schedule, so the same budget law might transfer to other resource-limited Gaussian refinement settings such as SLAM or dynamic scenes, an extension the paper does not claim.
  • Because the method sets children's opacity to 60% of the parent, its success suggests that opacity re-balancing after split is a free gain that other densification schemes could adopt independently of the geometric placement.
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 / 4 minor

Summary. The paper proposes a set of improvements to the densification pipeline of 3D Gaussian Splatting, organized around three questions: when to densify (Edge-Aware Score, EAS), how to densify (Long-Axis Split, LAS), and how to mitigate overfitting (Recovery-Aware Pruning, Multi-step Update, Growth Control). The method is evaluated on Mip-NeRF 360, Tanks and Temples, and Deep Blending, with quantitative comparisons against eight recent baselines, per-component ablations, and a same-budget comparison against TamingGS and 3DGS-MCMC. The paper reports consistent gains in PSNR and SSIM with markedly fewer Gaussians, and claims in the Abstract and Conclusion that these gains come 'without introducing additional training or inference overhead.' The main quantitative results are credible at face value, but the no-overhead claim is directly contradicted by the paper's own ablation table, and several non-densification hyperparameter changes in the Implementation Details are not ablated, so the attribution of the reported gains to the densification components is not yet established.

Significance. If the central claims withstand scrutiny, this is a useful contribution: a drop-in densification improvement for 3DGS that improves fidelity and compactness, with code released and component-level ablations. The empirical evidence does show that each proposed component helps on the combined benchmark, and the same-budget comparison in Appendix A.2 is a commendable attempt to control for Gaussian count. However, the paper's headline 'no overhead' claim is false under its own measurements, and the unablated hyperparameter changes noted in the Implementation Details weaken the attribution of the state-of-the-art numbers to the proposed densification techniques. The core idea is defensible and the experiments are extensive, but the manuscript needs substantive revision before the claims can be accepted as stated.

major comments (4)
  1. [Abstract / Conclusion; Table 2, Ablation Experiments] The central claim that the method works 'without introducing additional training or inference overhead' is contradicted by the ablation data in Table 2: Ours Full runs at 289 FPS while the w/o EAS row runs at 346 FPS, a 16% reduction in rendering speed, and the ablation text in 'Effect of EAS' explicitly acknowledges that 'EAS increases rendering overhead.' The same table also shows training time rising from 6.4 to 6.7 minutes. Please remove the no-overhead claim from the Abstract and Conclusion, or rephrase it as an explicit quality-versus-speed trade-off with the overhead quantified.
  2. [Appendix: Implementation Details; Tables 1 and 2] The implementation applied to Ours but not to the baselines includes position_lr_init=0.00004, position_lr_final=0.000002, opacity_reset_threshold=0.05, and removal of large-Gaussian pruning. These changes are catalogued in the Implementation Details but are never ablated. Since Table 2 shows that each proposed component contributes roughly 0.1-0.3 dB PSNR on the combined datasets, the 0.71 dB gain over 3DGS on Mip-NeRF 360 could plausibly be driven in part by these unablated hyperparameters. Please ablate the proposed components while keeping the original 3DGS hyperparameters, or apply the modified settings to all baselines, including the same-budget comparison in Appendix A.2.
  3. [Methods, Long-Axis Split; Table 2, w/o LAS row] The decision to abandon the clone operation is justified in the text by the premise that 'during optimization, the size of Gaussians tends to converge to a value that balances under-reconstruction and over-reconstruction,' but that premise is not proven and the w/o LAS ablation removes both LAS and the clone operation simultaneously. The reader therefore cannot tell whether the measured improvement comes from the long-axis split geometry, from dropping clone, or from both. Please add ablations that (i) keep clone with LAS and (ii) use the original split operator without clone, so the contribution of each design choice is isolated.
  4. [Appendix A.4, Eq. (9)] The claimed proof that Rs = R0 * sqrt(1 - d^2/L0^2) minimizes the geometric difference after splitting is presented as an 'intuitive geometric comparison,' but the written argument does not define the overlapping and non-overlapping area expressions that are being compared, nor does it derive the optimum of their difference. Since Eq. (9) is used to motivate a load-bearing design choice, please either provide a rigorous derivation or explicitly label the formula as a heuristic that is supported only by the sensitivity study in Figure 12.
minor comments (4)
  1. [Methods, Edge-Aware Score, Eq. (7)] The number Ns of sampled views used to compute the average edge-aware score is never specified; please state its value and the exact threshold used for the absolute gradient criterion in Eq. (8).
  2. [Experiments, Quantitative Analysis] The statement that 'all results represent the best performance obtained over three runs' is statistically weak; please report mean and standard deviation across runs, or justify why best-of-three is the appropriate summary for this benchmark.
  3. [Methods, Recovery-Aware Pruning] The choice of pruning at iterations 3300 and 6300 is motivated by the opacity reset schedule, but the exact relationship between these iterations and the 3DGS resets at 3K, 6K, 9K, and 12K is not stated; please make this connection explicit.
  4. [Methods, heading] The section heading 'Muti-view Update' contains a typo and should read 'Multi-view Update.'

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the densification components are defined independently, ablated against the same budget, and benchmarked against external baselines; the unablated hyperparameter changes are an attribution confound, not a circular reduction.

full rationale

The claimed derivation chain is self-contained rather than circular. Edge-Aware Score defines a new candidate-selection signal from edge weights and rendering alpha (Eqs. 6-8), and its contribution is directly ablated in Table 2 against a same-budget configuration; it is not derived from the quality metric it is supposed to improve. Long-Axis Split is motivated by a geometric area-minimization calculation (Eq. 9, Appendix A.4), with the split distance and opacity reduction treated as empirically chosen and separately ablated values (Appendix A.8), not as fitted predictions of the final PSNR or LPIPS. Recovery-Aware Pruning, Multi-step Update, and Growth Control are operational training procedures with explicit ablations (Table 2 and Appendices A.6, A.9). No load-bearing conclusion is justified by a self-citation: the paper compares against externally published baselines on the standard Mip-NeRF 360, Tanks and Temples, and Deep Blending benchmarks, and the central quality claims are supported by those comparisons. The unablated hyperparameter changes listed in Implementation Details (position_lr_init/final, opacity_reset_threshold, removal of large-Gaussian pruning) are a real fairness and attribution concern for the headline gains, but they do not make any reported result equivalent to its own input by construction; the proposed components retain independent empirical content. Similarly, the claim that cloning can be dropped rests on an unproven convergence premise and is not fully isolated by the w/o LAS ablation, but that is an experimental-support gap rather than circularity. Overall, the derivation does not reduce to its inputs, so the appropriate score is 0.

Assumptions & free parameters 9 free parameters · 5 assumptions · 0 invented entities

The method introduces many hand-tuned hyperparameters: split distance, opacity ratio, prune ratio, update intervals, growth curve, budgets, and thresholds. None of these are derived from theory; each is validated by ablations on the same benchmark datasets used for the final comparison. The geometric proof for the optimal child radius is an intuitive 2D argument, not a rigorous derivation. The method also relies on domain assumptions from prior 3DGS work, such as edge weights as perceptual proxy and absolute gradients as loss proxy, without independent validation.

free parameters (9)
  • EAS gradient threshold = 0.0003
    Used to select candidate Gaussians for splitting; adjusted from the 3DGS default as described in Implementation Details.
  • Split distance ratio d/L0 = 0.45
    Chosen empirically to balance shape difference and overlap; ablation in Appendix A.8.
  • Child opacity ratio = 0.6
    Child Gaussians initialized to 60% of parent opacity; empirically tuned in the ablation of opacity reduction rate.
  • RAP prune ratio and iterations = bottom 20% at iterations 3300 and 6300
    Hand-chosen to remove overfitted Gaussians early in training.
  • MU update intervals = N=5 between 15k and 22.5k, N=20 afterward
    Hand-chosen two-stage schedule; supporting ablation in Appendix Figure 13.
  • Growth Control curve exponent = square root (convex)
    Smooth convex curve chosen to delay the peak in Gaussian count until the end of densification.
  • Gaussian budgets per scene = 3M, 1.5M, 1M
    User-defined final budgets assigned by scene size categories.
  • Position learning rates = init 4e-5, final 2e-6
    Modified learning rates listed in Implementation Details.
  • Opacity reset threshold = 0.05
    Changed from the 3DGS default of 0.01.
assumptions (5)
  • standard math The 3DGS differentiable rendering and adaptive density control framework from Kerbl et al. is correct and applicable.
    Preliminaries restate the 3DGS formulation; the paper builds on it without reassessing its correctness.
  • ad hoc to paper Gaussian sizes converge to a balance between under- and over-reconstruction during optimization, making clone unnecessary.
    Stated in the Long-Axis Split section; not proven and not isolated in the ablations.
  • domain assumption Laplacian edge weights from training images are a valid proxy for perceptual importance in choosing split candidates.
    Motivates Edge-Aware Score; no perceptual validation beyond downstream metrics.
  • domain assumption The 2D ellipse simplification of the 3D split captures the relevant geometry for the optimal Rs derivation.
    Appendix A.4 simplifies a 3D Gaussian to a 2D ellipse; the proof is intuitive and visual.
  • domain assumption Absolute coordinate gradients approximate pixel-wise loss for identifying under-reconstruction.
    Used as the second criterion in EAS; follows the rationale of AbsGS.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improving Densification in 3D Gaussian Splatting for High-Fidelity Rendering." pith.science (2026). https://pith.science/paper/DUEWCOMZ

@misc{pith2026250812313,
  author       = {Pith},
  title        = {Pith review of: Improving Densification in 3D Gaussian Splatting for High-Fidelity Rendering},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DUEWCOMZ}},
  note         = {Machine review of arXiv:2508.12313}
}
read the original abstract

Although 3D Gaussian Splatting (3DGS) has achieved impressive performance in real-time rendering, its densification strategy often results in suboptimal reconstruction quality. In this work, we present a comprehensive improvement to the densification pipeline of 3DGS from three perspectives: when to densify, how to densify, and how to mitigate overfitting. Specifically, we propose an Edge-Aware Score to effectively select candidate Gaussians for splitting. We further introduce a Long-Axis Split strategy that reduces geometric distortions introduced by clone and split operations. To address overfitting, we design a set of techniques, including Recovery-Aware Pruning, Multi-step Update, and Growth Control. Our method enhances rendering fidelity without introducing additional training or inference overhead, achieving state-of-the-art performance with fewer Gaussians.

Figures

Figures reproduced from arXiv: 2508.12313 by the authors.

Figure 1
Figure 1. Compare the geometric differences before and af [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Evaluate the drop in PSNR after splitting using [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Qualitative comparison results among scenes garden, drjohnson, train. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison for evaluating the effectiveness of EAS. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Evaluating the impact of LAS on optimization [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Qualitative comparison for evaluating the effec [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 9
Figure 9. Figure 9: Qualitative comparison results among scenes bicycle, flowers, stump, treehill, bonsai. [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: Qualitative comparison results among scenes counter, kitchen, room, playroom, truck. [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]
Figure 11
Figure 11. Figure 11: Schematic illustration of the split operation when simplified to a 2D ellipse. [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]
Figure 12
Figure 12. Figure 12: The trend of LPIPS with respect to Rs in the stump scene. The red dot indicates the value we chose in practice. [PITH_FULL_IMAGE:figures/full_fig_p012_12.png]
Figure 13
Figure 13. Figure 13: The impact of different MU starting iterations on LPIPS, test scene is bicycle. The update interval for parameters is [PITH_FULL_IMAGE:figures/full_fig_p013_13.png]
Figure 14
Figure 14. Figure 14: The trend of LPIPS with respect to opacity reduction rate in the stump scene. The red dot indicates the value we [PITH_FULL_IMAGE:figures/full_fig_p014_14.png]
Figure 15
Figure 15. Figure 15: Testing the impact of the distance d between the sub-Gaussian and the original Gaussian centers on quality and [PITH_FULL_IMAGE:figures/full_fig_p014_15.png]
Figure 16
Figure 16. Figure 16: Growth curves of Gaussians with and without GC, test scene is bicycle. [PITH_FULL_IMAGE:figures/full_fig_p015_16.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. The Role of Initialization in 3D Gaussian Splatting

    cs.CV 2026-03 unverdicted novelty 6.0 of 10

    Dense initialization of 3DGS does not consistently beat sparse SfM initialization for standard novel views, but improves off-trajectory generalization; no densification method wins everywhere.

Reference graph

Works this paper leans on

29 extracted references · 12 canonical work pages · cited by 1 Pith paper

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    T.; Mildenhall, B.; Verbin, D.; Srinivasan, P

    Barron, J. T.; Mildenhall, B.; Verbin, D.; Srinivasan, P. P.; and Hedman, P. 2022. Mip-nerf 360: Unbounded anti-aliased neural radiance fields. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 5470--5479

  4. [4]

    Chen, Z.; Wang, F.; Wang, Y.; and Liu, H. 2024. Text-to-3d using gaussian splatting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 21401--21412

  5. [5]

    Cheng, K.; Long, X.; Yang, K.; Yao, Y.; Yin, W.; Ma, Y.; Wang, W.; and Chen, X. 2024. Gaussianpro: 3d gaussian splatting with progressive propagation. In Forty-first International Conference on Machine Learning

  6. [6]

    Fang, G.; and Wang, B. 2024. Mini-splatting: Representing scenes with a constrained number of gaussians. In European Conference on Computer Vision, 165--181. Springer

  7. [7]

    Fridovich-Keil, S.; Yu, A.; Tancik, M.; Chen, Q.; Recht, B.; and Kanazawa, A. 2022. Plenoxels: Radiance fields without neural networks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 5501--5510

  8. [8]

    Hedman, P.; Philip, J.; Price, T.; Frahm, J.-M.; Drettakis, G.; and Brostow, G. 2018. Deep blending for free-viewpoint image-based rendering. ACM Transactions on Graphics (ToG), 37(6): 1--15

Show all 29 references
  1. [9]

    Kerbl, B.; Kopanas, G.; Leimk \"u hler, T.; and Drettakis, G. 2023. 3D Gaussian Splatting for Real-Time Radiance Field Rendering. ACM Trans. Graph., 42(4): 139--1

  2. [10]

    Kheradmand, S.; Rebain, D.; Sharma, G.; Sun, W.; Tseng, Y.-C.; Isack, H.; Kar, A.; Tagliasacchi, A.; and Yi, K. M. 2024. 3d gaussian splatting as markov chain monte carlo. Advances in Neural Information Processing Systems, 37: 80965--80986

  3. [11]

    Knapitsch, A.; Park, J.; Zhou, Q.-Y.; and Koltun, V. 2017. Tanks and temples: Benchmarking large-scale scene reconstruction. ACM Transactions on Graphics (ToG), 36(4): 1--13

  4. [12]

    Lee, B.; Lee, H.; Sun, X.; Ali, U.; and Park, E. 2024. Deblurring 3d gaussian splatting. In European Conference on Computer Vision, 127--143. Springer

  5. [13]

    Lin, Y.; Dai, Z.; Zhu, S.; and Yao, Y. 2024. Gaussian-flow: 4d reconstruction with dynamic 3d gaussian particle. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 21136--21145

  6. [14]

    S.; Goel, R.; Kerbl, B.; Steinberger, M.; Carrasco, F

    Mallick, S. S.; Goel, R.; Kerbl, B.; Steinberger, M.; Carrasco, F. V.; and De La Torre, F. 2024. Taming 3dgs: High-quality radiance fields with limited resources. In SIGGRAPH Asia 2024 Conference Papers, 1--11

  7. [15]

    H.; and Davison, A

    Matsuki, H.; Murai, R.; Kelly, P. H.; and Davison, A. J. 2024. Gaussian splatting slam. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 18039--18048

  8. [16]

    P.; Tancik, M.; Barron, J

    Mildenhall, B.; Srinivasan, P. P.; Tancik, M.; Barron, J. T.; Ramamoorthi, R.; and Ng, R. 2021. Nerf: Representing scenes as neural radiance fields for view synthesis. Communications of the ACM, 65(1): 99--106

  9. [17]

    M \"u ller, T.; Evans, A.; Schied, C.; and Keller, A. 2022. Instant neural graphics primitives with a multiresolution hash encoding. ACM transactions on graphics (TOG), 41(4): 1--15

  10. [18]

    Rota Bul \`o , S.; Porzi, L.; and Kontschieder, P. 2024. Revising densification in gaussian splatting. In European Conference on Computer Vision, 347--362. Springer

  11. [19]

    L.; and Frahm, J.-M

    Schonberger, J. L.; and Frahm, J.-M. 2016. Structure-from-motion revisited. In Proceedings of the IEEE conference on computer vision and pattern recognition, 4104--4113

  12. [20]

    Shao, Z.; Wang, Z.; Li, Z.; Wang, D.; Lin, X.; Zhang, Y.; Fan, M.; and Wang, Z. 2024. Splattingavatar: Realistic real-time human avatars with mesh-embedded gaussian splatting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 1606--1616

  13. [21]

    Wang, P.; Wang, Y.; Wang, D.; Mohan, S.; Fan, Z.; Wu, L.; Cai, R.; Yeh, Y.-Y.; Wang, Z.; Liu, Q.; et al. 2025. Steepest Descent Density Control for Compact 3D Gaussian Splatting. In Proceedings of the Computer Vision and Pattern Recognition Conference, 26663--26672

  14. [22]

    Wu, G.; Yi, T.; Fang, J.; Xie, L.; Zhang, X.; Wei, W.; Liu, W.; Tian, Q.; and Wang, X. 2024. 4d gaussian splatting for real-time dynamic scene rendering. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 20310--20320

  15. [23]

    Yang, Z.; Gao, X.; Sun, Y.-T.; Huang, Y.; Lyu, X.; Zhou, W.; Jiao, S.; Qi, X.; and Jin, X. 2024. Spec-gaussian: Anisotropic view-dependent appearance for 3d gaussian splatting. Advances in Neural Information Processing Systems, 37: 61192--61216

  16. [24]

    Ye, Z.; Li, W.; Liu, S.; Qiao, P.; and Dou, Y. 2024. Absgs: Recovering fine details in 3d gaussian splatting. In Proceedings of the 32nd ACM International Conference on Multimedia, 1053--1061

  17. [25]

    Yu, Z.; Chen, A.; Huang, B.; Sattler, T.; and Geiger, A. 2024. Mip-splatting: Alias-free 3d gaussian splatting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 19447--19456

  18. [26]

    A.; Shechtman, E.; and Wang, O

    Zhang, R.; Isola, P.; Efros, A. A.; Shechtman, E.; and Wang, O. 2018. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recognition, 586--595

  19. [27]

    Zhang, Z.; Hu, W.; Lao, Y.; He, T.; and Zhao, H. 2024. Pixel-gs: Density control with pixel-aware gradient for 3d gaussian splatting. In European Conference on Computer Vision, 326--342. Springer

  20. [28]

    Zhou, H.; and Ni, Z. 2025. Perceptual-GS: Scene-adaptive Perceptual Densification for Gaussian Splatting. arXiv preprint arXiv:2506.12400

  21. [29]

    Zhou, X.; Lin, Z.; Shan, X.; Wang, Y.; Sun, D.; and Yang, M.-H. 2024. Drivinggaussian: Composite gaussian splatting for surrounding dynamic autonomous driving scenes. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 21634--21643

Pith tools

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