pith. machine review for the scientific record. sign in

arxiv: 2605.11266 · v1 · submitted 2026-05-11 · 💻 cs.CV · cs.GR· cs.LG

Recognition: 2 theorem links

· Lean Theorem

PG-3DGS: Optimizing 3D Gaussian Splatting to Satisfy Physics Objectives

Zachary Lee , Maxwell Jacobson , Yexiang Xue
Authors on Pith no claims yet

Pith reviewed 2026-05-13 06:30 UTC · model grok-4.3

classification 💻 cs.CV cs.GRcs.LG
keywords 3D Gaussian SplattingDifferentiable PhysicsPhysics-Guided OptimizationFunctional 3D ModelingAerodynamic LiftPouring Simulation3D Reconstruction
0
0 comments X

The pith

PG-3DGS couples differentiable physics simulation to 3D Gaussian parameters so that optimized shapes satisfy physical objectives such as pouring and lift while keeping visual quality.

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

The paper introduces Physics-Guided 3D Gaussian Splatting to move beyond purely visual 3D scene generation. It integrates physical objectives into the optimization loop so that the resulting shapes exhibit real behaviors like pouring liquids or generating aerodynamic lift. The coupling happens by running differentiable physics simulations directly on the Gaussian parameters and combining those losses with standard visual reconstruction terms. A sympathetic reader would care because this produces 3D assets that are immediately usable for physical tasks rather than requiring separate engineering steps after visual modeling. Experiments demonstrate gains on pouring and lift tasks in simulation plus higher measured lift on three 3D-printed aircraft compared with appearance-only baselines.

Core claim

By allowing physical objectives to guide the shape optimization process alongside visual losses, PG-3DGS produces geometries that are not only photometrically accurate but also physically functional. The model learns to adjust shapes so that the generated objects exhibit physically meaningful behaviors, for example, teapots that can pour and airplanes that can generate lift, without sacrificing visual quality. Bench-top physical lift tests with 3D-printed aircraft under identical airflow conditions show higher scale-measured lift for PG-3DGS generated structures than an appearance-matching baseline in all three cases.

What carries the argument

The direct coupling of differentiable physics simulation to the parameters of 3D Gaussian splats, which supplies gradients that steer the optimization toward both photometric and physical goals.

If this is right

  • Teapots and similar vessels generated by the method pour liquids as intended.
  • Aircraft shapes achieve higher lift forces both in simulation and in physical scale tests.
  • Visual rendering quality remains comparable to standard 3D Gaussian Splatting.
  • The same optimization loop can in principle be applied to other physics objectives once the differentiable simulator is available.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The framework could be extended to optimize for additional physical properties such as structural stability or fluid containment.
  • Direct use of the optimized Gaussian parameters for manufacturing would remove the usual reconstruction-to-CAD conversion step.
  • Similar physics-guided losses might be added to other 3D representations that support differentiable rendering.

Load-bearing premise

Differentiable physics simulation can be coupled directly to the Gaussian parameters without introducing instabilities or requiring task-specific tuning beyond the demonstrated cases.

What would settle it

If a PG-3DGS-optimized airplane is 3D-printed and tested under the same airflow conditions as a visual-only optimized version yet produces equal or lower lift on the scale, the claim of improved physical functionality collapses.

Figures

Figures reproduced from arXiv: 2605.11266 by Maxwell Jacobson, Yexiang Xue, Zachary Lee.

Figure 1
Figure 1. Figure 1: Our proposed Physics-Guided 3D Gaussian Splatting (PG-3DGS) is able to [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Method overview: Visual supervision uses calibrated images and cameras; physics [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparative results for two teapots. Top row: renders with ground truth view [PITH_FULL_IMAGE:figures/full_fig_p027_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Full Render Comparison (Teapots 01–06). All methods are rendered from the [PITH_FULL_IMAGE:figures/full_fig_p028_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Geometry cutaway Comparison (Teapots 01–06). The first column is the reference [PITH_FULL_IMAGE:figures/full_fig_p029_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Physics Simulation Comparison (Teapots 01–06). Snapshot taken at final frame [PITH_FULL_IMAGE:figures/full_fig_p030_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Training progression for Teapot 05. Midplane cross-section views are shown at [PITH_FULL_IMAGE:figures/full_fig_p031_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Trajectories for paper plane over optimization process. The z position of the [PITH_FULL_IMAGE:figures/full_fig_p032_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Comparison of airplane motion between the baseline (3DGS) and our method. [PITH_FULL_IMAGE:figures/full_fig_p033_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Benchtop physical lift tests under identical blower airflow. Each row corresponds [PITH_FULL_IMAGE:figures/full_fig_p034_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: This plot shows that while the satisfaction of the physics objective does increase [PITH_FULL_IMAGE:figures/full_fig_p036_11.png] view at source ↗
read the original abstract

Recent advances in Gaussian Splatting have enabled fast, high-fidelity 3D scene generation, yet these methods remain purely visual and lack an understanding of how shapes behave in the physical world. We introduce Physics-Guided 3D Gaussian Splatting (PG-3DGS), a framework that couples differentiable physics simulation with 3D Gaussian representations to generate 3D structures satisfying physics functionalities. By allowing physical objectives to guide the shape optimization process alongside visual losses, our approach produces geometries that are not only photometrically accurate but also physically functional. The model learns to adjust shapes so that the generated objects exhibit physically meaningful behaviors, for example, teapots that can pour and airplanes that can generate lift, without sacrificing visual quality. Experiments on pouring and aerodynamic lift tasks show that PG-3DGS improves physical functionality while preserving visual quality. In addition to simulation gains, bench-top physical lift tests with 3D-printed aircraft (Cessna, B-2 Spirit, and paper plane) under identical airflow conditions show higher scale-measured lift for PG-3DGS, generated structures than an appearance-matching baseline in all three cases. Our unified framework connects appearance-based reconstruction with physics-based reasoning, enabling end-to-end generation of 3D structures that both look realistic and function correctly.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper introduces Physics-Guided 3D Gaussian Splatting (PG-3DGS), which augments standard 3DGS optimization with differentiable physics losses so that the resulting Gaussian parameters produce shapes satisfying both photometric reconstruction and task-specific physical objectives (fluid pouring for teapots; aerodynamic lift for aircraft). Experiments report quantitative gains in simulated physical metrics while preserving visual quality, plus real-world scale measurements on three 3D-printed aircraft showing higher lift for the physics-guided models than an appearance-only baseline.

Significance. If the differentiable coupling between Gaussian parameters and the physics simulator is stable and reproducible, the work provides a concrete bridge between appearance-driven 3D reconstruction and physics-based reasoning. The real-world lift validation on printed models is a notable strength that moves beyond pure simulation. The approach could enable generation of functionally valid 3D assets without post-hoc engineering.

major comments (3)
  1. [§3] §3 (Method), the paragraph describing the physics coupling: the manuscript does not specify how the continuous Gaussian density field is converted into the input required by the physics simulator (mesh extraction, voxelization, or soft-density treatment). Without this conversion step and its gradient path, it is impossible to verify that back-propagation through means, covariances, and opacities remains stable for contact-rich or turbulent regimes.
  2. [§4.2] §4.2 (Experiments), the pouring and lift results: the physics loss weight is treated as a free hyper-parameter. No ablation is shown on its sensitivity, nor is it demonstrated that the same weight works across tasks without retuning. This directly bears on the claim that the framework generalizes beyond the two demonstrated tasks.
  3. [§4.3] §4.3 (Real-world validation): the lift measurements on the three 3D-printed models are reported as uniformly higher, but the manuscript provides no details on printing resolution, surface finishing, exact airflow conditions, or statistical testing. These omissions leave open whether the measured lift differences can be attributed to the physics-guided geometry rather than fabrication or measurement variance.
minor comments (2)
  1. [Figure 3] Figure 3 caption: the legend for the baseline vs. PG-3DGS curves is difficult to read at print size; consider increasing font size or adding a table of final metric values.
  2. [§3] Notation: the symbol for the combined loss (visual + physics) is introduced without an explicit equation number; add Eq. (X) for clarity when first used.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their thorough review and constructive comments. We address each of the major comments below and will revise the manuscript accordingly to improve clarity and completeness.

read point-by-point responses

  1. Referee: [§3] §3 (Method), the paragraph describing the physics coupling: the manuscript does not specify how the continuous Gaussian density field is converted into the input required by the physics simulator (mesh extraction, voxelization, or soft-density treatment). Without this conversion step and its gradient path, it is impossible to verify that back-propagation through means, covariances, and opacities remains stable for contact-rich or turbulent regimes.

    Authors: We agree that this information is essential for reproducibility and verification. We will revise the manuscript to fully specify the conversion from the Gaussian density field to the physics simulator's input format, along with the corresponding gradient propagation path. This addition will address concerns regarding stability in contact-rich or turbulent regimes. revision: yes

  2. Referee: [§4.2] §4.2 (Experiments), the pouring and lift results: the physics loss weight is treated as a free hyper-parameter. No ablation is shown on its sensitivity, nor is it demonstrated that the same weight works across tasks without retuning. This directly bears on the claim that the framework generalizes beyond the two demonstrated tasks.

    Authors: We acknowledge the value of such an ablation for supporting generalizability. We will add experiments in the revised version that ablate the physics loss weight across a range of values for both the pouring and lift tasks, reporting effects on physical performance and visual fidelity. We will also demonstrate the use of a consistent weight across tasks. revision: yes

  3. Referee: [§4.3] §4.3 (Real-world validation): the lift measurements on the three 3D-printed models are reported as uniformly higher, but the manuscript provides no details on printing resolution, surface finishing, exact airflow conditions, or statistical testing. These omissions leave open whether the measured lift differences can be attributed to the physics-guided geometry rather than fabrication or measurement variance.

    Authors: We thank the referee for this feedback. We will update the real-world validation section to include specifics on the 3D printing parameters (resolution and material), surface finishing process, exact airflow conditions during testing, and results of statistical analysis (including means, variances, and significance tests) over repeated measurements to confirm that the lift differences are due to the optimized geometry. revision: yes

Circularity Check

0 steps flagged

No significant circularity; physics objectives are externally sourced

full rationale

The paper's central derivation couples an external differentiable physics simulator to 3D Gaussian parameters for tasks such as pouring and lift generation. Physical objectives (fluid flow, lift coefficients) originate from the simulator rather than being defined in terms of the fitted Gaussian means, covariances, or opacities. No equations reduce the claimed prediction to a self-fit, no load-bearing self-citations are invoked to justify uniqueness, and no ansatz is smuggled via prior author work. The optimization simply adds independent physics losses to visual losses, preserving self-contained structure against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the existence of a differentiable physics simulator whose gradients can be back-propagated into Gaussian parameters, plus the assumption that visual and physical losses can be balanced without one dominating the other. No new physical constants or particles are introduced.

free parameters (1)
  • physics loss weight
    Scalar balancing visual and physical objectives; must be chosen per task.
axioms (1)
  • domain assumption Differentiable physics simulator exists and is stable for the chosen tasks
    Invoked when the paper states that physical objectives guide shape optimization.

pith-pipeline@v0.9.0 · 5541 in / 1419 out tokens · 62835 ms · 2026-05-13T06:30:22.888826+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages · 4 internal anchors

  1. [1]

    L. Yang, Z. Zhang, Y. Song, S. Hong, R. Xu, Y. Zhao, W. Zhang, B. Cui, M.-H. Yang, Diffusion models: A comprehensive survey of methods and applications, ACM Comput. Surv. 56 (4) (Nov. 2023). doi:10.1145/3626235. URLhttps://doi.org/10.1145/3626235 40

    work page doi:10.1145/3626235 2023

  2. [2]

    Z. Xing, Q. Feng, H. Chen, Q. Dai, H. Hu, H. Xu, Z. Wu, Y.-G. Jiang, A survey on video diffusion models, ACM Comput. Surv. 57 (2) (Nov. 2024).doi:10.1145/3696415. URLhttps://doi.org/10.1145/3696415

    work page doi:10.1145/3696415 2024

  3. [3]

    W. X. Zhao, K. Zhou, J. Li, T. Tang, X. Wang, Y. Hou, Y. Min, B. Zhang, J. Zhang, Z. Dong, Y. Du, C. Yang, Y. Chen, Z. Chen, J. Jiang, R. Ren, Y. Li, X. Tang, Z. Liu, P. Liu, J.-Y. Nie, J.-R. Wen, A survey of large language models, arXiv preprint arXiv:2303.18223 (2023). URLhttp://arxiv.org/abs/2303.18223

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  4. [4]

    2451–2463,https://worldmodels.github.io

    D.Ha, J.Schmidhuber, Recurrentworldmodelsfacilitatepolicyevolution, in: Advances in Neural Information Processing Systems 31, Curran As- sociates, Inc., 2018, pp. 2451–2463,https://worldmodels.github.io. URL https://papers.nips.cc/paper/ 7512-recurrent-world-models-facilitate-policy-evolution

    work page 2018

  5. [5]

    J. Ding, Y. Zhang, Y. Shang, Y. Zhang, Z. Zong, J. Feng, Y. Yuan, H. Su, N. Li, N. Sukiennik, et al., Understanding world or predicting future? a comprehensive survey of world models, ACM Computing Surveys 58 (3) (2025) 1–38

    work page 2025

  6. [6]

    Mildenhall, B

    B. Mildenhall, B. Mildenhall, P. P. Srinivasan, P. P. Srinivasan, M. Tancik, M.Tancik, J.T.Barron, J.T.Barron, R.Ramamoorthi, R.Ramamoorthi, R. Ng, R. Ng, NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis, arXiv: Computer Vision and Pattern Recognition (2020).doi:10.1145/3503250

    work page doi:10.1145/3503250 2020

  7. [7]

    https://doi.org/10.1145/3592433 Xiaonan Kong and Riley G

    B. Kerbl, G. Kopanas, T. Leimkühler, G. Drettakis, 3D Gaussian Splat- ting for Real-Time Radiance Field Rendering, ACM Transactions on Graphics (2023).doi:10.1145/3592433

    work page doi:10.1145/3592433 2023

  8. [8]

    Poole, A

    B. Poole, A. Jain, J. T. Barron, B. Mildenhall, Dreamfusion: Text-to-3d using 2d diffusion, arXiv (2022)

    work page 2022

  9. [9]

    C.-H. Lin, J. Gao, L. Tang, T. Takikawa, X. Zeng, X. Huang, K. Kreis, S. Fidler, M.-Y. Liu, T.-Y. Lin, Magic3d: High-resolution text-to-3d content creation, in: IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 41

    work page 2023

  10. [10]

    G. Wu, T. Yi, J. Fang, L. Xie, X. Zhang, W. Wei, W. Liu, Q. Tian, X. Wang, 4D Gaussian Splatting for Real-Time Dynamic Scene Render- ing, Computer Vision and Pattern Recognition (2023).doi:10.1109/ cvpr52733.2024.01920

    work page arXiv 2023

  11. [11]

    J. Tang, J. Ren, H. Zhou, Z. Liu, G. Zhang, DreamGaussian: Generative Gaussian Splatting for Efficient 3D Content Creation, International Conference on Learning Representations (2023).doi:10.48550/arxiv. 2309.16653

    work page internal anchor Pith review doi:10.48550/arxiv 2023

  12. [12]

    In: CVPR, pp

    Z. Liang, Q. Zhang, Y. Feng, Y. Shan, K. Jia, GS-IR: 3D Gaussian Splat- ting for Inverse Rendering, Computer Vision and Pattern Recognition (2023).doi:10.1109/cvpr52733.2024.02045

    work page doi:10.1109/cvpr52733.2024.02045 2023

  13. [13]

    Z. Cao, Z. Chen, L. Pan, Z. Liu, Physx-3d: Physical-grounded 3d asset generation, arXiv preprint arXiv:2507.12465 (2025)

    work page arXiv 2025

  14. [14]

    Schenck, D

    C. Schenck, D. Fox, SPNets: Differentiable Fluid Dynamics for Deep Neural Networks, Conference on Robot Learning (2018).doi:null

    work page 2018

  15. [15]

    Xu, C.-y

    M. Xu, C.-y. Song, D. Levin, D. Hyde, A Differentiable Material Point Method Framework for Shape Morphing, IEEE Transactions on Visualiza- tion and Computer Graphics (2024).doi:10.1109/tvcg.2025.3591729

    work page doi:10.1109/tvcg.2025.3591729 2024

  16. [16]

    Y. Hu, Y. Hu, L. Anderson, L. Anderson, T.-M. Li, T.-M. Li, Q. Sun, Q. Sun, N. Carr, N. A. Carr, J. Ragan-Kelley, J. Ragan-Kelley, F. Du- rand, F. Durand, DiffTaichi: Differentiable Programming for Physical Simulation, International Conference on Learning Representations (2019). doi:null

    work page 2019

  17. [17]

    P. Holl, V. Koltun, K. Um, N. Thuerey, phiflow: A Differentiable PDE Solving Framework for Deep Learning via Physical Simulations (2020). URLhttps://telecom-paris.hal.science/hal-03142278

    work page 2020

  18. [18]

    Takahashi, J

    T. Takahashi, J. Liang, Y.-L. Qiao, M. C. Lin, Differentiable Fluids with Solid Coupling for Learning and Control, AAAI Conference on Artificial Intelligence (2021).doi:10.1609/aaai.v35i7.16764

    work page doi:10.1609/aaai.v35i7.16764 2021

  19. [19]

    Franz, H

    A. Franz, H. Wei, L. Guastoni, N. Thuerey, PICT - A Differentiable, GPU- Accelerated Multi-Block PISO Solver for Simulation-Coupled Learning 42 Tasks in Fluid Dynamics, Journal of Computational Physics (2025). doi:10.48550/arxiv.2505.16992

    work page doi:10.48550/arxiv.2505.16992 2025

  20. [20]

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations

    M. Raissi, P. Perdikaris, G. Karniadakis, Physics-informed neu- ral networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equa- tions, Journal of Computational Physics 378 (2019) 686–707. doi:https://doi.org/10.1016/j.jcp.2018.10.045. URL https://www.sciencedirect.com/science/article/pii/ S00...

    work page doi:10.1016/j.jcp.2018.10.045 2019

  21. [21]

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, L. Yang, Physics-informed machine learning, Nature Reviews Physics 3 (6) (2021) 422–440.doi:10.1038/s42254-021-00314-5. URLhttps://doi.org/10.1038/s42254-021-00314-5

    work page doi:10.1038/s42254-021-00314-5 2021

  22. [22]

    S. Wang, Y. Teng, P. Perdikaris, Understanding and mitigating gradient flow pathologies in physics-informed neural networks, SIAM Journal on Scientific Computing 43 (5) (2021) A3055–A3081.arXiv:https: //doi.org/10.1137/20M1318043,doi:10.1137/20M1318043. URLhttps://doi.org/10.1137/20M1318043

    work page doi:10.1137/20m1318043 2021

  23. [23]

    Y. Feng, X. Feng, Y. Shang, Y. Jiang, C. Yu, Z. Zong, T. Shao, H. Wu, K. Zhou, C. Jiang, Y. Yang, Gaussian Splashing: Unified Particles for Versatile Motion Synthesis and Rendering, Computer Vision and Pattern Recognition (2024).doi:10.1109/cvpr52734.2025.00057

    work page doi:10.1109/cvpr52734.2025.00057 2024

  24. [24]

    Zhang, H.-X

    T. Zhang, H.-X. Yu, R. Wu, B. Y. Feng, C. Zheng, N. Snavely, J. Wu, W. T. Freeman, PhysDreamer: Physics-based interaction with 3d objects via video generation, in: European Conference on Computer Vision, Springer, 2024

    work page 2024

  25. [25]

    B. Chen, H. Jiang, S. Liu, S. Gupta, Y. Li, H. Zhao, S. Wang, Physgen3d: Crafting a miniature interactive world from a single image, CVPR (2025)

    work page 2025

  26. [26]

    Dhariwal, A

    P. Dhariwal, A. Nichol, Diffusion Models Beat GANs on Image Synthesis, Neural Information Processing Systems (2021).doi:null

    work page 2021

  27. [27]

    J. Ho, J. Ho, T. Salimans, T. Salimans, Classifier-Free Diffusion Guidance, arXiv.org (2022).doi:10.48550/arxiv.2207.12598. 43

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2207.12598 2022

  28. [28]

    Deep Unsupervised Learning using Nonequilibrium Thermodynamics

    J. Sohl-Dickstein, E. A. Weiss, N. Maheswaranathan, S. Ganguli, Deep Unsupervised Learning using Nonequilibrium Thermodynamics, arXiv:1503.03585 [cond-mat, q-bio, stat] (Nov. 2015).doi:10.48550/ arXiv.1503.03585. URLhttp://arxiv.org/abs/1503.03585

    work page internal anchor Pith review arXiv 2015

  29. [29]

    Y. Song, S. Ermon, Generative Modeling by Estimating Gradients of the Data Distribution, arXiv:1907.05600 [cs, stat] (Oct. 2020). doi: 10.48550/arXiv.1907.05600. URLhttp://arxiv.org/abs/1907.05600

    work page doi:10.48550/arxiv.1907.05600 1907

  30. [30]

    Croitoru, V

    F.-A. Croitoru, V. Hondru, R. T. Ionescu, M. Shah, Diffusion Models in Vision: A Survey, IEEE Transactions on Pattern Analysis and Machine Intelligence 45 (9) (2023) 10850–10869, arXiv:2209.04747 [cs]. doi: 10.1109/TPAMI.2023.3261988. URLhttp://arxiv.org/abs/2209.04747

    work page doi:10.1109/tpami.2023.3261988 2023

  31. [31]

    Q. Xu, J. Liu, M. Wong, C. Chen, Y.-S. Ong, Precise-physics driven text-to-3d generation, arXiv preprint arXiv:2403.12438 (2024)

    work page arXiv 2024

  32. [32]

    Bridson, Fluid Simulation for Computer Graphics, CRC Press, Taylor & Francis Group, CRC Press is an imprint of the Taylor & Francis Group, an informa Business, 2016

    R. Bridson, Fluid Simulation for Computer Graphics, CRC Press, Taylor & Francis Group, CRC Press is an imprint of the Taylor & Francis Group, an informa Business, 2016

    work page 2016

  33. [33]

    Chorin, Numerical solution of the navier–stokes equations, Mathe- matics of Computation 22 (10 1968).doi:10.2307/2004575

    A. Chorin, Numerical solution of the navier–stokes equations, Mathe- matics of Computation 22 (10 1968).doi:10.2307/2004575

    work page doi:10.2307/2004575 1968

  34. [34]

    J. Stam, J. Stam, Stable fluids, International Conference on Computer Graphics and Interactive Techniques (1999). doi:10.1145/311535. 311548

    work page doi:10.1145/311535 1999

  35. [35]

    Lee, [dataset] pg-3dgs supporting dataset: Multiview teapot and airplane images, depth maps, masks, and camera parameters (2026)

    Z. Lee, [dataset] pg-3dgs supporting dataset: Multiview teapot and airplane images, depth maps, masks, and camera parameters (2026). doi:10.17632/xz7tkg2zhd.1. URLhttps://doi.org/10.17632/xz7tkg2zhd.1

    work page doi:10.17632/xz7tkg2zhd.1 2026

  36. [36]

    Z. Lee, M. Jacobson, Y. Xue, Physics-guided 3d gaussian splatting official implementation [software] (2026).doi:10.5281/zenodo.19392335. URLhttps://doi.org/10.5281/zenodo.19392335 44

    work page doi:10.5281/zenodo.19392335 2026

  37. [37]

    Yariv, J

    L. Yariv, J. Gu, Y. Kasten, Y. Lipman, Volume rendering of neural implicit surfaces, in: Thirty-Fifth Conference on Neural Information Processing Systems, 2021. Appendix A. Efficient tiled evaluation of the Gaussian-induced solid mask This appendix describes an efficient strategy for evaluating the Gaussian- induced solid mask defined in Sec. 5.3.1. The m...

    work page 2021