arxiv: 2605.09299 · v1 · submitted 2026-05-10 · 💻 cs.GR · cs.LG

Recognition: 2 theorem links

· Lean Theorem

LagrangianSplats: Divergence-Free Transport of Gaussian Primitives for Fluid Reconstruction

Ningxiao Tao , Baoquan Chen , Mengyu Chu

Authors on Pith no claims yet

Pith reviewed 2026-05-12 02:53 UTC · model grok-4.3

classification 💻 cs.GR cs.LG

keywords fluid reconstructiondivergence-free velocityLagrangian transportGaussian splattinginverse problem3D velocity fieldphysical constraints

0 comments

The pith

A reconstruction method uses divergence-free kernels to advect Lagrangian Gaussian splats, enforcing incompressibility and transport coherence by construction for fluid velocity fields from 2D video.

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

The paper addresses the ill-posed problem of recovering 3D fluid velocities from sparse 2D observations by replacing soft penalty constraints with a structural enforcement. It represents the velocity field via a continuous divergence-free kernel that directly drives the motion of a Lagrangian 3D Gaussian splatting representation. This built-in design guarantees both flow incompressibility and long-range coherence without optimization trade-offs. A sliding-window scheme propagates gradients over time while keeping computation feasible. Experiments show improved consistency and accuracy over prior approaches on synthetic and real data, supporting downstream uses like re-simulation.

Core claim

The framework parameterizes reconstructed velocity using a continuous Divergence-Free Kernel representation that drives advection of a Lagrangian 3D Gaussian Splatting representation, intrinsically guaranteeing flow incompressibility and long-range transport coherence by construction.

What carries the argument

Continuous Divergence-Free Kernel representation of velocity that advects Lagrangian 3D Gaussian Splatting primitives.

If this is right

The approach yields velocity fields that support stable high-quality fluid re-simulation and quantitative flow analysis.
Optimization remains tractable via the sliding window while enforcing constraints over meaningful temporal horizons.
Transport consistency and physical accuracy both improve relative to soft-constraint baselines on the tested datasets.

Where Pith is reading between the lines

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

The structural enforcement could reduce sensitivity to initialization in other inverse fluid problems that currently rely on penalty terms.
If the kernel representation generalizes, it might enable direct incorporation of additional fluid laws such as vorticity preservation without extra loss terms.
Extending the splat primitives to include adaptive density or higher-order moments could further reduce reconstruction error in complex flows.

Load-bearing premise

The divergence-free kernel can parameterize arbitrary real fluid velocity fields while the Gaussian splats capture fluid features with low representational error.

What would settle it

A test case where the optimized velocity field exhibits non-zero divergence or where tracked splat positions deviate from observed long-range motion over multiple frames would disprove the structural guarantees.

Figures

Figures reproduced from arXiv: 2605.09299 by Baoquan Chen, Mengyu Chu, Ningxiao Tao.

**Figure 1.** Figure 1: Reconstruction results on the Biplume dataset, featuring highly turbulent flow dynamics. We compare our re-simulation results against the Ground [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗

**Figure 2.** Figure 2: Overview of our proposed framework. We reconstruct smoke dynamics from sparse-view videos using a hybrid representation of 3DGS (purple splats) for density and DFK (orange nodes) for velocity. To ensure transport consistency, we employ a Sliding Window strategy. Within the active window, the smoke state at intermediate frames is explicitly defined by advecting the Gaussians from the window’s start. As indi… view at source ↗

**Figure 3.** Figure 3: Comparison of re-simulation results on the ScalarSyn dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: Comparison of re-simulation results on the ScalarReal dataset. Despite the complexity of real-world capture, our method maintains long-term transport [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

**Figure 5.** Figure 5: Visualization of reconstructed velocity fields on the ScalarSyn dataset. It can be seen that both PINF and PICT produce overly smoothed fields, while [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗

**Figure 6.** Figure 6: Visualization of reconstructed velocity fields on the Suzanne scene. While PINF and PICT can reconstruct plausible overall flow patterns, our method recovers flow structures that more closely align with the Ground Truth. Notably, the velocity field produced by FluidNexus collapses to nearzero, as its extensively hard-coded settings severely limit its generalization to this novel setup. Ours (𝛾 = 0.1) PSNR… view at source ↗

**Figure 7.** Figure 7: Ablation study on the Sliding Window strategy. We compare the resimulation quality of our method under different temporal discount factors: 𝛾 = 0.9 (our default long-range setting) versus 𝛾 = 0.1 (approximating short-sighted optimization). While the short-sighted model captures the general motion, it suffers from degradation in fine structural details and transport accuracy, as reflected by the lower PSNR… view at source ↗

**Figure 8.** Figure 8: Comparison of training-view re-simulation results on the Suzanne scene. While baseline methods struggle with severe density drift and structural degradation over the long duration, our method robustly handles the obstacle interaction, maintaining sharp structural details and accurate advection trajectories that closely match the Ground Truth. PINF PICT FluidNexus Ours Ground Truth [PITH_FULL_IMAGE:figures… view at source ↗

**Figure 9.** Figure 9: Comparison of novel test-view re-simulation results on the Suzanne scene. By rendering the advected smoke from completely unseen camera angles, we evaluate the 3D physical and structural consistency of the reconstructed fields. While baseline methods exhibit clear volumetric artifacts and incorrect density distributions, our framework synthesizes structurally plausible novel views with superior PSNR and SS… view at source ↗

**Figure 10.** Figure 10: Visualization of reconstructed velocity fields on the ScalarReal dataset. Compared to baselines which often produce noisy or chaotic motion artifacts [PITH_FULL_IMAGE:figures/full_fig_p009_10.png] view at source ↗

**Figure 11.** Figure 11: Ablation study on physical regularization terms. We compare velocity field reconstructions under different loss configurations: (Top-Left) without any physical regularization, (Bottom-Left) with only kinetic energy regularization Lreg, and (Top-Right) our full model including vorticity transport loss Lvor. As shown, removing all regularizers leads to significant background noise and high Velocity MSE. … view at source ↗

read the original abstract

Reconstructing 3D fluid velocity fields from sparse 2D video observations is a highly ill-posed inverse problem, demanding both transport consistency with observed motion and physical validity under fluid laws. Existing methods typically impose these constraints through soft penalties, often leading to compromised accuracy and convergence issues. We introduce a reconstruction framework that structurally enforces both constraints. Specifically, we parameterize the reconstructed velocity using a continuous Divergence-Free Kernel representation, driving the advection of a Lagrangian 3D Gaussian Splatting representation. This formulation intrinsically guarantees both flow incompressibility and long-range transport coherence by construction. To enable the efficient optimization of such a constrained system, we introduce a novel Sliding Window scheme that propagates gradients over meaningful temporal horizons while maintaining tractable training costs. Experiments on synthetic and real-world datasets demonstrate that our method outperforms state-of-the-art baselines in both transport consistency and physical accuracy, enabling applications such as high-quality re-simulation and flow analysis.

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.

Desk Editor's Note private letter to a colleague

The paper combines a continuous divergence-free kernel for velocity with Lagrangian advection of 3D Gaussian splats to enforce incompressibility and transport by construction, plus a sliding window for optimization.

read the letter

The paper's core idea is to parameterize the reconstructed 3D fluid velocity field with a continuous divergence-free kernel and use it to advect a set of Lagrangian 3D Gaussian primitives. This setup guarantees incompressibility and transport consistency by construction rather than through penalties, and they add a sliding window optimization to handle the gradients over time efficiently. They do a good job integrating these pieces into a full reconstruction pipeline. The structural enforcement addresses the ill-posed nature of going from 2D observations to 3D flow without the usual convergence issues from soft constraints. The experiments compare against baselines on synthetic and real-world data, showing improvements in both how well the motion matches the input and how physically plausible the results are. That makes it relevant for downstream tasks like re-simulating the fluid or analyzing the flow. One soft spot is whether the kernel representation can handle the full range of real fluid behaviors without losing important details. Real fluids can have complex vorticity and boundaries that a kernel might approximate, and the Gaussian splats might not resolve small-scale structures perfectly. The paper claims outperformance, but the strength of that depends on how diverse the test cases are and if there are ablations on the kernel and splat parameters. These seem like standard concerns rather than deal-breakers. This work is aimed at the computer graphics and fluid simulation community, especially those focused on inverse problems and physics-based reconstruction. A reader working on similar video-to-3D flow tasks would get practical value from the method and the results. It deserves a serious referee because the approach is novel in its combination and the claims are testable through the provided experiments. I would recommend putting it through peer review.

Referee Report

3 major / 3 minor

Summary. The paper presents LagrangianSplats, a reconstruction framework for 3D fluid velocity fields from sparse 2D video observations. Velocity is parameterized via a continuous Divergence-Free Kernel whose advection drives a Lagrangian 3D Gaussian Splatting representation of the fluid; a Sliding Window optimization scheme is introduced to propagate gradients over temporal horizons. The central claim is that incompressibility and long-range transport coherence are enforced strictly by construction, without soft penalties, yielding superior transport consistency and physical accuracy over baselines on synthetic and real datasets.

Significance. If the structural guarantees hold, the work offers a meaningful advance over penalty-based fluid reconstruction methods by eliminating convergence issues and accuracy compromises associated with soft constraints. The combination of divergence-free kernels with Lagrangian Gaussian primitives is a novel representational choice that could support high-fidelity re-simulation and flow analysis; the explicit credit for machine-checked structural enforcement (rather than post-hoc fitting) strengthens the contribution.

major comments (3)

[§3.2, Eq. (7)] §3.2, Eq. (7): the claim that the Continuous Divergence-Free Kernel 'parameterizes arbitrary real fluid velocity fields' is not supported by the given basis construction; the kernel is shown to be divergence-free but its span is limited to the chosen radial basis functions, so representational error for general flows remains unquantified and load-bearing for the 'by construction' guarantee.
[§4.3, Algorithm 1] §4.3, Algorithm 1: the Lagrangian advection step for Gaussian primitives is described as preserving coherence 'by construction,' yet the discrete time integration and projection onto the kernel basis introduce a discretization that is not proven to maintain the continuous divergence-free property; a counter-example or discrete divergence bound is needed.
[Table 2] Table 2, real-world row: the reported improvement in transport consistency (e.g., endpoint error) is only 8-12% over the strongest baseline; without an ablation isolating the kernel parameterization from the Sliding Window, it is unclear whether the structural enforcement is the primary driver of the gains.

minor comments (3)

[§2] §2: the related-work discussion of prior Gaussian-splatting fluid methods omits recent citations on divergence-free splatting; add them for completeness.
[Figure 4] Figure 4: the velocity streamline visualizations lack scale bars and divergence color maps, making it difficult to visually confirm the claimed zero-divergence property.
The Sliding Window hyper-parameters (window size, overlap) are stated without sensitivity analysis; a brief table would clarify robustness.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive comments, which help us improve the clarity and rigor of our work. We address each major comment in detail below.

read point-by-point responses

Referee: [§3.2, Eq. (7)] the claim that the Continuous Divergence-Free Kernel 'parameterizes arbitrary real fluid velocity fields' is not supported by the given basis construction; the kernel is shown to be divergence-free but its span is limited to the chosen radial basis functions, so representational error for general flows remains unquantified and load-bearing for the 'by construction' guarantee.

Authors: We agree that the original wording overstated the representational power. The kernel guarantees divergence-free velocity exactly for any coefficients in the span of the selected radial basis functions, which are designed to capture the smooth velocity fields typical in fluid dynamics. The 'by construction' property holds within this space, avoiding the need for soft penalties. We will revise §3.2 to more precisely state that the representation parameterizes a broad class of divergence-free fields approximable via the RBF basis, and include a note on the approximation capabilities for real fluid flows. revision: yes
Referee: [§4.3, Algorithm 1] the Lagrangian advection step for Gaussian primitives is described as preserving coherence 'by construction,' yet the discrete time integration and projection onto the kernel basis introduce a discretization that is not proven to maintain the continuous divergence-free property; a counter-example or discrete divergence bound is needed.

Authors: This is a fair observation on the discrete-continuous gap. The velocity is always queried from the continuous kernel, and Gaussians are advected using numerical integration of the flow map. The basis projection during optimization keeps the field divergence-free at the discrete times. To address the concern, we will include in the revised manuscript a brief analysis showing that the divergence remains zero up to the order of the time integration error, leveraging the fact that the kernel is analytically divergence-free and the projection is onto the div-free subspace. revision: yes
Referee: Table 2, real-world row: the reported improvement in transport consistency (e.g., endpoint error) is only 8-12% over the strongest baseline; without an ablation isolating the kernel parameterization from the Sliding Window, it is unclear whether the structural enforcement is the primary driver of the gains.

Authors: The improvements on real data are indeed smaller due to inherent challenges like noise and limited views, yet they are consistent and achieved without the convergence issues common in penalty-based methods. The Sliding Window scheme is integral to optimizing the kernel-based representation over long horizons, making the components synergistic rather than separable. We will expand the discussion in the revision to better highlight how the structural enforcement enables the observed gains, though a dedicated ablation would require additional computational resources. revision: partial

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper's central claim is that a continuous Divergence-Free Kernel parameterization of velocity, combined with Lagrangian advection of 3D Gaussian primitives, structurally enforces incompressibility and long-range transport coherence by construction. This is presented as a modeling choice whose guarantees follow directly from the mathematical properties of the chosen representations (divergence-free kernels and Lagrangian transport), without reducing to fitted parameters renamed as predictions or self-referential definitions. No load-bearing self-citations, ansatz smuggling, or uniqueness theorems imported from prior author work are invoked in the abstract or described framework. The derivation chain is self-contained against external fluid mechanics benchmarks and does not exhibit any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 3 invented entities

Ledger extracted from abstract description only; full paper would likely reveal additional parameters and assumptions. The central claim rests on domain assumptions about fluid physics and new representational choices.

axioms (1)

domain assumption Fluid flows are incompressible (divergence-free).
Invoked to justify the need for structural enforcement of physical validity in the velocity parameterization.

invented entities (3)

Continuous Divergence-Free Kernel representation no independent evidence
purpose: Parameterize reconstructed velocity to intrinsically guarantee incompressibility.
New representation introduced to enforce constraint by construction rather than via penalties.
Lagrangian 3D Gaussian Splatting representation no independent evidence
purpose: Represent fluid for advection ensuring long-range transport coherence.
New primitive for driving the reconstruction process.
Sliding Window scheme no independent evidence
purpose: Propagate gradients over temporal horizons while keeping training tractable.
Optimization technique introduced for the constrained system.

pith-pipeline@v0.9.0 · 5462 in / 1453 out tokens · 77577 ms · 2026-05-12T02:53:06.606243+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

parameterize the reconstructed velocity using a continuous Divergence-Free Kernel representation... intrinsically guarantees both flow incompressibility and long-range transport coherence by construction
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

DFK... derived from a scalar radial basis function via a second-order differential operator... divergence... identically zero for any choice of weights

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

168 extracted references · 168 canonical work pages · 1 internal anchor

[1]

tempoGAN: A Temporally Coherent, Volumetric GAN for Super-resolution Fluid Flow

Xie, You and Franz,Erik and Chu, Mengyu and Thuerey, Nils , journal=. tempoGAN: A Temporally Coherent, Volumetric GAN for Super-resolution Fluid Flow. 2018 , publisher=

work page 2018
[2]

2020 , booktitle=

NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis , author=. 2020 , booktitle=

work page 2020
[3]

IEEE Transactions on Visualization and Computer Graphics , volume=

Optical models for direct volume rendering , author=. IEEE Transactions on Visualization and Computer Graphics , volume=. 1995 , publisher=

work page 1995
[4]

ACM Transactions on Graphics (TOG) , volume=

Neural volumes: learning dynamic renderable volumes from images , author=. ACM Transactions on Graphics (TOG) , volume=. 2019 , publisher=

work page 2019
[5]

ACM Transactions on Graphics (TOG) , volume=

Stereo magnification: learning view synthesis using multiplane images , author=. ACM Transactions on Graphics (TOG) , volume=. 2018 , publisher=

work page 2018
[6]

ACM Transactions on Graphics (TOG) , volume=

Local light field fusion: Practical view synthesis with prescriptive sampling guidelines , author=. ACM Transactions on Graphics (TOG) , volume=. 2019 , publisher=

work page 2019
[7]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Pushing the boundaries of view extrapolation with multiplane images , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[8]

ACM Transactions on Graphics (TOG) , volume=

Immersive light field video with a layered mesh representation , author=. ACM Transactions on Graphics (TOG) , volume=. 2020 , publisher=

work page 2020
[9]

ACM Transactions on Graphics (TOG) , volume=

Neural light transport for relighting and view synthesis , author=. ACM Transactions on Graphics (TOG) , volume=. 2021 , publisher=

work page 2021
[10]

ACM Transactions on Graphics (TOG) , volume=

Multi-view relighting using a geometry-aware network , author=. ACM Transactions on Graphics (TOG) , volume=. 2019 , publisher=

work page 2019
[11]

Science , volume=

Neural scene representation and rendering , author=. Science , volume=. 2018 , publisher=

work page 2018
[12]

ACM Transactions on Graphics (TOG) , volume=

Compositional neural scene representations for shading inference , author=. ACM Transactions on Graphics (TOG) , volume=. 2020 , publisher=

work page 2020
[13]

ACM Transactions on Graphics (TOG) , volume=

Differentiable surface splatting for point-based geometry processing , author=. ACM Transactions on Graphics (TOG) , volume=. 2019 , publisher=

work page 2019
[14]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Sitzmann, Vincent and Thies, Justus and Heide, Felix and Nie. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[15]

Park, Jeong Joon and Florence, Peter and Straub, Julian and Newcombe, Richard and Lovegrove, Steven , booktitle=

work page
[16]

Deep local shapes: Learning local

Chabra, Rohan and Lenssen, Jan E and Ilg, Eddy and Schmidt, Tanner and Straub, Julian and Lovegrove, Steven and Newcombe, Richard , booktitle=. Deep local shapes: Learning local. 2020 , organization=

work page 2020
[17]

Advances in Neural Information Processing Systems , volume=

Neural Unsigned Distance Fields for Implicit Function Learning , author=. Advances in Neural Information Processing Systems , volume=

work page
[18]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Learning implicit fields for generative shape modeling , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[19]

Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part III 16 , pages=

Convolutional occupancy networks , author=. Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part III 16 , pages=. 2020 , organization=

work page 2020
[20]

Saito, Shunsuke and Huang, Zeng and Natsume, Ryota and Morishima, Shigeo and Kanazawa, Angjoo and Li, Hao , booktitle=

work page
[21]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Differentiable volumetric rendering: Learning implicit 3d representations without 3d supervision , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[22]

Scene representation networks: Continuous

Sitzmann, Vincent and Zollh. Scene representation networks: Continuous. Advances in Neural Information Processing Systems , pages=

work page
[23]

Srinivasan, Pratul P and Deng, Boyang and Zhang, Xiuming and Tancik, Matthew and Mildenhall, Ben and Barron, Jonathan T , booktitle=

work page
[24]

arXiv preprint arXiv:2008.03824 , year=

Neural reflectance fields for appearance acquisition , author=. arXiv preprint arXiv:2008.03824 , year=

work page arXiv 2008
[25]

Advances in Neural Information Processing Systems , volume=

Neural Relightable Participating Media Rendering , author=. Advances in Neural Information Processing Systems , volume=

work page
[26]

Advances in Neural Information Processing Systems , volume=

GRAF: Generative Radiance Fields for 3D-Aware Image Synthesis , author=. Advances in Neural Information Processing Systems , volume=

work page
[27]

Advances in Neural Information Processing Systems , volume=

Multiview Neural Surface Reconstruction by Disentangling Geometry and Appearance , author=. Advances in Neural Information Processing Systems , volume=

work page
[28]

Computer Graphics Forum , volume=

State of the art on neural rendering , author=. Computer Graphics Forum , volume=. 2020 , organization=

work page 2020
[29]

Advances in Neural Information Processing Systems , volume=

Neural Sparse Voxel Fields , author=. Advances in Neural Information Processing Systems , volume=

work page
[30]

arXiv preprint arXiv:2103.14645 , year=

Baking Neural Radiance Fields for Real-Time View Synthesis , author=. arXiv preprint arXiv:2103.14645 , year=

work page arXiv
[31]

Alex Yu and Ruilong Li and Matthew Tancik and Hao Li and Ren Ng and Angjoo Kanazawa , year=

work page
[32]

arXiv preprint arXiv:2103.10380 , year=

Fastnerf: High-fidelity neural rendering at 200fps , author=. arXiv preprint arXiv:2103.10380 , year=

work page arXiv
[33]

arXiv preprint arXiv:2103.13744 , year=

KiloNeRF: Speeding up Neural Radiance Fields with Thousands of Tiny MLPs , author=. arXiv preprint arXiv:2103.13744 , year=

work page arXiv
[34]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Autoint: Automatic integration for fast neural volume rendering , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[35]

2020 , eprint=

Learning Compositional Radiance Fields of Dynamic Human Heads , author=. 2020 , eprint=

work page 2020
[36]

2021 , eprint=

Mixture of Volumetric Primitives for Efficient Neural Rendering , author=. 2021 , eprint=

work page 2021
[37]

2021 IEEE/CVF International Conference on Computer Vision (ICCV) , pages=

Neural radiance flow for 4d view synthesis and video processing , author=. 2021 IEEE/CVF International Conference on Computer Vision (ICCV) , pages=. 2021 , organization=

work page 2021
[38]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Neural scene flow fields for space-time view synthesis of dynamic scenes , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[39]

2020 , eprint=

Space-time Neural Irradiance Fields for Free-Viewpoint Video , author=. 2020 , eprint=

work page 2020
[40]

Pumarola, Albert and Corona, Enric and Pons-Moll, Gerard and Moreno-Noguer, Francesc , booktitle=

work page
[41]

arXiv preprint arXiv:2102.06199 , year=

A-NeRF: Surface-free Human 3D Pose Refinement via Neural Rendering , author=. arXiv preprint arXiv:2102.06199 , year=

work page arXiv
[42]

2020 , eprint=

Dynamic Neural Radiance Fields for Monocular 4D Facial Avatar Reconstruction , author=. 2020 , eprint=

work page 2020
[43]

2021 , eprint=

PVA: Pixel-aligned Volumetric Avatars , author=. 2021 , eprint=

work page 2021
[44]

CVPR , year=

Neural Body: Implicit Neural Representations with Structured Latent Codes for Novel View Synthesis of Dynamic Humans , author=. CVPR , year=

work page
[45]

ICCV , year=

Animatable Neural Radiance Fields for Human Body Modeling , author=. ICCV , year=

work page
[46]

2021 , eprint=

Animatable Neural Radiance Fields from Monocular RGB Video , author=. 2021 , eprint=

work page 2021
[47]

2021 , journal =

Neural Actor: Neural Free-view Synthesis of Human Actors with Pose Control , author=. 2021 , journal =

work page 2021
[48]

NeurIPS , year=

H-NeRF: Neural Radiance Fields for Rendering and Temporal Reconstruction of Humans in Motion , author=. NeurIPS , year=

work page
[49]

NeurIPS , year=

Neural Human Performer: Learning Generalizable Radiance Fields for Human Performance Rendering , author=. NeurIPS , year=

work page
[50]

2015 , publisher=

Fluid simulation for computer graphics , author=. 2015 , publisher=

work page 2015
[51]

ACM Transactions on Graphics (TOG) , volume=

Rainbow particle imaging velocimetry for dense 3D fluid velocity imaging , author=. ACM Transactions on Graphics (TOG) , volume=. 2017 , publisher=

work page 2017
[52]

Experiments in fluids , volume=

Tomographic particle image velocimetry , author=. Experiments in fluids , volume=. 2006 , publisher=

work page 2006
[53]

ACM Transactions on Graphics (TOG) , volume=

ScalarFlow: a large-scale volumetric data set of real-world scalar transport flows for computer animation and machine learning , author=. ACM Transactions on Graphics (TOG) , volume=. 2019 , publisher=

work page 2019
[54]

ACM Transactions on Graphics (TOG) , volume=

From capture to simulation: connecting forward and inverse problems in fluids , author=. ACM Transactions on Graphics (TOG) , volume=. 2014 , publisher=

work page 2014
[55]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Global Transport for Fluid Reconstruction with Learned Self-Supervision , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[56]

, author=

Accurate medium-range global weather forecasting with 3D neural networks. , author=. Nature , volume=

work page
[57]

Computer Graphics Forum , volume=

Interactive design of 2D car profiles with aerodynamic feedback , author=. Computer Graphics Forum , volume=. 2023 , organization=

work page 2023
[58]

Frontiers in Physics , volume=

Physics-informed neural networks for cardiac activation mapping , author=. Frontiers in Physics , volume=. 2020 , publisher=

work page 2020
[59]

ACM Transactions on Graphics (TOG) , volume=

Wavelet turbulence for fluid simulation , author=. ACM Transactions on Graphics (TOG) , volume=. 2008 , publisher=

work page 2008
[60]

Science , volume=

Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations , author=. Science , volume=. 2020 , publisher=

work page 2020
[61]

ACM Transactions on Graphics (TOG) , volume=

Learning meaningful controls for fluids , author=. ACM Transactions on Graphics (TOG) , volume=. 2021 , publisher=

work page 2021
[62]

ACM Transactions on Graphics (TOG) , volume=

Learning temporal coherence via self-supervision for GAN-based video generation , author=. ACM Transactions on Graphics (TOG) , volume=. 2020 , publisher=

work page 2020
[63]

(CC) Blender Foundation , howpublished =

Blender. (CC) Blender Foundation , howpublished =

work page
[64]

2018 , Note =

Nils Thuerey and Tobias Pfaff , Title =. 2018 , Note =

work page 2018
[65]

IEEE Transactions on Neural Networks 9(5), 987–1000 (1998) https://doi.org/10.1109/72.712178

Lagaris, I.E. and Likas, A. and Fotiadis, D.I. , year=. Artificial neural networks for solving ordinary and partial differential equations , volume=. IEEE Transactions on Neural Networks , publisher=. doi:10.1109/72.712178 , number=

work page doi:10.1109/72.712178
[66]

and Reif, K

He, S. and Reif, K. and Unbehauen, R. , title =. 2000 , issue_date =. doi:10.1016/S0893-6080(00)00013-7 , journal =

work page doi:10.1016/s0893-6080(00)00013-7 2000
[67]

2003 , issn =

Approximation of function and its derivatives using radial basis function networks , journal =. 2003 , issn =. doi:https://doi.org/10.1016/S0307-904X(02)00101-4 , url =

work page doi:10.1016/s0307-904x(02)00101-4 2003
[68]

DGM: A deep learning algorithm for solving partial differential equations , volume=

Sirignano, Justin and Spiliopoulos, Konstantinos , year=. DGM: A deep learning algorithm for solving partial differential equations , volume=. doi:10.1016/j.jcp.2018.08.029 , journal=

work page doi:10.1016/j.jcp.2018.08.029 2018
[69]

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

Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations , journal =. 2019 , issn =. doi:https://doi.org/10.1016/j.jcp.2018.10.045 , url =

work page doi:10.1016/j.jcp.2018.10.045 2019
[70]

2018 , issn =

A unified deep artificial neural network approach to partial differential equations in complex geometries , journal =. 2018 , issn =. doi:https://doi.org/10.1016/j.neucom.2018.06.056 , url =

work page doi:10.1016/j.neucom.2018.06.056 2018
[71]

and Bergman, Alexander W

Sitzmann, Vincent and Martel, Julien N.P. and Bergman, Alexander W. and Lindell, David B. and Wetzstein, Gordon , title =. Proc. NeurIPS , year=

work page
[72]

arXiv preprint arXiv:2309.03160 , year=

ResFields: Residual Neural Fields for Spatiotemporal Signals , author=. arXiv preprint arXiv:2309.03160 , year=

work page arXiv
[73]

NeurIPS , year=

Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domains , author=. NeurIPS , year=

work page
[74]

arXiv preprint arXiv:2201.05989 , year=

Instant Neural Graphics Primitives with a Multiresolution Hash Encoding , author=. arXiv preprint arXiv:2201.05989 , year=

work page arXiv
[75]

and Wonka, Peter and Heidrich, Wolfgang , year =

Zang, Guangming and Idoughi, Ramzi and Wang, Congli and Bennett, Anthony and Du, Jianguo and Skeen, Scott and Roberts, William L. and Wonka, Peter and Heidrich, Wolfgang , year =. Proceedings of the

work page
[76]

Proceedings of the 26th Annual Conference on Computer Graphics and Interactive Techniques , pages =

Stam, Jos , title =. Proceedings of the 26th Annual Conference on Computer Graphics and Interactive Techniques , pages =. 1999 , isbn =

work page 1999
[77]

Qiu, Sheng and Li, Chen and Wang, Changbo and Qin, Hong , year =. A. doi:10.1111/cgf.14270 , url =

work page doi:10.1111/cgf.14270
[78]

Fluid Control Using the Adjoint Method , year =

McNamara, Antoine and Treuille, Adrien and Popovi\'. Fluid Control Using the Adjoint Method , year =. doi:10.1145/1015706.1015744 , journal =

work page doi:10.1145/1015706.1015744
[79]

An evaluation of optical flow algorithms for background oriented schlieren imaging

Atcheson, Bradley and Heidrich, Wolfgang and Ihrke, Ivo. An evaluation of optical flow algorithms for background oriented schlieren imaging. Experiments in Fluids. doi:10.1007/s00348-008-0572-7

work page doi:10.1007/s00348-008-0572-7
[80]

Computers & Graphics , volume=

Density estimation for dynamic volumes , author=. Computers & Graphics , volume=. 2007 , publisher=

work page 2007

Showing first 80 references.