Pith. sign in

REVIEW 2 major objections 6 minor 78 references

A Query-Efficient Stochastic Volume Rendering Framework for Time-Varying Implicit Neural Volumes

T0 review · 2 major / 6 minor · reviewed 2026-07-31 · grok-4.5

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

Pith's one-line read Time-varying neural volumes can be volume-rendered directly at interactive rates without resampling, retraining, or caches.

desk verdict Solid TVCG-style systems paper: direct interactive rendering of unmodified time-varying scientific INRs at real FPS, with honest limits on approximate majorants. read the letter →

arxiv 2607.28047 v1 pith:DMJBKWV6 submitted 2026-07-30 cs.GR cs.LG

Alper Sahistan , Haichao Miao , Zhimin Li , Peer-Timo Bremer , Joshua A Levine , Valerio Pascucci This is my paper
classification cs.GRcs.LG
keywords volumerenderingraytracingimplicitneuralrepresentationstimeseriesdatascientificvisualizationdeltatrackingqueryreduction
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

Scientific volumes stored as time-varying implicit neural networks replace cheap memory lookups with expensive network evaluations, so ordinary ray marching is too slow and intermediate grids or caches break under continuous time navigation. This paper shows that delta tracking—stochastic free-flight sampling under local density upper bounds—can be turned into a query-reduction engine for those networks. A four-stage pipeline separates ray traversal on ray-tracing cores from batched network evaluation on tensor cores, then further cuts queries with adaptive ray budgets and pruning of near-uniform cells. The result is that many original time-varying networks render at roughly 30–40 frames per second at 1024² with shadows and converge to high-fidelity images, while switching the continuous time parameter costs only a few milliseconds. A sympathetic reader cares because dynamic CT and similar modalities often exist only as these networks; the work makes interactive exploration of the continuous time domain practical without destroying the representation.

What carries the argument

A four-stage wavefront pipeline that decouples RT-core macrocell traversal and delta tracking from tensor-core batched INR evaluation, augmented by adaptive ray budgeting and homogeneity-based query pruning that together minimize and amortize neural inferences while preserving estimator correctness under practical approximations.

What would settle it

Render the same original network with a single loose global bound versus the paper’s lattice macrocells: if box artifacts persist after ghost-pass refinement, or if converged images diverge from a dense reference in PSNR/FLIP while frame rate collapses below interactive levels on typical dynamic-CT or CFD networks, the claim fails.

Watch

Extended reading notes

Core claim

Many time-varying implicit neural volumes can be rendered directly from their original networks at interactive rates by treating delta tracking as a neural-query reducer inside a four-stage heterogeneous pipeline, reaching about 30–40 FPS at 1024² with ray-traced shadows and roughly 1–2 ms timestep updates, without requiring resampling, retraining, or caching as a prerequisite.

Load-bearing premise

Finite lattice samples and linear time interpolation are good enough to set local density upper bounds so that the stochastic tracker does not systematically skip real structure or leave permanent box artifacts.

Editorial extensions

If this is right

  • Dynamic X-ray CT and other deforming-object volumes that exist only as continuous neural fields become interactively explorable in continuous time.
  • Scientists can keep the original compact network as the working representation instead of baking grids or surrogate caches before visualization.
  • Timestep changes cost a few milliseconds, so scrubbing and animation of continuous time are practical on a single high-end GPU.
  • Secondary effects such as ray-traced shadows remain affordable because the same query-reduction path applies to shadow rays.
  • Conservative homogeneity pruning and blue-noise ray budgets further raise frame rate with bounded early-frame noise and limited bias.

Reading between the lines

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

  • The same decouple-and-batch pattern should transfer to other expensive continuous fields (for example physics-informed networks) whenever evaluation cost dwarfs traversal.
  • If training exposed certified local extrema or interval bounds, the lattice-and-ghost-pass stage could shrink or disappear, tightening majorants and raising frame rate further.
  • Interactive continuous-time scrubbing may change analysis workflows that today jump only between discrete reconstructed volumes.
  • When temporal coherence is high, a lightweight cache layered on this pipeline could compound speedups; when users scrub time aggressively, the cache-free path remains the fallback.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 6 minor

Summary. The paper presents a query-efficient stochastic volume rendering system for time-varying implicit neural representations (INRs). It reinterprets delta tracking as a neural-query reduction mechanism and implements a four-stage wavefront pipeline that decouples RT-core macrocell traversal from batched tensor-core INR evaluation. Two complementary query-reduction strategies—adaptive ray budgeting (residual histogram, STBN, and residual-weighted STBN) and homogeneity-based query pruning—are added on top. Across six INRs (FFN dynamic-CT, SIREN CFD, CoordNet), the system reports ~30–40 FPS at 1024² on an RTX 4090 with ray-traced shadows, ~1–2 ms timestep updates via interpolated working macrocell grids, and progressive convergence to high-fidelity images, without requiring resampling, retraining, or caching as a prerequisite.

Significance. Interactive visualization of unmodified time-varying INRs is a genuine bottleneck in scientific visualization, especially for dynamic CT where the INR is often the only practical representation. The work makes a solid systems contribution: the traverse–evaluate wavefront design, RT-core sparse GAS traversal, and tensor-core batching are well motivated and supported by controlled ablations (ray marcher vs naïve delta tracker vs wavefront DDA vs full RT pipeline in Fig. 7; memory/update costs in Table 1 and Fig. 6; budgeting and pruning trade-offs in Figs. 8–9). Code availability and honest limitation discussion (macrocell underestimation, vendor coupling) strengthen the contribution. If the reported rates and update latencies hold under broader use, the framework meaningfully expands the practical utility of continuous neural volumes in SciVis.

major comments (2)
  1. [§4.1.1, §6] §4.1.1 and §6 (Figs. 4, 11): The central performance claim is defensible, but estimator correctness under approximate majorants needs a sharper statement. Lattice sampling plus temporal lerp can underestimate extrema; underestimation is not merely a quality issue—it can cause rays to skip dense structure (invalid majorant), whereas overestimation only increases null collisions. Ghost passes mitigate but do not guarantee coverage. Please state explicitly under which operating modes (no pruning; majorants verified/loose; ghost-pass budget) the estimator remains unbiased versus when structured bias/artifacts are accepted, and quantify residual underestimation rate or FLIP after a fixed ghost-pass budget on the high-temporal-variation SIREN cases.
  2. [§5.3, Fig. 9] §5.3 / Fig. 9: Homogeneity pruning is presented as a cost-effective extension, but the critical threshold ε_c is dataset-dependent (0.12 vs 0.16) and quality collapses once pruned cells overlap heterogeneous structure. For the claim that pruning “improves efficiency with limited impact on image quality” when used conservatively, add guidance or an automatic cap (e.g., relative to transfer-function Lipschitz scale or fraction of non-empty cells) so users can stay below ε_c without per-dataset search. Otherwise the practical takeaway remains brittle.
minor comments (6)
  1. [Fig. 7] Fig. 7: Ray marcher and naïve delta tracker are near-flat at <1 FPS; consider a log-scale inset or a table of absolute sample counts / INR queries per frame so the ~125× and ~2× gains are easier to audit beyond FPS.
  2. [§4.2.1] §4.2.1: Residual reprojection after camera motion is described but not measured. A short PSNR/FPS trace for a tumble or orbit sequence would show whether residual-driven schemes retain their early-frame advantage under interaction.
  3. [Table 1] §5.1 / Table 1: GAS sizes are marked with ∗ (average over timesteps). Report min–max or a small bar to show sensitivity to TF and t, since empty-cell culling is part of the memory story.
  4. [§4.1.4] Listing 5 and §4.1.4: Clarify numeric format end-to-end (weights FP16; activations; transfer-function opacity precision) and whether any observed quality gap vs a pure FP32 evaluator was measured.
  5. [§2] Related work: briefly position against other SciVis INR renderers and cache systems beyond Zavorotny et al. [69] (e.g., any concurrent interactive INR volume work) so the “direct, unmodified INR” claim is scoped against the closest alternatives.
  6. Minor typos/notation: “sciVis” vs “SciVis” inconsistency; Fig. 1 bar chart labels are hard to read in the author version; define FLIP on first use in the main text (currently mostly in captions).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: empirical systems paper with measured FPS/PSNR against external baselines

full rationale

This is a GPU systems/visualization paper whose central claims are measured end-to-end performance (~30–40 FPS at 1024², 1–2 ms timestep updates) and image quality (PSNR/FLIP vs long-accumulation ground truth), not first-principles predictions. Delta tracking is the classical Woodcock estimator (Eqs. 1–4, Listings 1–3), applied as a known query-reduction mechanism; the four-stage wavefront pipeline, RT-core traversal, and tensor-core batching are engineering contributions evaluated against ray marching, naïve delta tracking, and DDA wavefront baselines (Fig. 7). Free parameters (ρ, ε, macrocell resolution, α) are user knobs that trade speed for quality; they do not force the reported FPS or convergence numbers by construction. Self-citations to prior multi-density Woodcock work are background, not load-bearing uniqueness theorems. Approximate macrocell extrema (lattice sampling, temporal lerp, ghost passes) are acknowledged limitations that can bias majorants, but that is a correctness/approximation issue, not circular derivation. No step reduces a claimed prediction to its own fitted inputs.

Assumptions & free parameters 5 free parameters · 5 assumptions · 2 invented entities

Load-bearing content is mostly standard Monte Carlo volume rendering plus engineering approximations needed because INRs lack analytic cell extrema. Performance claims rest on hardware behavior and user-chosen thresholds, not on new physical postulates.

free parameters (5)
  • homogeneity threshold ε (and falloff δ) = δ≈0.006; ε user-set, critical ~0.12–0.16
    Controls when INR queries are replaced by a per-cell expected scalar; quality drops sharply past dataset-dependent ε_c (~0.12–0.16). δ=0.006 chosen experimentally.
  • ray budget fraction ρ and residual steer α = examples ρ∈{0.25,0.5,0.75}, α∈{0.5,0.75}
    Decide which pixels cast rays each frame in STBN / residual-weighted schemes; trade FPS vs early-frame error.
  • macrocell spatial/temporal resolution (N³×T) = 64³×20 (FFN), 64³×30 (SIREN) in main experiments
    Sets majorant tightness, memory, GAS rebuild cost, and null-collision rate; authors recommend ~64³×20–30 as operating point.
  • EMA rate / center weight for pruned-cell expected scalar = α=0.1 max(w_c,0.01)
    α=0.1 max(w_c,0.01) updates ŝ_c from real evaluations; hand-chosen smoothing.
  • lattice sampling density for macrocell extrema = 9³ lattice / 4³-cell tiles
    9³ lattice per 4³ tile used to approximate continuous INR bounds; density is a design choice affecting underestimation risk.
assumptions (5)
  • standard math Delta tracking with a valid majorant σ̄≥σ_t(x) is an unbiased estimator of the emission-absorption volume rendering integral.
    Invoked throughout §3 and as motivation in §4; classical Woodcock/delta tracking.
  • domain assumption Local majorants derived from transfer-function extrema over approximate per-macrocell scalar min/max remain usable for free-flight sampling on continuous INRs.
    §3–§4.1.1; exact extrema are intractable, so lattice + temporal lerp + ghost passes stand in.
  • domain assumption Batched WMMA evaluation of the given MLP architectures on tensor cores is numerically adequate (including 16-bit weights) for visualization-quality scalars.
    §4.1.4 fused eval kernel; correctness of images assumes half-precision inference matches the intended field closely enough.
  • ad hoc to paper Per-pixel residual |C_n−C̄_{n−1}| is a useful proxy for where additional rays reduce perceived/error most under progressive accumulation.
    §4.2.1 budgeting schemes; heuristic, not derived from a perception theorem.
  • ad hoc to paper Near-uniform macrocells (Δs<ε) may substitute a constant expected scalar for pointwise INR queries with acceptable bias for SciVis transfer functions when ε is conservative.
    §4.2.2; authors acknowledge bias and block artifacts if ε is aggressive.
invented entities (2)
  • Ghost passes (hidden cosine-TF sweeps to refine macrocell bounds)
    purpose: Force visitation of macrocells that loose majorants would otherwise skip, so underestimated bounds can be corrected at runtime.
    Introduced in §4.1.1 / Fig. 4 as a paper-specific mitigation for lattice/temporal underestimation; not a physical entity, but an algorithmic construct load-bearing for artifact control.
  • Four-stage traverse–evaluate wavefront state (TraversalState + INR query buffers) independent evidence
    purpose: Decouple serial-per-ray delta tracking from batched tensor-core inference while retaining RT-core macrocell traversal.
    Core systems abstraction in §4.1; engineering structure rather than a new scientific object.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Query-Efficient Stochastic Volume Rendering Framework for Time-Varying Implicit Neural Volumes." pith.science (2026). https://pith.science/paper/DMJBKWV6

@misc{pith2026260728047,
  author       = {Pith},
  title        = {Pith review of: A Query-Efficient Stochastic Volume Rendering Framework for Time-Varying Implicit Neural Volumes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DMJBKWV6}},
  note         = {Machine review of arXiv:2607.28047}
}
read the original abstract

Time-varying implicit neural representations (INRs) provide a compact representation of scientific volumes and, for modalities such as dynamic X-ray computed tomography (CT), are often the only practical way to represent the data. However, interactive volume rendering of INRs is challenging, as cheap memory lookups are replaced by expensive neural inferences, hindering the performance. Therefore, conventional volume rendering methods such as ray marching with dense sampling are often impractical. While resampling, caching, and retraining can mitigate this cost, they compromise convenience and accuracy and become impractical for time-varying data. We tackle these challenges using a query-efficient stochastic volume rendering framework based on delta tracking. Our system employs a four-stage pipeline that exploits heterogeneous parallelism, using ray tracing cores for traversal and tensor cores for batched neural evaluation. Furthermore, we present strategies to reduce INR queries via ray budgeting and query pruning, thereby increasing per-frame performance. Using our renderer, many time-varying INRs can be rendered directly from their original representation. The system achieves ~30-40 FPS at 1024x1024 resolution on an RTX 4090 GPU and converges to high-fidelity images. Moreover, the system enables interactive temporal exploration of the continuous domain, with timestep updates taking approximately 1-2 ms.

Figures

Figures reproduced from arXiv: 2607.28047 by the authors.

Figure 1
Figure 1. Volume-rendered frames of a time-varying implicit neural representation (INR) encoding X-ray tomography of a physically [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of our INR visualization framework: Each frame begins with [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Illustration of the lattice-shaped sampling process for the macro [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Comparison of ghost-pass macrocell refinement against ground [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Homogeneity-based query pruning on the S05_700 dataset. a) [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Data structure update time (lower is better) versus macrocell [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Frames per second performance (higher is better) of various volume rendering methods against increasing macrocell grid resolution for [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Evaluation of ray budgeting strategies. The top row shows image quality (higher is better) as a function of elapsed time (seconds) for the three [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Homogeneity-based query pruning on S05_700 and cylinder [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Frames per second performance over time with and without [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

78 extracted references · 12 canonical work pages

  1. [1]

    Ahrens, B

    J. Ahrens, B. Geveci, and C. Law. ParaView: An end-user tool for large-data visualization. In C. D. Hansen and C. R. Johnson, eds.,Visualization Hand- book. 2005. 5, 7

    2005

  2. [2]

    Andersson, J

    P. Andersson, J. Nilsson, and T. Akenine-Möller. Visualizing and Commu- nicating Errors in Rendered Images. In A. Marrs, P. Shirley, and I. Wald, eds.,Ray Tracing Gems II, chap. 19, pp. 301–320. 2021. 4

    2021

  3. [3]

    Baeza Rojo and T

    I. Baeza Rojo and T. Günther. Vector field topology of time-dependent flows in a steady reference frame.IEEE Transactions on Visualization and Computer Graphics (Proc. IEEE Scientific Visualization), 2019. 6

    2019

  4. [4]

    Bauer, Q

    D. Bauer, Q. Wu, H. Gadirov, and K.-L. Ma. Gscache: Real-time radiance caching for volume path tracing using 3d gaussian splatting.arXiv preprint arXiv:2507.19718, 2025. 2

    arXiv 2025

  5. [5]

    Bauer, Q

    D. Bauer, Q. Wu, and K.-L. Ma. FoV olNet: Fast V olume Rendering using Foveated Deep Neural Networks .IEEE Transactions on Visualization & Computer Graphics, 29(01):515–525, 2023. doi: 10.1109/TVCG.2022. 3209498 2

    doi:10.1109/tvcg.2022 2023

  6. [6]

    Brownlee and D

    C. Brownlee and D. Demarle. Fast volumetric gradient shading approxima- tions for scientific ray tracing. In A. Marrs, P. Shirley, and I. Wald, eds.,Ray Tracing Gems II. Apress Berkeley, CA, 2021. doi: 10.1007/978-1-4842 -7185-8 9

    work page doi:10.1007/978-1-4842 2021

  7. [7]

    Childs, E

    H. Childs, E. Brugger, B. Whitlock, J. Meredith, S. Ahern, D. Pugmire et al. VisIt: An End-User Tool For Visualizing and Analyzing Very Large Data,

  8. [8]

    Di and F

    S. Di and F. Cappello. Fast error-bounded lossy hpc data compression with sz. In2016 ieee international parallel and distributed processing symposium (ipdps), pp. 730–739. IEEE, 2016. 2

    2016

Show all 78 references
  1. [9]

    Dyken, A

    L. Dyken, A. Sewell, W. Usher, N. Debardeleben, S. Petruzza, and S. Kumar. V olume encoding gaussians: Transfer function-agnostic 3d gaussians for volume rendering.arXiv preprint arXiv:2504.13339, 2025. 2

    2025

  2. [10]

    J. Fong, M. Wrenninge, C. Kulla, and R. Habel. Production V olume Rendering: SIGGRAPH 2017 Course. InACM SIGGRAPH 2017 Courses, SIGGRAPH ’17, 2017. doi: 10.1145/3084873.3084907 2

    2017

  3. [11]

    Georgiev and M

    I. Georgiev and M. Fajardo. Blue-noise dithered sampling. InACM SIG- GRAPH 2016 Talks, SIGGRAPH ’16, art. no. 35, 1 p. Association for Computing Machinery, 2016. doi: 10.1145/2897839.2927430 2

    2016

  4. [12]

    Guenter, M

    B. Guenter, M. Finch, S. Drucker, D. Tan, and J. Snyder. Foveated 3d graphics.ACM Trans. Graph., 31(6), art. no. 164, 2012. doi: 10.1145/ 2366145.2366183 2

    2012

  5. [13]

    Günther, A

    T. Günther, A. Kuhn, and H. Theisel. MCFTLE: Monte Carlo Rendering of Finite-Time Lyapunov Exponent Fields.Computer Graphics Forum, 35,

  6. [14]

    Hachisuka, W

    T. Hachisuka, W. Jarosz, R. P. Weistroffer, K. Dale, G. Humphreys, M. Zwicker et al. Multidimensional adaptive sampling and reconstruc- tion for ray tracing.ACM Transactions on Graphics (Proceedings of SIG- GRAPH), 27(3):33:1–33:10, 2008. doi: 10/fm6c2w 5

    2008

  7. [15]

    Hadwiger, C

    M. Hadwiger, C. Sigg, H. Scharsach, K. Bühler, and M. Gross. Real-Time Ray-Casting and Advanced Shading of Discrete Isosurfaces.Computer Graphics Forum, 24, 2005. doi: 10.1111/j.1467-8659.2005.00855.x 2

    2005

  8. [16]

    J. Han, K. Tang, and C. Wang. Moe-inr: Implicit neural representation with mixture-of-experts for time-varying volumetric data compression.IEEE Transactions on Visualization and Computer Graphics, pp. 1–11, 2025. doi: 10.1109/TVCG.2025.3633893 2

    2025

  9. [17]

    Han and C

    J. Han and C. Wang. Coordnet: Data generation and visualization generation for time-varying volumes via a coordinate-based neural network.IEEE Transactions on Visualization and Computer Graphics, 29(12):4951–4963,

  10. [18]

    Han and F

    J. Han and F. Yang. Dcinr: a divide-and-conquer implicit neural repre- sentation for compressing time-varying volumetric data in hours.IEEE Transactions on Visualization and Computer Graphics, 2025. doi: 10.1109/ TVCG.2025.3564255 2

    2025

  11. [19]

    M. Han, A. Sewell, J. Insley, J. Knowles, V . A. Mateevitsi, M. E. Papka et al. Toward distributed 3d gaussian splatting for high-resolution isosurface visu- alization. In2025 IEEE International Conference on eScience (eScience), pp. 331–332. IEEE, 2025. 2

    2025

  12. [20]

    Hofmann, J

    N. Hofmann, J. Martschinke, K. Engel, and M. Stamminger. Neural De- noising for Path Tracing of Medical V olumetric Data.ACM Transactions on Graphics (Proceedings of SIGGRAPH ’20), 3(2), 2020. doi: 10.1145/ 3406181 1, 2

    2020

  13. [21]

    Ikits, J

    M. Ikits, J. Kniss, A. Lefohn, and C. Hansen. V olume Ren- dering Techniques. InGPU Gems. 2004. Available at https://developer.nvidia.com/sites/all/modules/custom/ gpugems/books/GPUGems/gpugems_ch39.html, Accessed: 24 June

    2004

  14. [22]

    Kerbl, G

    B. Kerbl, G. Kopanas, T. Leimkühler, G. Drettakis, et al. 3d gaussian splatting for real-time radiance field rendering. 2023. 2

    2023

  15. [23]

    Koskela, K

    M. Koskela, K. Immonen, T. Viitanen, P. Jääskeläinen, J. Multanen, and J. Takala. Foveated instant preview for progressive rendering. InSIG- GRAPH Asia 2017 Technical Briefs, art. no. 10. Association for Computing Machinery, 2017. doi: 10.1145/3145749.3149423 2

    2017

  16. [24]

    Krüger and R

    J. Krüger and R. Westermann. Acceleration Techniques for GPU-based V ol- ume Rendering. InProceedings of IEEE Visualization, IEEE Visualization Conference, 2003. doi: 10.1109/VISUAL.2003.1250384 2

    2003 arXiv

  17. [25]

    P. Kutz, R. Habel, Y . K. Li, and J. Novák. Spectral and decomposition tracking for rendering heterogeneous volumes.ACM Transactions on Graphics (Proceedings of SIGGRAPH 2017), 36(4), 2017. doi: 10.1145/ 3072959.3073665 2

    2017

  18. [26]

    Laine, T

    S. Laine, T. Karras, and T. Aila. Megakernels considered harmful: wave- front path tracing on gpus. InProceedings of the 5th High-Performance Graphics Conference, HPG ’13, pp. 137–143. Association for Computing Machinery, 2013. doi: 10.1145/2492045.2492060 3

    2013

  19. [27]

    Lindstrom

    P. Lindstrom. Fixed-rate compressed floating-point arrays.IEEE transac- tions on visualization and computer graphics, 20(12):2674–2683, 2014. doi: 10.1109/TVCG.2014.2346458 2

    2014

  20. [28]

    Ljung, C

    P. Ljung, C. Lundström, and A. Ynnerman. Multiresolution Interblock Interpolation in Direct V olume Rendering. InEUROVIS - Eurographic- s/IEEE VGTC Symposium on Visualization, 2006. doi: 10.2312/VisSym/ EuroVis06/259-266 2

    2006 doi

  21. [29]

    Y . Lu, K. Jiang, J. A. Levine, and M. Berger. Compressive neural representa- tions of volumetric scalar fields.Computer Graphics Forum, 40(3):135–146, 6 2021. doi: 10.1111/cgf.14295 2

    2021 doi

  22. [30]

    Martschinke, S

    J. Martschinke, S. Hartnagel, B. Keinert, K. Engel, and M. Stamminger. Adaptive Temporal Sampling for V olumetric Path Tracing of Medical Data. Computer Graphics Forum, 2019. doi: 10.1111/cgf.13771 1, 2

    2019 doi

  23. [31]

    Mildenhall, P

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, and R. Ng. Nerf: representing scenes as neural radiance fields for view synthesis.Commun. ACM, 65(1):99–106, 2021. doi: 10.1145/3503250 2

    2021 doi

  24. [32]

    K. A. Mohan, M. Ferrucci, C. Divin, G. A. Stevenson, and H. Kim. Dis- tributed stochastic optimization of a neural representation network for time-space tomography reconstruction.IEEE Transactions on Computa- tional Imaging, 11:362–376, 2025. doi: 10.1109/TCI.2025.3547265 1, 6

    2025

  25. [33]

    Morrical, A

    N. Morrical, A. Sahistan, U. Güdükbay, I. Wald, and V . Pascucci. Quick Clusters: A GPU-Parallel Partitioning for Efficient Path Tracing of Unstruc- tured V olumetric Grids.IEEE Transactions on Visualization and Computer Graphics, 29(01), 2023. doi: 10.1109/TVCG.2022.3209418 2, 3

    2023

  26. [34]

    Morrical, W

    N. Morrical, W. Usher, I. Wald, and V . Pascucci. Efficient Space Skipping and Adaptive Sampling of Unstructured V olumes Using Hardware Acceler- ated Ray Tracing. In2019 IEEE Visualization Conference, 2019. doi: 10. 1109/VISUAL.2019.8933539 2

    2019

  27. [35]

    Morrical, I

    N. Morrical, I. Wald, W. Usher, and V . Pascucci. Accelerating unstructured mesh point location with RT cores.IEEE Transactions on Visualization and Computer Graphics, pp. 1–14, In Press. doi: 10.1109/TVCG.2020. 3042930 2

    2020 doi

  28. [36]

    Morrical, S

    N. Morrical, S. Zellmann, A. Sahistan, P. Shriwise, and V . Pascucci. Attribute-Aware RBFs: Interactive Visualization of Time Series Particle V olumes Using RT Core Range Queries.IEEE Transactions on Visual- ization and Computer Graphics, 30(1), 2023. doi: 10.1109/TVCG.2023. 3327366 2

    2023 doi

  29. [37]

    T. Müller. tiny-cuda-nn. https://github.com/NVlabs/ tiny-cuda-nn, 2021. Version 2.0, released April 21, 2021. 9 10 © 2026 IEEE. This is the author’s version of the article that has been published in IEEE Transactions on Visualization and Computer Graphics. The final version of...

    2021

  30. [38]

    Nehab, P

    D. Nehab, P. V . Sander, J. Lawrence, N. Tatarchuk, and J. R. Isidoro. Accel- erating real-time shading with reverse reprojection caching. InProceedings of the 22nd ACM SIGGRAPH/EUROGRAPHICS Symposium on Graphics Hardware, GH ’07, pp. 25–35. Eurographics Association, 2007. 6

    2007

  31. [39]

    Novák, A

    J. Novák, A. Selle, and W. Jarosz. Residual ratio tracking for estimating attenuation in participating media.ACM Trans. Graph., 33(6), 2014. doi: 10.1145/2661229.2661292 2

    2014

  32. [40]

    Vingelmann, and F

    NVIDIA, P. Vingelmann, and F. H. Fitzek. CUDA release: 13. 6

  33. [41]

    NVIDIA OptiX Ray Tracing Engine

    NVIDIA Corporation. NVIDIA OptiX Ray Tracing Engine. Available at https://developer.nvidia.com/optix, Accessed: 27 March 2026. 6

    2026

  34. [42]

    Patney, M

    A. Patney, M. Salvi, J. Kim, A. Kaplanyan, C. Wyman, N. Benty et al. Towards foveated rendering for gaze-tracked virtual reality.ACM Trans. Graph., 35(6), art. no. 179, 2016. doi: 10.1145/2980179.2980246 2

    2016

  35. [43]

    Perlin and E

    K. Perlin and E. M. Hoffert. Hypertexture.SIGGRAPH Comput. Graph., 23(3), 1989. doi: 10.1145/74334.74359 2

    1989

  36. [44]

    Pharr, W

    M. Pharr, W. Jakob, and G. Humphreys. 11 V olume Scattering. InPhysically based rendering: From theory to implementation, pp. 697–736. The MIT Press, 4 ed., 2023. 2

    2023

  37. [45]

    S. Popinet. Free computational fluid dynamics.ClusterWorld, 2(6), 2004. 6

    2004

  38. [46]

    Popinet, M

    S. Popinet, M. Smith, and C. Stevens. Experimental and numerical study of the turbulence characteristics of airflow around a research vessel.Journal of Atmospheric and Oceanic Technology, 21(10):1575–1589, 2004. doi: 10. 1175/1520-0426(2004)021<1575:EANSOT>2.0.CO;2 6

    2004

  39. [47]

    A. W. Reed, H. Kim, R. Anirudh, K. A. Mohan, K. Champley, J. Kang et al. Dynamic ct reconstruction from limited views with implicit neural repre- sentations and parametric motion fields, 2021. doi: 10.1109/ICCV48922. 2021.00226 6

    2021

  40. [48]

    Reiser, S

    C. Reiser, S. Peng, Y . Liao, and A. Geiger. Kilonerf: Speeding up neural ra- diance fields with thousands of tiny MLPs. InProceedings of the IEEE/CVF international conference on computer vision, pp. 14335–14345, 2021. doi: 10.1109/ICCV48922.2021.01407 2

    2021

  41. [49]

    Reiser, R

    C. Reiser, R. Szeliski, D. Verbin, P. Srinivasan, B. Mildenhall, A. Geiger et al. Merf: Memory-efficient radiance fields for real-time view synthesis in unbounded scenes.ACM Transactions on Graphics (ToG), 42(4):1–12,

  42. [50]

    Rousselle, C

    F. Rousselle, C. Knaus, and M. Zwicker. Adaptive sampling and recon- struction using greedy error minimization. InProceedings of the 2011 SIGGRAPH Asia Conference, SA ’11, art. no. 159, 12 pp. Association for Computing Machinery, 2011. doi: 10.1145/2024156.2024193 5

    2011

  43. [51]

    Röttger, S

    S. Röttger, S. Guthe, D. Weiskopf, T. Ertl, and W. Straßer. Smart Hardware- Accelerated V olume Rendering. InProceedings of the Symposium on Data Visualisation 2003, vol. 3 ofVISSYM ’03, 2003. 2

    2003

  44. [52]

    Sahistan, S

    A. Sahistan, S. Demirci, I. Wald, S. Zellmann, J. Barbosa, N. Morrical et al. Visualization of large non-trivially partitioned unstructured data with native distribution on high-performance computing systems.IEEE Transactions on Visualization and Computer Graphics, pp. 1–14, 2...

    2024

  45. [53]

    Sahistan, S

    A. Sahistan, S. Zellmann, H. Miao, N. Morrical, I. Wald, and V . Pascucci. Materializing inter-channel relationships with multi-density woodcock tracking.IEEE Transactions on Visualization and Computer Graphics, 32(3):2726–2740, 2026. doi: 10.1109/TVCG.2026.3653310 2

    2026

  46. [54]

    Sahistan, S

    A. Sahistan, S. Zellmann, N. Morrical, V . Pascucci, and I. Wald. Multi- Density Woodcock Tracking: Efficient & High-Quality Rendering for Multi-Channel V olumes. InEurographics Symposium on Parallel Graphics and Visualization, 2025. doi: 10.2312/pgv.20251150 2

    2025 doi

  47. [55]

    Saragadam, J

    V . Saragadam, J. Tan, G. Balakrishnan, R. G. Baraniuk, and A. Veeraragha- van. Miner: Multiscale implicit neural representation. InEuropean Confer- ence on Computer Vision, pp. 318–333. Springer, 2022. doi: 10.1007/978 -3-031-20050-2_19 2

    2022 doi

  48. [56]

    Sitzmann, J

    V . Sitzmann, J. N. Martel, A. W. Bergman, D. B. Lindell, and G. Wetzstein. Implicit neural representations with periodic activation functions. InProc. NeurIPS, 2020. 6

    2020

  49. [57]

    Szirmay-Kalos, B

    L. Szirmay-Kalos, B. Tóth, and M. Magdics. Free Path Sampling in High Resolution Inhomogeneous Participating Media.Computer Graphics Fo- rum, 30(1), 2011. doi: 10.1111/j.1467-8659.2010.01831.x 2

    2011

  50. [58]

    Tang and C

    K. Tang and C. Wang. Ecnr: Efficient compressive neural representation of time-varying volumetric datasets.arXiv preprint arXiv:2311.12831, 2023. 2

    2023 arXiv

  51. [59]

    Viola, A

    I. Viola, A. Kanitsar, and M. Groller. Importance-driven volume rendering. InIEEE Visualization 2004, pp. 139–145, 2004. doi: 10.1109/VISUAL. 2004.48 2

    2004 doi

  52. [60]

    I. Wald, G. Johnson, J. Amstutz, C. Brownlee, A. Knoll, J. Jeffers et al. OSPRay - A CPU ray tracing framework for scientific visualization.IEEE Transactions on Visualization and Computer Graphics, 23(1):931–940,

  53. [61]

    I. Wald, W. Usher, N. Morrical, L. Lediaev, and V . Pascucci. RTX beyond ray tracing: Exploring the use of hardware ray tracing cores for tet-mesh point location. InProceedings of High-Performance Graphics - Short Papers, HPG ’19, 2019. doi: 10.2312/hpg.20191189 2

    2019 doi

  54. [62]

    I. Wald, S. Zellmann, W. Usher, N. Morrical, U. Lang, and V . Pascucci. Ray Tracing Structured AMR Data Using ExaBricks.IEEE Transactions on Visualization and Computer Graphics, 27(2), 2021. doi: 10.1109/TVCG. 2020.3030470 2

    2021

  55. [63]

    Wolfe, N

    A. Wolfe, N. Morrical, T. Akenine-Möller, and R. Ramamoorthi. Spatiotem- poral Blue Noise Masks. In A. Ghosh and L.-Y . Wei, eds.,Eurographics Symposium on Rendering. The Eurographics Association, 2022. doi: 10. 2312/sr.20221161 2, 6

    2022

  56. [64]

    Woodcock, T

    E. Woodcock, T. Murphy, H. P., and L. T.C. Techniques used in the GEM Code for Monte Carlo Neutronics Calculation in Reactors and Other Sys- tems of Complex Geometry. Technical report, Argonne National Laboratory,

  57. [65]

    Q. Wu, W. Usher, S. Petruzza, S. Kumar, F. Wang, I. Wald et al. VisIt- OSPRay: Toward an Exascale V olume Visualization System. In H. Childs and F. Cucchietti, eds.,Eurographics Symposium on Parallel Graphics and Visualization. The Eurographics Association, 2018. doi: 10.2312/...

    2018 doi

  58. [66]

    L. Yang, S. Liu, and M. Salvi. A survey of temporal antialiasing techniques. Computer Graphics Forum, 39(2):607–621, 2020. doi: 10.1111/cgf.14018 6

    2020 doi

  59. [67]

    J. Ye, A. Xie, S. Jabbireddy, Y . Li, X. Yang, and X. Meng. Rectangular Mapping-based Foveated Rendering . In2022 IEEE on Conference Virtual Reality and 3D User Interfaces (VR), pp. 756–764. IEEE Computer Society,

  60. [68]

    Y . Yue, K. Iwasaki, B.-Y . Chen, Y . Dobashi, and T. Nishita. Unbiased, Adaptive Stochastic Sampling for Rendering Inhomogeneous Participating Media.ACM Transactions on Graphics, 29(6), 2010. doi: 10.1145/1882261 .1866199 2

    2010 doi

  61. [69]

    Zavorotny, Q

    D. Zavorotny, Q. Wu, D. Bauer, and K.-L. Ma. From Cluster to Desktop: A Cache-Accelerated INR framework for Interactive Visualization of Tera- Scale Data. In G. Reina, S. Rizzi, and C. Gueunet, eds.,Eurographics Symposium on Parallel Graphics and Visualization. The Eurographic...

    2025 doi

  62. [70]

    Zellmann, D

    S. Zellmann, D. Seifried, N. Morrical, I. Wald, W. Usher, J. Law-Smith et al. Point containment queries on ray tracing cores for AMR flow visualiza- tion.Computing in Science Engineering, 2022. doi: 10.1109/mcse.2022. 3153677 2

    2022 doi

  63. [71]

    Zellmann, Q

    S. Zellmann, Q. Wu, A. Sahistan, K.-L. Ma, and I. Wald. Beyond ExaBricks: GPU V olume Path Tracing of AMR Data.Computer Graphics Forum, 43(3),

  64. [72]

    doi: 10.1109/VR51125.2022.00097 2

    2022

  65. [78]

    Zvonek and A

    J. Zvonek and A. Gillette. Pruning AMR: Efficient visualization of implicit neural representations via weight matrix analysis, 2025. doi: 10.48550/ arXiv.2512.02967 9 11

    2025 doi

  66. [2016]

    doi: 10.1111/cgf.12914 1, 2

    doi

  67. [2017]

    doi: 10.1109/TVCG.2016.2599041 5, 7

    2016

  68. [2022]

    doi: 10.1109/TVCG.2022.3197203 2, 6

    2022

  69. [2023]

    doi: 10.1145/3592426 2

    doi

  70. [2024]

    doi: 10.1111/cgf.15095 2

    doi

Pith tools

Reviewed July 31, 2026 · model on record in the stance chip above.