Pith. sign in

REVIEW 4 major objections 4 minor 26 references

Parametric Integration with Neural Integral Operators

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A trained neural operator can approximate the light transport integral from a five-dimensional projection of incident radiance, yielding noise-free shading from a single sample per pixel.

desk verdict A clearly-specified denoise-before-shading pipeline with a genuinely new idea, but the central projection-sufficiency assumption is unproven and the evaluation is qualitative only. read the letter →

arxiv 2507.17440 v1 pith:7D7SPP2L submitted 2025-07-23 cs.GR

Christoph Schied , Alexander Keller This is my paper · ORCID
classification cs.GR
keywords neuralintegraloperatorparametricintegrationpre-shadingdenoisingprojectedirradiancematerial-agnosticreal-timerenderingnoise-to-noisetrainingU-Net
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that the right place to denoise a noisy rendered image is before material shading, not after. It replaces the BSDF integral for reflected radiance with a learned neural operator that acts on a low-dimensional projection of incident radiance, a vector of irradiance coefficients. Because the projected irradiance is smoother than the shaded result, filtering it preserves material and geometric detail that post-shading filters tend to blur. The authors show that this pipeline runs in real time with data from only one frame, trains without clean references using a noise-to-noise scheme, and can be combined with existing denoisers and temporal anti-aliasing.

What carries the argument

The central object is the projected irradiance $\mu^* = \int E(\omega,\omega_r) L_i(x,\omega) \cos\theta\, d\omega$, a vector-valued integral evaluated by Monte Carlo with one path sample plus next-event estimation. $E$ has five components per color channel: the constant 1, which is the diffuse irradiance, and four BSDF-derived lobes, making $\mu^*$ a 15-dimensional quantity that is smoother than the reflected radiance integral. Around this object the pipeline is built: a U-Net $U_\theta$ denoises the normalized, logarithm-compressed projected irradiance, and a small ResNet $M_\theta$, the material decoder, maps the denoised coefficients together with material and view parameters to three softmax weights and an intensity, forming the final color as a convex combination of black, base color, and white multiplied by the filtered irradiance.

What would settle it

Construct a material and two lighting environments that are visually distinct under the full BSDF integral but have identical five-coefficient projected irradiance at every surface point; if the trained decoder produces nearly identical images for both, the projection is lossy and the central approximation fails on that input. Concretely, replace a broad area light with a sharp spot light tuned so that the constant-plus-four-lobe coefficients match, and compare the rendered highlights.

Watch

Extended reading notes

Core claim

The paper's central claim is the approximation $L_r(x,\omega_r) \approx M_\theta(x,\omega_r, \int E(\omega,\omega_r) L_i(x,\omega) \cos\theta\, d\omega)$: a small network, the material decoder, can act as a neural integral operator, mapping a projected-irradiance vector $\mu^*$ to the reflected radiance that the full BSDF integral would produce. The projection basis $E$ is built from the constant function plus four spherical-Gaussian lobes derived from a production BSDF model, so the five coefficients per color channel capture the information the decoder needs. Denoising is done in this projected-irradiance space by a U-Net, and the decoder then reconstructs shading as a convex combination of black, base color, and white scaled by the filtered irradiance. The authors validate the idea by comparing 1-spp neural shading against 2048-spp references.

Load-bearing premise

The 15 irradiance coefficients (five per color channel) are assumed to capture enough of the incident radiance distribution for the decoder to reconstruct the true shading for every material and lighting the network will meet; the training data, with only four light directions per example, does not prove this sufficiency.

Editorial extensions

If this is right

  • Single-frame, 1-spp inputs can be rendered effectively noise-free in real time without temporal accumulation or motion vectors.
  • Because denoising happens on irradiance rather than shaded color, fine material and geometric details are preserved that post-shading filters blur.
  • The material decoder is trained independently of scene geometry and materials, then the denoiser is trained noise-to-noise, making training fast and scene-agnostic.
  • The pipeline plugs into existing sampling algorithms and can be combined with denoisers and temporal anti-aliasing, and it supports 2x2 subsampling.
  • The same parametric-integration idea extends beyond shading to general functionals of solutions of integral equations, as the authors note.

Reading between the lines

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

  • If the five-coefficient projection is sufficient, the same operator scheme could be applied to other integrals in rendering, such as indirect illumination or light field reconstruction, where the integrand is high-dimensional but its low-order projection is smooth.
  • The projection basis is derived from one specific BSDF family; swapping it for a learned basis could extend the method to anisotropic materials or to other appearance models without redesigning the decoder.
  • The consistency constraint in the noise-to-noise loss suggests the denoiser could be trained entirely in projected space, so the material decoder could be retuned offline without retraining the denoiser.
  • A direct stress test would be to render scenes whose incident radiance shares the same 5-coefficient projection but differs in the full angular distribution; if the decoder cannot reproduce the shading differences, the projection is lossy and a higher-dimensional basis would be needed.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes a neural shading pipeline for real-time rendering at extremely low sampling rates (1 sample per pixel). Instead of denoising the final shaded image, the method first projects incident radiance onto a five-dimensional basis built from a constant plus four Disney BSDF lobes, denoises this projected irradiance with a U-Net, and then applies a trained material decoder M_θ that approximates the reflected radiance integral (Eq. (2)). The authors argue that filtering before shading preserves material detail and that the resulting pipeline is material-agnostic, single-frame, and compatible with existing denoisers and temporal anti-aliasing. Training is split into a decoder pass and a dependent or independent denoiser pass, with Noise2Noise-style losses and a consistency constraint. The central claim is that Eq. (2), the neural surrogate for the light transport integral, is faithful enough to produce practically noise-free images from a single frame. The paper presents qualitative comparisons in Fig. 5, but no quantitative error measurements, no baseline comparisons, and no runtime data.

Significance. If the proposed approximation holds, the paper offers an attractive alternative to post-shading denoisers: denoising the integrand in a higher-dimensional projected-irradiance space is conceptually elegant and could enable high-quality real-time rendering at 1 spp. The training protocol is described in detail, including loss functions, optimizer settings, and the independent denoiser variant in Sect. 3.3, which helps guard against circularity concerns. The decoupling of the material decoder from scene geometry and the possibility of training it on procedural data are also practical strengths. However, the significance is currently not demonstrated quantitatively. The load-bearing assumption—that a five-dimensional projection of incident radiance is sufficient for the decoder to reconstruct reflected radiance over the full range of Disney BSDF parameters—is not backed by either a representation theorem or an experiment that measures the approximation error. The paper is a promising research direction, but the magnitude of the claimed benefits (real-time, single-frame, noise-free) requires stronger evidence than the provided visual comparisons.

major comments (4)
  1. [Eq. (2), Sect. 2.1, Appendix A.1] The central approximation in Eq. (2) is only justified if the five-dimensional projected irradiance μ* retains enough information about the incident radiance L_i to determine the reflected radiance L_r. The four non-constant basis functions in A.1 are fixed Disney lobes, but a general Disney BSDF with continuous roughness, specular, and metallic parameters is not in their span, so the projection L_i → μ* is generally not injective. The paper provides no representation theorem, no error bound, and no experiment isolating the approximation error of M_θ. Because every claimed benefit (denoising before shading, material-agnostic filtering, and 1-spp real-time rendering) depends on Eq. (2) being a faithful surrogate, this is a load-bearing gap. Please either add a formal statement with explicit assumptions or provide a quantitative study that varies BSDF parameters and incident lighting complexity (e.g., point lights vs. environment maps) and measures the error of M_θ against the true integral in Eq. (1).
  2. [Sect. 3.1] The material decoder is trained with only four uniformly sampled incident directions per training example. The input to Eq. (2) is an integral over the whole hemisphere, and four point samples do not span the space of incident radiance distributions; consequently, the decoder may only learn to invert the projection for sparse point-light configurations. Moreover, the reference L_r is computed from the same four-sample finite sum, so the training loss does not measure error with respect to the true integral. Please train and evaluate with full hemisphere integration (or at least many more directions) to establish that M_θ approximates the integral operator rather than a particular Monte Carlo estimator.
  3. [Sect. 2.4, Fig. 5] The paper's central claims—noise-free 1-spp rendering, preservation of material detail, and real-time performance—are supported only by visual comparisons in Fig. 5. There are no error metrics (e.g., RMSE, SSIM, or relative MSE against the reference), no baseline comparisons to post-shading denoisers such as Bako et al., no ablation of the dependent vs. independent denoiser training, and no runtime measurements anywhere in the manuscript. Please add quantitative evaluation and ablations so that the claims can be verified.
  4. [Sect. 2.2, Eq. (7)] The denoiser output is multiplied by the blurred irradiance μ̂_R,0 to restore dynamic range. The paper intentionally omits edge-stopping heuristics, arguing that they may mask fine geometry. This design choice is plausible, but no analysis or experiment is provided to show how the wide a-trous blur behaves across geometric discontinuities, nor how much detail is lost compared to a post-shading denoiser. Please add an experiment or at least a discussion quantifying the trade-off between noise reduction and detail preservation, since preserving detail is a central motivation of the method.
minor comments (4)
  1. [Sect. 3.2] The text uses the German word 'und' in 'μ_A und μ_B'; this should be 'and'.
  2. [Eqs. (5) and (6)] The normalization uses max{·, ε} to avoid division by zero, but for a color channel with irradiance below ε, the division is by ε, which can amplify values instead of smoothly compressing them; the statement that 'zero components remain zero' only holds for exactly zero channels. Consider a smoother treatment or a brief justification.
  3. [Title and Sect. 4.3] The term 'material-agnostic denoising' is imprecise, since the decoder is restricted to one specific isotropic BSDF model (as acknowledged in Sect. 4.3). Consider using 'material-parameter-agnostic denoising' or clarifying the scope in the abstract.
  4. [References] Reference [24] points to a Wikipedia page for irradiance; a standard textbook or the original irradiance-caching reference would be more appropriate.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Eq. (2) is a supervised approximation trained against the BSDF target, not a fitted input renamed as a prediction.

full rationale

The central approximation (2) is not derived from Eq. (1) by construction; it is a learned surrogate M_theta trained with a relative-MSE loss against reference L_r values computed by evaluating the Disney BSDF (Sect. 3.1, A.1). Using the same BSDF model to define the projection basis E (one constant plus four lobes) and to generate supervised training references is a basis-choice/matching strategy, not circular: the decoder is not given the target L_r as an input, and the training distribution spans material parameters, directions, and radiance values. The denoiser (Sect. 3.2/3.3) is trained Noise2Noise on two independent projected-irradiance estimates, optionally with the material decoder in the loop, but the decoder is trained first without the denoiser; no fitted parameter is later renamed a prediction. The paper claims no uniqueness theorem and relies on no author-self-citation for the approximation: references to the authors' previous work ([15], [16], [22]) appear in discussion/future work as context, not as load-bearing justification for Eq. (2). The projection-sufficiency assumption (that 5 components capture enough incident-radiance information) is an unproven representational premise and a correctness risk, but it is not a circular reduction: the neural operator could fail without the derivation being equivalent to its input. Evaluation on unseen scenes (Fig. 5) provides independent, if only visual, evidence.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central approximation rests on the choice of the projection basis (free parameters), the sufficiency of that projection (domain assumption), and the generalization of networks trained on synthetic data (domain assumption). No new physical entities are introduced.

free parameters (6)
  • Projection basis lobe parameters (metallic, specular, roughness for E1-E4) = E1: (0.5, 0.5, 0.1); E2: (1.0, 0.0, 0.1); E3: (0.0, 1.0, 0.1); E4: (1.0, 1.0, 0.6)
    Chosen by hand from the Disney BSDF model (Sect. A.1). They define the projected irradiance representation mu and therefore the information available to the material decoder.
  • Number of basis components d = 5
    The dimensionality of the projection space. Chosen by hand; the central claim that the projection is sufficient depends on this choice.
  • Training radiance range [0,16]^3 = 16 per channel
    Used for procedurally generated training examples (Sect. 3.1). The paper states this range is a free parameter, though intensity is decoupled.
  • BSDF output clamp maximum = 16.0
    Restricts the output range of the BSDF during training to aid convergence (Sect. 3.1). This limits the functions the decoder must approximate.
  • Minimum roughness = 0.1
    Limits material roughness to avoid Dirac delta degenerate BSDFs (Sect. 3.1). This changes the target distribution.
  • Importance sampling roughness clamp = 0.5
    Clamps roughness during BSDF importance sampling to ensure coverage of highlights (Sect. 3.1). Affects training sample distribution.
assumptions (6)
  • standard math Monte Carlo and quasi-Monte Carlo integration evaluate the projected irradiance integral in (2) without bias.
    Invoked in Sect. 2.1 to justify computing mu by summing path contributions. Standard MC convergence.
  • domain assumption The finite-dimensional projection onto E(omega,omega_r) captures sufficient information for the material decoder to reconstruct the reflected radiance integral for the target BSDF family.
    This is the core representational assumption. The paper provides no proof that d=5 components are sufficient, only empirical images.
  • standard math The a-trous wavelet filter preserves average image brightness, so the denoiser can restore absolute radiance.
    Invoked in Sect. 2.2 before Eq (7). This is a property of the a-trous transform.
  • domain assumption The U-Net can learn to invert the logarithmic compression and map noisy projected irradiance to the clean projected irradiance.
    The success of the denoiser depends on the representational capacity of the U-Net and the separability of noise from signal in the projected space.
  • domain assumption Noise2Noise training with two independent noisy observations converges to the same result as regression to the clean signal.
    Borrowed from Lehtinen et al. [11]; assumes the noise has zero mean conditioned on the input.
  • domain assumption Training on procedurally generated random materials and lights generalizes to real scenes.
    The networks are trained without ray-traced scene data for the decoder, and with synthetic per-pixel material randomization for the denoiser. Generalization to test scenes is assumed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Parametric Integration with Neural Integral Operators." pith.science (2026). https://pith.science/paper/7D7SPP2L

@misc{pith2026250717440,
  author       = {Pith},
  title        = {Pith review of: Parametric Integration with Neural Integral Operators},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7D7SPP2L}},
  note         = {Machine review of arXiv:2507.17440}
}
read the original abstract

Real-time rendering imposes strict limitations on the sampling budget for light transport simulation, often resulting in noisy images. However, denoisers have demonstrated that it is possible to produce noise-free images through filtering. We enhance image quality by removing noise before material shading, rather than filtering already shaded noisy images. This approach allows for material-agnostic denoising (MAD) and leverages machine learning by approximating the light transport integral operator with a neural network, effectively performing parametric integration with neural operators. Our method operates in real-time, requires data from only a single frame, seamlessly integrates with existing denoisers and temporal anti-aliasing techniques, and is efficient to train. Additionally, it is straightforward to incorporate with physically based rendering algorithms.

Figures

Figures reproduced from arXiv: 2507.17440 by the authors.

Figure 1
Figure 1. Comparison of the components of the projected irradiance 𝜇 at 1 sample per pixel (1 spp), the denoised projected irradiance 𝜇 ′ , and the ground truth reference 𝜇 ∗ . Note that the first component (top row) actually is the classic irradiance. The geometry is recognizable as the cos 𝜗𝑟 term and next event estimation are included in the estimates. Next event estimation contributes shadows from visibility and quadratic… view at source ↗
Figure 2
Figure 2. Visualization of the denoiser inputs irradiance (𝜇𝑅,0, 𝜇𝐺,0, 𝜇𝐺,0), normals, depth, and the the blurred irradiance (𝜇ˆ𝑅,0, 𝜇ˆ𝐺,0, 𝜇ˆ𝐺,0) that is practically noise-free. Dividing the irradiance by the blurred irradiance normalizes scale, yielding almost gray images. The remaining high dynamic range noise is compressed by applying a logarithm. Note that black components remain black after the transformation. ReLU U-Ne… view at source ↗
Figure 3
Figure 3. As compared to the original U-Net [20], we only use one convolution per layer and ResNets [6] for the stacked 1x1 convolutions, resulting in a more efficient architecture. The network has about 9 million weights [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: The image produced by the material decoder 𝑀𝜃,𝑅 in (11) is quite close to the reference. For illustration, we show a false color visualization of the normalized outputs and the intensity 𝐼 of the material decoder neural network (10) as used in the convex combination (1…
Figure 5
Figure 5. Figure 5: Comparison of a) using the trained material decoder 𝑀𝜃 in (2) for shading at 1 and averaged 2048 samples per pixel (spp), and to the projected irradiance 𝜇 ∗ , approximated by 2048 spp, b) the reference as provided by the Unreal Engine BSDF, and c) material agnostic de…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 20 canonical work pages

  1. [1]

    ACM Transactions on Graphics (Proceedings of SIGGRAPH 2017) 36(4) (2017)

    Bako, S., Vogels, T., McWilliams, B., Meyer, M., Nov´ak, J., Harvill, A., Sen, P., DeRose, T., Rousselle, F.: Kernel-predicting convolutional networks for denoising Monte Carlo render- ings. ACM Transactions on Graphics (Proceedings of SIGGRAPH 2017) 36(4) (2017)

    work page 2017

  2. [2]

    Binder, N., Fricke, S., Keller, A.: Massively parallel path space filtering. In: A. Keller (ed.) Monte Carlo and Quasi-Monte Carlo Methods, MCQMC 2020, Oxford, United Kingdom, August 10–14, pp. 149–168. Springer (2022)

    work page 2022

  3. [3]

    In: Proceedings of the Conference on High Perfor- mance Graphics, HPG ’10, pp

    Dammertz, H., Sewtz, D., Hanika, J., Lensch, H.P.A.: Edge-avoiding `a-trous wavelet trans- form for fast global illumination filtering. In: Proceedings of the Conference on High Perfor- mance Graphics, HPG ’10, pp. 67–75. Eurographics Association (2010)

    work page 2010

  4. [4]

    Game Developers Conference

    Deligiannis, J., Schmid, J.: ”It Just Works: Ray-Traced Reflections in ’Battlefield V’. Game Developers Conference. (2019). URL https://www.gdcvault.com/play/1026282/ It-Just-Works-Ray-Traced

    work page arXiv 2019

  5. [5]

    In: International Conference on Learning Representations (2022)

    Guibas, J., Mardani, M., Li, Z., Tao, A., Anandkumar, A., Catanzaro, B.: Efficient token mixing for transformers via adaptive Fourier neural operators. In: International Conference on Learning Representations (2022). URL https://openreview.net/forum?id=EXHG-A3jlM Parametric Integration with Neural Integral Operators 15

    work page 2022

  6. [6]

    He, K., Zhang, X., Ren, S., Sun, J.: Identity mappings in deep residual networks. In: B. Leibe, J. Matas, N. Sebe, M. Welling (eds.) Computer Vision – ECCV 2016, pp. 630–645. Springer International Publishing, Cham (2016)

    work page 2016

  7. [7]

    Holschneider, M., Kronland-Martinet, R., Morlet, J., Tchamitchian, P.: A real-time algorithm for signal analysis with the help of the wavelet transform. In: J. Combes, A. Grossmann, P. Tchamitchian (eds.) Wavelets - Time-Frequency Methods and Phase Space, pp. 286–297. Springer (1989). DOI 10.1007/978-3-642-75988-8 28

    work page doi:10.1007/978-3-642-75988-8 1989

  8. [8]

    Keller, A.: Quasi-Monte Carlo Methods for Photorealistic Image Synthesis. Ph.D. thesis, University of Kaiserslautern, Germany (1998)

    work page 1998

Show all 26 references
  1. [9]

    Keller, A., Dahm, K., Binder, N.: Path space filtering. In: R. Cools, D. Nuyens (eds.) Monte Carlo and Quasi-Monte Carlo Methods 2014, pp. 423–436. Springer (2016)

    2016

  2. [10]

    Synthesis lectures in computer graphics and animation

    K ˇriv´anek, J., Gautron, P.: Practical Global Illumination with Irradiance Caching. Synthesis lectures in computer graphics and animation. Morgan & Claypool (2009)

    2009

  3. [11]

    Lehtinen, J., Munkberg, J., Hasselgren, J., Laine, S., Karras, T., Aittala, M., Aila, T.: Noise2Noise: Learning image restoration without clean data. Proc. International Confer- ence on Machine Learning (ICML) (2018)

    2018

  4. [12]

    In: International Conference on Learning Representations (2021)

    Li, Z., Kovachki, N.B., Azizzadenesheli, K., liu, B., Bhattacharya, K., Stuart, A., Anandkumar, A.: Fourier neural operator for parametric partial differential equations. In: International Conference on Learning Representations (2021). URL https://openreview.net/forum?id= c8P9NQVtmnO

    2021

  5. [13]

    Nature Machine Intelligence 3(3), 218–229 (2021)

    Lu, L., Jin, P., Pang, G., Zhang, Z., Karniadakis, G.E.: Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators. Nature Machine Intelligence 3(3), 218–229 (2021). DOI 10.1038/s42256-021-00302-5. URL https://doi.org/ 10.1038/s42256-0...

    2021 doi

  6. [14]

    In: ACM SIGGRAPH 2012 Courses, SIGGRAPH ’12 (2012)

    McAuley, S., Hill, S., Hoffman, N., Gotanda, Y., Smits, B., Burley, B., Martinez, A.: Practical physically-based shading in film and game production. In: ACM SIGGRAPH 2012 Courses, SIGGRAPH ’12 (2012). DOI 10.1145/2343483.2343493

    2012

  7. [15]

    ACM Transactions on Graphics 41(4), 102:1–102:15 (2022)

    M¨ uller, T., Evans, A., Schied, C., Keller, A.: Instant neural graphics primitives with a mul- tiresolution hash encoding. ACM Transactions on Graphics 41(4), 102:1–102:15 (2022). DOI 10.1145/3528223.3530127

    2022

  8. [16]

    ACM Transactions on Graphics (SIGGRAPH) 40(4), 36:1–36:16 (2021)

    M¨ uller, T., Rousselle, F., Nov´ak, J., Keller, A.: Real-time neural radiance caching for path tracing. ACM Transactions on Graphics (SIGGRAPH) 40(4), 36:1–36:16 (2021). DOI 10.1145/3450626.3459812

    2021

  9. [17]

    Computer Graphics Forum 36(4), 65–78 (2017)

    Nalbach, O., Arabadzhiyska, E., Mehta, D., Seidel, H.P., Ritschel, T.: Deep shading: Convo- lutional neural networks for screen space shading. Computer Graphics Forum 36(4), 65–78 (2017). DOI 10.1111/cgf.13225

    2017 doi

  10. [18]

    Morgan Kaufmann, fourth edition (2023)

    Pharr, M., Jacob, W., Humphreys, G.: Physically Based Rendering - From Theory to Imple- mentation. Morgan Kaufmann, fourth edition (2023)

    2023

  11. [19]

    ACM Transactions on Graphics 21(3), 267–276 (2002)

    Reinhard, E., Stark, M., Shirley, P., Ferwerda, J.: Photographic tone reproduction for digital images. ACM Transactions on Graphics 21(3), 267–276 (2002)

    2002

  12. [20]

    Ronneberger, O., Fischer, P., Brox, T.: U-Net: Convolutional networks for biomedical image segmentation. In: N. Navab, J. Hornegger, W.M. Wells, A.F. Frangi (eds.) Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015, pp. 234–241. Springer International Pub...

    2015

  13. [21]

    ACM Transactions on Graphics42(4) (2023)

    Vaidyanathan, K., Salvi, M., Wronski, B., Akenine-M¨oller, T., Ebelin, P., Lefohn, A.: Random- access neural compression of material textures. ACM Transactions on Graphics42(4) (2023). DOI 10.1145/3592407

    2023 doi

  14. [22]

    In: Interspeech 2024, pp

    Van keirsbilck, M., Keller, A.: Conformer without convolutions. In: Interspeech 2024, pp. 3475–3479 (2024). DOI 10.21437/Interspeech.2024-445

    2024 doi

  15. [23]

    In: 3rd Eurographics Workshop on Rendering

    Ward, G., Heckbert, P.: Irradiance gradients. In: 3rd Eurographics Workshop on Rendering. Bristol, United Kingdom (1992) 16 Christoph Schied and Alexander Keller

    1992

  16. [24]

    URL https://en.wikipedia.org/wiki/Irradiance

    Wikipedia: Irradiance (2025). URL https://en.wikipedia.org/wiki/Irradiance

    2025

  17. [25]

    In: ACM SIGGRAPH 2023 Courses, SIGGRAPH ’23 (2023)

    Wyman, C., Kettunen, M., Lin, D., Bitterli, B., Yuksel, C., Jarosz, W., Kozlowski, P.: A gentle introduction to ReSTIR path reuse in real-time. In: ACM SIGGRAPH 2023 Courses, SIGGRAPH ’23 (2023). DOI 10.1145/3587423.3595511

    2023

  18. [26]

    ACM Transactions on Graphics 43(3) (2024)

    Zeltner, T., Rousselle, F., Weidlich, A., Clarberg, P., Nov´ak, J., Bitterli, B., Evans, A., Davi- doviˇc, T., Kallweit, S., Lefohn, A.: Real-time neural appearance models. ACM Transactions on Graphics 43(3) (2024). DOI 10.1145/3659577 A Appendix A.1 Disney Principled BSDF Mod...

    2024 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.