Pith. sign in

REVIEW 4 major objections 5 minor 100 references

Neural Inverse Rendering from Propagating Light

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A physics-based renderer inverts flash-lidar light echoes into accurate 3D geometry.

desk verdict A genuinely new time-resolved radiance cache for physically-based inverse rendering from flash lidar; the simulated geometry results are strong, and the main soft spot is the captured-data calibration premise, which is acknowledged but not stress-tested. read the letter →

arxiv 2506.05347 v1 pith:MZZZMF6Z submitted 2025-06-05 cs.CV

Anagh Malik , Benjamin Attal , Andrew Xie , Matthew O'Toole , David B. Lindell This is my paper
classification cs.CV
keywords inverserenderingtransientimagingflashlidarneuralradiancecacheglobalilluminationtime-resolvedfieldsrelighting
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that multi-view flash lidar videos, which record light as it propagates through a scene, can be inverted with a physically based renderer to recover accurate geometry even under strong indirect light. The key claim is that direct and multiply-scattered light are explained by a single physical model, so the recovered geometry, normals, and materials are consistent with all measured echoes rather than only the first return. This matters because conventional lidar reconstruction discards indirect light, losing information that this method exploits. The authors show that the approach improves normal accuracy on simulated scenes, cutting mean absolute error from 22.80 to 8.45 relative to the best baseline, and supports new capabilities such as time-resolved relighting and direct/indirect decomposition.

What carries the argument

The time-resolved radiance cache: a neural representation that stores the infinite-bounce radiance arriving at any point from any direction, replacing the recursive evaluation of the rendering equation with a cache lookup and a single integration. For each point on a primary sensor ray, secondary rays are cast and volume-rendered against the same density field used for geometry, and two small MLPs predict time-resolved direct and indirect outgoing radiance conditioned on a hash-encoded appearance feature and the light-source position. This removes the need to trace recursive light paths while preserving the physics of multi-bounce light transport.

What would settle it

In a scene with known ground-truth geometry, place a specular reflector so that a two-bounce path reaches a checkerboard corner before the direct reflection; if the recovered normal or depth map shows a localized error centered at that corner, the first-peak calibration assumption is the cause. Alternatively, compare the trilaterated light-source position against an independently measured mechanical position and check whether the depth errors correlate with the discrepancy.

Watch

Extended reading notes

Core claim

The central claim is that a time-resolved extension of neural radiance caching makes physically based inverse rendering from captured propagating-light measurements tractable, and that doing so yields better geometry than models that ignore or only approximate indirect transport. The rendering equation is evaluated with a direct term, an analytic delta-source with inverse-square falloff and time delay, and an indirect term obtained by querying a cache of time-resolved radiance at any point and direction. The cache is rendered by volume-casting secondary rays against the same density field that defines scene geometry, and it is decomposed into direct and indirect parts with a split-sum approximation. Optimizing this model against multi-view lidar measurements recovers density, normals, and Disney-GGX material parameters, with the direct and indirect components each constrained by a radiometric prior. On four simulated scenes the method reduces normal MAE from 22.80 to 8.45 and depth L1 error from 0.47 to 0.21 compared with the strongest baseline, while on captured scenes it produces qualitatively cleaner normals in regions dominated by inter-reflections.

Load-bearing premise

The flash lidar source is treated as a calibrated point light with known 3D position and measured directional intensity profile, with the position estimated by trilateration of first time-of-flight peaks that are assumed to be direct-only; if indirect light biases those peaks, the physically based model will attribute the mismatch to geometry and normals.

Editorial extensions

If this is right

  • Time-resolved relighting of captured scenes becomes possible: the recovered geometry and materials let a user place a new pulsed source and render the resulting propagating light from novel viewpoints.
  • The same physical model can be supervised by continuous-wave time-of-flight measurements or steady-state intensity images, and it still produces videos of light propagation, so the approach does not strictly require ultrafast lidar hardware.
  • Because the renderer is physically grounded, it avoids artifacts of unconstrained neural baselines such as mirror-copy geometry used to explain specular reflections or depth shifts used to explain diffuse inter-reflections.
  • Material parameters (albedo, roughness, metalness) are recovered as a byproduct, enabling relighting and material-aware editing of the captured scene.
  • Direct and indirect components of the measured light transport are separated automatically, which is useful for applications that need to distinguish first-return geometry from multi-bounce contamination.

Reading between the lines

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

  • The paper leaves implicit that jointly refining the light-source position and intensity profile during optimization could close the observed gap between simulated and captured performance, since the method's sensitivity to calibration is stated as the likely cause of that gap.
  • The cache is conditioned on light-source position, so the framework could in principle be extended to non-line-of-sight imaging by treating the hidden source position or relay-surface geometry as latent variables to be optimized.
  • Because CW-ToF and intensity supervision suffice to recover transient videos, the method points toward extracting light-in-flight from conventional time-of-flight sensors in consumer devices, albeit at coarser temporal resolution.
  • A testable extension: if the first-time-of-flight peaks used for source calibration are biased by indirect light, recomputing the source position with a robust estimator that discards outliers should measurably improve recovered normals on captured scenes.
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

4 major / 5 minor

Summary. This paper presents a method for inverse rendering from multi-viewpoint, time-resolved flash lidar measurements. It extends neural radiance caching to the time domain, decomposing incident radiance into direct and indirect components, and optimizes a Zip-NeRF-style density field, normals, materials, and a radiance cache against the captured transient measurements. The method is evaluated on simulated scenes (Cornell box, pots, peppers, kitchen) and captured scenes (globe, house, spheres, statue), reporting view-synthesis metrics and, for simulation, depth and normal error. The authors also demonstrate time-resolved relighting and training from continuous-wave ToF or intensity images. The central claim is that physically-based modeling of multi-bounce light enables state-of-the-art geometry reconstruction under strong indirect light.

Significance. If the results are robust, the paper would be the first to couple a physically-based time-resolved renderer with a neural radiance cache for inverting captured flash lidar data. The simulated geometry results are substantially better than the FWP++ baseline (MAE 8.45 vs 22.80, L1 depth 0.21 vs 0.47), and the method enables novel capabilities such as multi-view time-resolved relighting and direct/indirect decomposition. The authors provide a reproducible setup, including code and data, and their evaluation includes a well-motivated FWP++ baseline. However, the significance is tempered by the absence of quantitative geometry metrics on captured data and by the sensitivity of the physical model to source calibration, as discussed in the major comments.

major comments (4)
  1. [Section 5.2, Appendix D.1] The abstract and Section 5 claim state-of-the-art 3D reconstruction under strong indirect light from captured measurements, but Table 1 omits depth and normal metrics for captured scenes (the text states 'we omit depth and normal metrics for the captured experiments, as there is no ground truth reference'), and Figure 5 reports only qualitative normals. The point-source calibration in Appendix D.1 trilaterates the first ToF peak at checkerboard corners, implicitly assuming that peak is a direct-only path. Under strong indirect transport or with a finite SPAD footprint that includes a nearer surface, the first peak can be biased; since the rendering model is physically constrained, such a systematic error cannot be explained by lighting and will be absorbed into geometry, normals, and BRDF parameters. The simulated experiments use exact source parameters and cannot expose this fragility. Please add a sensitivity analysis (e.g., perturb the source position and intensity profile by realistic amounts and quantify the resulting depth and normal error on synthetic scenes) or provide an independent ground-truth measurement for at least one captured scene.
  2. [Eq. (6), Eq. (11)] The direct illumination term in Eq. (6) is written as a delta function times the source intensity divided by distance squared, with no visibility or transmittance factor; Eq. (11) uses the same unoccluded L_dir_i for the cache's direct component. For any surface point in shadow, the model predicts non-zero direct light, so the optimizer must compensate through geometry, normals, or materials. This contradicts the 'physically based' claim and could contribute to the captured-data degradation noted in Section 5.2. Please clarify whether visibility is handled implicitly by the density field (e.g., through transmittance along the source-to-point ray) or by the learned cache, and if not, include it or provide evidence (e.g., a synthetic scene with a hard shadow) that the omission does not affect geometry recovery.
  3. [Eq. (12)] The split-sum approximation factorizes the indirect outgoing radiance integral into a product of an integrated BRDF and an integrated incident radiance. This is exact only when one of the factors is constant; for a specular GGX BRDF and spatially and time-varying indirect illumination, the error is scene-dependent. The neural networks f_indir and L_indir_i,Omega can compensate in principle, but the paper does not quantify the approximation error or compare the cache against a full path-traced reference. Because the paper's central advantage is physical accuracy, please provide a validation of this approximation, such as a comparison of the split-sum cache output with a Monte Carlo reference on a synthetic transient scene.
  4. [Section 5.2, Table 1] The paper reports that on captured data FWP++ slightly outperforms the proposed method on view synthesis (PSNR 28.45 vs 27.39, LPIPS 0.32 vs 0.33, T-IOU 0.55 vs 0.54) and attributes this to calibration sensitivity. This admission means that the claimed state-of-the-art advantage on real data is not supported by the quantitative evidence; the contribution is primarily demonstrated in simulation. I recommend either strengthening the captured evaluation (e.g., with an independently scanned geometry or a calibration-error analysis) or revising the abstract and conclusion to state that state-of-the-art geometry recovery is demonstrated in simulation and that captured results are qualitative or concept demonstrations.
minor comments (5)
  1. [Appendix B.2] In the loss hyperparameters paragraph, 'Lgeom' should be 'lambda_geom' and 'disortion' should be 'distortion'.
  2. [Eq. (14)] The definition of alpha(L) as '||sum_tau L||^(-beta)' would benefit from a clarifying sentence that the exponent applies to the scalar sum of the discretized radiance vector.
  3. [Table C.2] Reporting four numeric metrics to two decimals without error bars or multiple seeds makes it impossible to assess the statistical significance of the differences; consider adding variance or a statistical test.
  4. [Section 5.2] The statement 'we recover more accurate geometry than FWP++' in the comparison paragraph is not supported by any quantitative metric on captured scenes; consider rephrasing to 'qualitatively more plausible for regions with indirect light'.
  5. [Abstract] The phrase 'the first system for physically based, neural inverse rendering from multi-viewpoint videos of propagating light' is a strong claim; if the authors are aware of any concurrent or unpublished work beyond GaNI, a brief discussion would help position the contribution.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the physics-based forward model is optimized against transient measurements, with independent synthetic ground truth and external baselines.

full rationale

This paper's derivation chain is self-contained. The forward model is the transient rendering equation (Eq. 5) with direct delta-source radiance (Eq. 6) and a radiance cache (Eq. 7); the optimization loss (Eq. 13) supervises the rendered transient radiance directly against the captured lidar histograms, with no evaluation metric or target geometry injected into the loss. The direct/indirect decomposition follows from the linearity of the rendering integral and is enforced as a physical fixed-point consistency constraint (Eq. 15), not by fitting the decomposition to a target. Simulated experiments use independent ground-truth geometry from a modified Mitsuba transient renderer and report MAE and L1 depth against that ground truth, so the geometry claim is externally validated. Captured experiments honestly omit depth and normal metrics because no ground truth exists (Section 5.2), and the paper explicitly flags sensitivity to source calibration and model mismatch as a limitation; these are correctness risks, not circular steps. The self-citations (T-NeRF, FWP, Attal et al.) refer to prior architectures and baselines; FWP++ is a controlled ablation of the cache without the physical model, and T-NeRF is an external baseline. No load-bearing claim reduces by construction to a fitted parameter, a self-citation chain, or a uniqueness theorem imported from the authors. Therefore the circularity score is 0.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The paper introduces no new physical entities; the time-resolved radiance cache is a learned representation, not a new medium or force. The main assumptions are the point-source calibration model, the BRDF class, and the split-sum factorization. The listed free parameters are training hyperparameters and per-dataset scales; the network weights themselves are fit to measurements as in all inverse rendering.

free parameters (4)
  • beta (tonemap prior exponent) = 1 (synthetic), 2 (cache, captured), 1 (physical, captured)
    Manually chosen in Eq. 14 to weight darker regions; changes the loss landscape and affects recovered geometry.
  • lambda_normals (normal loss weight) = varies per dataset
    Regularization weight that varies per dataset, chosen by hand; affects how tightly normals follow density gradients.
  • normalization_scale = 600
    Coordinate scaling for all captured scenes (Table D.1); a hand-chosen constant that defines the metric scale of the reconstructed geometry.
  • time_bin_count = dataset-dependent
    Number of discrete time bins in radiance vectors; chosen based on the lidar waveform and scene depth range.
assumptions (6)
  • domain assumption The rendering equation (Eq. 1) is the correct model for light transport in the scenes.
    Used as the basis of Eq. 5; assumes no participating media and geometric optics.
  • domain assumption The flash lidar source is a point light with known position and known directional intensity profile (Eq. 6, Appendix D.1).
    If the point-source model or calibration is inaccurate, the direct light term is biased.
  • domain assumption Disney-GGX BRDF (Eqs. 17-19) parameterizes all surface reflectance in the scenes.
    Materials that violate this (e.g., subsurface scattering, iridescence) are not modeled.
  • ad hoc to paper Split-sum approximation (Eq. 12) is an adequate factorization of the indirect light integral.
    This approximation is known to be lossy for glossy interreflections and is adopted to make the cache efficient.
  • domain assumption The time-resolved cache neural networks (Eqs. 9-12) can represent the full indirect transport within discretized time bins.
    Capacity and discretization limits could cause missed indirect paths.
  • standard math Background math: volume rendering quadrature (Eqs. 3-4, 7-8), NeRF density fields, Monte Carlo integration.
    Standard results assumed without proof.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Neural Inverse Rendering from Propagating Light." pith.science (2026). https://pith.science/paper/MZZZMF6Z

@misc{pith2026250605347,
  author       = {Pith},
  title        = {Pith review of: Neural Inverse Rendering from Propagating Light},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MZZZMF6Z}},
  note         = {Machine review of arXiv:2506.05347}
}
read the original abstract

We present the first system for physically based, neural inverse rendering from multi-viewpoint videos of propagating light. Our approach relies on a time-resolved extension of neural radiance caching -- a technique that accelerates inverse rendering by storing infinite-bounce radiance arriving at any point from any direction. The resulting model accurately accounts for direct and indirect light transport effects and, when applied to captured measurements from a flash lidar system, enables state-of-the-art 3D reconstruction in the presence of strong indirect light. Further, we demonstrate view synthesis of propagating light, automatic decomposition of captured measurements into direct and indirect components, as well as novel capabilities such as multi-view time-resolved relighting of captured scenes.

Figures

Figures reproduced from arXiv: 2506.05347 by the authors.

Figure 1
Figure 1. We introduce a method to model and invert multi-view, time-resolved measurements of propagating light from a flash lidar [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Method overview. (a) Our time-resolved renderer combines physically-based rendering for primary rays (left inset), and neural rendering for an indirect radiance cache along secondary rays (right inset). (b) The incident radiance Li at a sensor pixel is a function of the outgoing radiance Lo from each point x(t) along a sensor ray, which integrates incident direct light L dir i and indirect light L cache i from a pul… view at source ↗
Figure 3
Figure 3. Multi-view capture setup. An elevation arm controls [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Simulated results. Compared to the baselines our method recovers more accurate normals and similar or improved intensity [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Results on the captured dataset. Our method recovers more accurate normals compared to FWP++ (cols. 3, 5) due to its [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 7
Figure 7. Figure 7: Our approach recovers time-resolved videos of propa [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

100 extracted references · 80 canonical work pages

  1. [1]

    Bayesian time-of-flight for realtime shape, illumination and albedo.IEEE Trans

    Amit Adam, Christoph Dann, Omer Yair, Shai Mazor, and Sebastian Nowozin. Bayesian time-of-flight for realtime shape, illumination and albedo.IEEE Trans. Pattern Anal. Mach. Intell., 39(5):851–864, 2016. 3

    2016

  2. [2]

    Apple vision pro, 2024

    Apple Inc. Apple vision pro, 2024. Accessed: 2024-11-04. 1

    2024

  3. [3]

    T ¨oRF: Time-of-flight radiance fields for dynamic scene view synthesis.Proc

    Benjamin Attal, Eliot Laidlaw, Aaron Gokaslan, Changil Kim, Christian Richardt, James Tompkin, and Matthew O’Toole. T ¨oRF: Time-of-flight radiance fields for dynamic scene view synthesis.Proc. NeurIPS, 2021. 2, 4, 14

    2021

  4. [4]

    Flash cache: Reducing bias in radiance cache based inverse rendering

    Benjamin Attal, Dor Verbin, Ben Mildenhall, Peter Hedman, Jonathan T Barron, Matthew O’Toole, and Pratul P Srini- vasan. Flash cache: Reducing bias in radiance cache based inverse rendering. InProc. ECCV, 2024. 2, 3, 5, 13

    2024

  5. [5]

    Zip-NeRF: Anti-aliased grid- based neural radiance fields

    Jonathan T Barron, Ben Mildenhall, Dor Verbin, Pratul P Srinivasan, and Peter Hedman. Zip-NeRF: Anti-aliased grid- based neural radiance fields. InProc. ICCV, 2023. 5, 13

    2023

  6. [6]

    Rendering resources, 2016

    Benedikt Bitterli. Rendering resources, 2016. https://benedikt-bitterli.me/resources/. 6

    2016

  7. [7]

    Barron, and Hendrik P.A

    Mark Boss, Varun Jampani, Raphael Braun, Ce Liu, Jonathan T. Barron, and Hendrik P.A. Lensch. Neural-pil: Neural pre-integrated lighting for reflectance decomposition. InProc. NeurIPS, 2021. 3

    2021

  8. [8]

    Barron, Hendrik P.A

    Mark Boss, Andreas Engelhardt, Abhishek Kar, Yuanzhen Li, Deqing Sun, Jonathan T. Barron, Hendrik P.A. Lensch, and Varun Jampani. SAMURAI: Shape And Material from Unconstrained Real-world Arbitrary Image collections. In Proc. NeurIPS, 2022. 3

    2022

Show all 100 references
  1. [9]

    Physically-based shading at Disney

    Brent Burley and Walt Disney Animation Studios. Physically-based shading at Disney. InACM SIGGRAPH Courses, 2012. 4, 13

    2012

  2. [10]

    Learning implicit fields for generative shape modeling

    Zhiqin Chen and Hao Zhang. Learning implicit fields for generative shape modeling. InProc. CVPR, 2019. 2, 3

    2019

  3. [11]

    Rendering synthetic objects into real scenes: bridging traditional and image-based graphics with global il- lumination and high dynamic range photography

    Paul Debevec. Rendering synthetic objects into real scenes: bridging traditional and image-based graphics with global il- lumination and high dynamic range photography. InProc. SIGGRAPH, 1998. 3

    1998

  4. [12]

    Depth-supervised NeRF: Fewer views and faster training for free

    Kangle Deng, Andrew Liu, Jun-Yan Zhu, and Deva Ra- manan. Depth-supervised NeRF: Fewer views and faster training for free. InProc. CVPR, 2022. 2

    2022

  5. [13]

    CRC Press, 2017

    Pinliang Dong and Qi Chen.LiDAR remote sensing and ap- plications. CRC Press, 2017. 1

    2017

  6. [14]

    V ol- ume rendering.ACM SIGGRAPH, 22(4):65–74, 1988

    Robert A Drebin, Loren Carpenter, and Pat Hanrahan. V ol- ume rendering.ACM SIGGRAPH, 22(4):65–74, 1988. 3

    1988

  7. [15]

    Non-line-of-sight imaging.Nat

    Daniele Faccio, Andreas Velten, and Gordon Wetzstein. Non-line-of-sight imaging.Nat. Rev. Phys., 2(6):318–327,

  8. [16]

    Single-photon sensitive light-in-fight imaging.Nat

    Genevieve Gariepy, Nikola Krstaji ´c, Robert Henderson, Chunyong Li, Robert R Thomson, Gerald S Buller, Bar- mak Heshmat, Ramesh Raskar, Jonathan Leach, and Daniele Faccio. Single-photon sensitive light-in-fight imaging.Nat. Commun., 6(1):6021, 2015. 2

    2015

  9. [17]

    An eval- uation of computational imaging techniques for heteroge- neous inverse scattering

    Ioannis Gkioulekas, Anat Levin, and Todd Zickler. An eval- uation of computational imaging techniques for heteroge- neous inverse scattering. InProc. ECCV, 2016. 4

    2016

  10. [18]

    Phasor imaging: A generalization of correlation- based time-of-flight imaging.ACM Trans

    Mohit Gupta, Shree K Nayar, Matthias B Hullin, and Jaime Martin. Phasor imaging: A generalization of correlation- based time-of-flight imaging.ACM Trans. Graph., 34(5): 1–18, 2015. 2, 3

    2015

  11. [19]

    Inverse global illumination using a neural radiometric prior

    Saeed Hadadan, Geng Lin, Jan Nov ´ak, Fabrice Rousselle, and Matthias Zwicker. Inverse global illumination using a neural radiometric prior. InProc. SIGGRAPH, pages 1–11,

  12. [20]

    Low-budget transient imaging using photonic mixer devices.ACM Trans

    Felix Heide, Matthias B Hullin, James Gregson, and Wolf- gang Heidrich. Low-budget transient imaging using photonic mixer devices.ACM Trans. Graph., 32(4):1–10, 2013. 2

    2013

  13. [21]

    Neural lidar fields for novel view synthesis

    Shengyu Huang, Zan Gojcic, Zian Wang, Francis Williams, Yoni Kasten, Sanja Fidler, Konrad Schindler, and Or Litany. Neural lidar fields for novel view synthesis. InProc. ICCV,

  14. [22]

    Woodhead Publishing, 2020

    Andrew S Huntington.InGaAs avalanche photodiodes for ranging and Lidar. Woodhead Publishing, 2020. 2

    2020

  15. [23]

    Neural active structure-from-motion in dark and textureless environment.Proc

    Kazuto Ichimaru, Diego Thomas, Takafumi Iwaguchi, and Hiroshi Kawasaki. Neural active structure-from-motion in dark and textureless environment.Proc. ACCV, 2024. 2

    2024

  16. [24]

    Non-line-of- sight reconstruction using efficient transient rendering.ACM Trans

    Julian Iseringhausen and Matthias B Hullin. Non-line-of- sight reconstruction using efficient transient rendering.ACM Trans. Graph., 39(1):1–14, 2020. 2

    2020

  17. [25]

    Wenzel Jakob, S ´ebastien Speierer, Nicolas Roussel, and De- lio Vicini. Dr. jit: A just-in-time compiler for differentiable rendering.ACM Trans. Graph., 41(4):1–19, 2022. 1, 3

    2022

  18. [26]

    Bidirectional render- ing of vector light transport

    Adrian Jarabo and Victor Arellano. Bidirectional render- ing of vector light transport. InComputer Graphics Forum, pages 96–105. Wiley Online Library, 2018. 3

    2018

  19. [27]

    A framework for tran- sient rendering.ACM Trans

    Adrian Jarabo, Julio Marco, Adolfo Munoz, Raul Buisan, Wojciech Jarosz, and Diego Gutierrez. A framework for tran- sient rendering.ACM Trans. Graph., 33(6):1–10, 2014. 3

    2014

  20. [28]

    TensoIR: Tensorial inverse rendering

    Haian Jin, Isabella Liu, Peijia Xu, Xiaoshuai Zhang, Song- fang Han, Sai Bi, Xiaowei Zhou, Zexiang Xu, and Hao Su. TensoIR: Tensorial inverse rendering. InProc. CVPR, 2023. 2, 3, 14

    2023

  21. [29]

    The rendering equation

    James T Kajiya. The rendering equation. InProc. SIG- GRAPH, 1986. 2, 3

    1986

  22. [30]

    Real shading in Unreal Engine 4.ACM SIG- GRAPH Courses, 2013

    Brian Karis. Real shading in Unreal Engine 4.ACM SIG- GRAPH Courses, 2013. 5

    2013

  23. [31]

    First-photon imaging.Science, 343(6166): 58–61, 2014

    Ahmed Kirmani, Dheera Venkatraman, Dongeek Shin, An- drea Colac ¸o, Franco NC Wong, Jeffrey H Shapiro, and Vivek K Goyal. First-photon imaging.Science, 343(6166): 58–61, 2014. 2

    2014

  24. [32]

    PlatoNeRF: 3D reconstruction in Plato’s cave via single-view two-bounce lidar

    Tzofi Klinghoffer, Xiaoyu Xiang, Siddharth Somasundaram, Yuchen Fan, Christian Richardt, Ramesh Raskar, and Rakesh Ranjan. PlatoNeRF: 3D reconstruction in Plato’s cave via single-view two-bounce lidar. InProc. CVPR, 2024. 2, 3

    2024

  25. [33]

    Optical ranging system employing a high power injection laser diode.IEEE Trans

    Walter Koechner. Optical ranging system employing a high power injection laser diode.IEEE Trans. Aerosp. Electron. Syst., AES-4(1):81–91, 1968. 2

    1968

  26. [34]

    Time-of-flight camera—an introduction.Tech- nical white paper, 2014

    Larry Li et al. Time-of-flight camera—an introduction.Tech- nical white paper, 2014. 2

    2014

  27. [35]

    Differentiable monte carlo ray tracing through edge sampling.ACM Trans

    Tzu-Mao Li, Miika Aittala, Fr ´edo Durand, and Jaakko Lehti- nen. Differentiable monte carlo ray tracing through edge sampling.ACM Trans. Graph., 37(6):1–11, 2018. 2, 3

    2018

  28. [36]

    Lidar for autonomous driving: The principles, challenges, and trends for automo- tive lidar and perception systems.IEEE Signal Process

    You Li and Javier Ibanez-Guzman. Lidar for autonomous driving: The principles, challenges, and trends for automo- tive lidar and perception systems.IEEE Signal Process. Mag., 37(4):50–61, 2020. 1

    2020

  29. [37]

    Three-dimensional imaging through scattering media based on confocal diffuse tomography.Nat

    David B Lindell and Gordon Wetzstein. Three-dimensional imaging through scattering media based on confocal diffuse tomography.Nat. Commun., 11(1):4517, 2020. 3

    2020

  30. [38]

    Towards transient imaging at interactive rates with single- photon detectors

    David B Lindell, Matthew O’Toole, and Gordon Wetzstein. Towards transient imaging at interactive rates with single- photon detectors. InProc. ICCP, 2018. 2

    2018

  31. [39]

    Wave-based non-line-of-sight imaging using fast fk migra- tion.ACM Trans

    David B Lindell, Gordon Wetzstein, and Matthew O’Toole. Wave-based non-line-of-sight imaging using fast fk migra- tion.ACM Trans. Graph., 38(4):1–13, 2019. 3, 8

    2019

  32. [40]

    NeRF as a non-distant environment emitter in physics- based inverse rendering

    Jingwang Ling, Ruihan Yu, Feng Xu, Chun Du, and Shuang Zhao. NeRF as a non-distant environment emitter in physics- based inverse rendering. InProc. SIGGRAPH, 2024. 3

    2024

  33. [41]

    Non-line-of-sight imaging using phasor-field virtual wave optics.Nature, 572(7771): 620–623, 2019

    Xiaochun Liu, Ib ´on Guill ´en, Marco La Manna, Ji Hyun Nam, Syed Azer Reza, Toan Huu Le, Adrian Jarabo, Diego Gutierrez, and Andreas Velten. Non-line-of-sight imaging using phasor-field virtual wave optics.Nature, 572(7771): 620–623, 2019. 3

    2019

  34. [42]

    NERO: Neural geometry and BRDF reconstruction of reflec- tive objects from multiview images.ACM Trans

    Yuan Liu, Peng Wang, Cheng Lin, Xiaoxiao Long, Jiepeng Wang, Lingjie Liu, Taku Komura, and Wenping Wang. NERO: Neural geometry and BRDF reconstruction of reflec- tive objects from multiview images.ACM Trans. Graph., 42 (4):1–22, 2023. 2, 3, 13

    2023

  35. [43]

    Neural vol- umes: learning dynamic renderable volumes from images

    Stephen Lombardi, Tomas Simon, Jason Saragih, Gabriel Schwartz, Andreas Lehrmann, and Yaser Sheikh. Neural vol- umes: learning dynamic renderable volumes from images. ACM Trans. Graph., 38(4):1–14, 2019. 3

    2019

  36. [44]

    Reparameterizing discontinuous integrands for differentiable rendering.ACM Trans

    Guillaume Loubet, Nicolas Holzschuch, and Wenzel Jakob. Reparameterizing discontinuous integrands for differentiable rendering.ACM Trans. Graph., 38(6):1–14, 2019. 2

    2019

  37. [45]

    Transien- tangelo: Few-viewpoint surface reconstruction using single- photon lidar.Proc

    Weihan Luo, Anagh Malik, and David B Lindell. Transien- tangelo: Few-viewpoint surface reconstruction using single- photon lidar.Proc. WACV, 2025. 2

    2025

  38. [46]

    Neural microfacet fields for inverse render- ing

    Alexander Mai, Dor Verbin, Falko Kuester, and Sara Fridovich-Keil. Neural microfacet fields for inverse render- ing. InProc. ICCV, 2023. 3, 8

    2023

  39. [47]

    Transient neural radiance fields for lidar view synthesis and 3D reconstruc- tion

    Anagh Malik, Parsa Mirdehghan, Sotiris Nousias, Kiri- akos N Kutulakos, and David B Lindell. Transient neural radiance fields for lidar view synthesis and 3D reconstruc- tion. InProc. NeurIPS, 2023. 2, 4, 6, 14, 15, 17, 18

    2023

  40. [48]

    Kutulakos, and David B

    Anagh Malik, Noah Juravsky, Ryan Po, Gordon Wetzstein, Kiriakos N. Kutulakos, and David B. Lindell. Flying with photons: Rendering novel views of propagating light. In Proc. ECCV, 2024. 2, 5, 6, 15, 17, 18, 19

    2024

  41. [49]

    Camera calibrator app.https : / / www

    Mathworks. Camera calibrator app.https : / / www . mathworks . com / help / vision / ref / cameracalibrator-app.html, 2020. 15

    2020

  42. [50]

    Optical models for direct volume rendering

    Nelson Max. Optical models for direct volume rendering. IEEE Trans. Vis. Comput. Graph., 1(2):99–108, 1995. 3

    1995

  43. [51]

    NeRF: Representing scenes as neural radiance fields for view syn- thesis.Commun

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. NeRF: Representing scenes as neural radiance fields for view syn- thesis.Commun. ACM, 65(1):99–106, 2021. 2, 3

    2021

  44. [52]

    Srinivasan, and Jonathan T

    Ben Mildenhall, Peter Hedman, Ricardo Martin-Brualla, Pratul P. Srinivasan, and Jonathan T. Barron. NeRF in the dark: High dynamic range view synthesis from noisy raw images. InProc. CVPR, 2022. 5

    2022

  45. [53]

    3D Gaussian ray tracing: Fast tracing of particle scenes.Proc

    Nicolas Moenne-Loccoz, Ashkan Mirzaei, Or Perel, Ric- cardo de Lutio, Janick Martinez Esturo, Gavriel State, Sanja Fidler, Nicholas Sharp, and Zan Gojcic. 3D Gaussian ray tracing: Fast tracing of particle scenes.Proc. SIGGRAPH Asia, 2024. 8

    2024

  46. [54]

    Towards 3D vision with low-cost single-photon cameras

    Fangzhou Mu, Carter Sifferman, Sacha Jungerman, Yiquan Li, Mark Han, Michael Gleicher, Mohit Gupta, and Yin Li. Towards 3D vision with low-cost single-photon cameras. In Proc. CVPR, 2024. 2

    2024

  47. [55]

    Instant neural graphics primitives with a multires- olution hash encoding.ACM Trans

    Thomas M ¨uller, Alex Evans, Christoph Schied, and Alexan- der Keller. Instant neural graphics primitives with a multires- olution hash encoding.ACM Trans. Graph. (SIGGRAPH), 41(4):1–15, 2022. 5, 6

    2022

  48. [56]

    Single view reflectance capture using mul- tiplexed scattering and time-of-flight imaging.ACM Trans

    Nikhil Naik, Shuang Zhao, Andreas Velten, Ramesh Raskar, and Kavita Bala. Single view reflectance capture using mul- tiplexed scattering and time-of-flight imaging.ACM Trans. Graph., 30(6):1–10, 2011. 2

    2011

  49. [57]

    A light transport model for mitigating multipath interference in time-of-flight sensors

    Nikhil Naik, Achuta Kadambi, Christoph Rhemann, Shahram Izadi, Ramesh Raskar, and Sing Bing Kang. A light transport model for mitigating multipath interference in time-of-flight sensors. InProc. CVPR, pages 73–81, 2015. 3

    2015

  50. [58]

    Mitsuba 2: A retargetable forward and inverse renderer.ACM Trans

    Merlin Nimier-David, Delio Vicini, Tizian Zeltner, and Wen- zel Jakob. Mitsuba 2: A retargetable forward and inverse renderer.ACM Trans. Graph., 38(6):1–17, 2019. 2, 3, 6

    2019

  51. [59]

    Flowed time of flight radiance fields

    Mikhail Okunev, Marc Mapeke, Benjamin Attal, Christian Richardt, Matthew O’Toole, and James Tompkin. Flowed time of flight radiance fields. InProc. ECCV, 2024. 2

    2024

  52. [60]

    Temporal frequency probing for 5D transient analysis of global light transport.ACM Trans

    Matthew O’Toole, Felix Heide, Lei Xiao, Matthias B Hullin, Wolfgang Heidrich, and Kiriakos N Kutulakos. Temporal frequency probing for 5D transient analysis of global light transport.ACM Trans. Graph., 33(4):1–11, 2014. 2, 3

    2014

  53. [61]

    Reconstructing transient images from single-photon sensors

    Matthew O’Toole, Felix Heide, David B Lindell, Kai Zang, Steven Diamond, and Gordon Wetzstein. Reconstructing transient images from single-photon sensors. InProc. CVPR,

  54. [62]

    Confocal non-line-of-sight imaging based on the light-cone transform.Nature, 555(7696):338–341, 2018

    Matthew O’Toole, David B Lindell, and Gordon Wetzstein. Confocal non-line-of-sight imaging based on the light-cone transform.Nature, 555(7696):338–341, 2018. 2

    2018

  55. [63]

    Ellipsoidal path connections for time-gated ren- dering.ACM Trans

    Adithya Pediredla, Ashok Veeraraghavan, and Ioannis Gkioulekas. Ellipsoidal path connections for time-gated ren- dering.ACM Trans. Graph., 38(4):1–12, 2019. 3

    2019

  56. [64]

    Path tracing estima- tors for refractive radiative transfer.ACM Trans

    Adithya Pediredla, Yasin Karimi Chalmiani, Mat- teo Giuseppe Scopelliti, Maysamreza Chamanzar, Srinivasa Narasimhan, and Ioannis Gkioulekas. Path tracing estima- tors for refractive radiative transfer.ACM Trans. Graph., 39 (6):1–15, 2020. 3

    2020

  57. [65]

    MIT Press, 2023

    Matt Pharr, Wenzel Jakob, and Greg Humphreys.Physi- cally based rendering: From theory to implementation. MIT Press, 2023. 1, 3, 13

    2023

  58. [66]

    A signal-processing framework for inverse rendering

    Ravi Ramamoorthi and Pat Hanrahan. A signal-processing framework for inverse rendering. InProc. SIGGRAPH, 2001. 3

    2001

  59. [67]

    Gated fields: Learning scene re- construction from gated videos

    Andrea Ramazzina, Stefanie Walz, Pragyan Dahal, Mario Bijelic, and Felix Heide. Gated fields: Learning scene re- construction from gated videos. InProc. CVPR, 2024. 2

    2024

  60. [68]

    5D time-light transport matrix: What can we reason about scene properties? Tech- nical report, MIT, 2008

    Raskar Ramesh and James Davis. 5D time-light transport matrix: What can we reason about scene properties? Tech- nical report, MIT, 2008. 3

    2008

  61. [69]

    Seeing around corners with edge- resolved transient imaging.Nat

    Joshua Rapp, Charles Saunders, Juli ´an Tachella, John Murray-Bruce, Yoann Altmann, Jean-Yves Tourneret, Stephen McLaughlin, Robin MA Dawson, Franco NC Wong, and Vivek K Goyal. Seeing around corners with edge- resolved transient imaging.Nat. Commun., 11(1):5929,

  62. [70]

    Advances in single-photon lidar for autonomous vehicles: Working principles, chal- lenges, and recent advances.IEEE Signal Process

    Joshua Rapp, Julian Tachella, Yoann Altmann, Stephen McLaughlin, and Vivek K Goyal. Advances in single-photon lidar for autonomous vehicles: Working principles, chal- lenges, and recent advances.IEEE Signal Process. Mag., 37(4):62–71, 2020. 2

    2020

  63. [71]

    Urban radiance fields

    Konstantinos Rematas, Andrew Liu, Pratul P Srini- vasan, Jonathan T Barron, Andrea Tagliasacchi, Thomas Funkhouser, and Vittorio Ferrari. Urban radiance fields. In Proc. CVPR, 2022. 2

    2022

  64. [72]

    Non-line-of-sight transient rendering.Comput

    Diego Royo, Jorge Garc ´ıa, Adolfo Mu ˜noz, and Adrian Jarabo. Non-line-of-sight transient rendering.Comput. Graph., 107:84–92, 2022. 6

    2022

  65. [73]

    Virtual mirrors: Non-line-of-sight imaging beyond the third bounce.ACM Trans

    Diego Royo, Talha Sultan, Adolfo Mu ˜noz, Khadijeh Masumnia-Bisheh, Eric Brandt, Diego Gutierrez, Andreas Velten, and Julio Marco. Virtual mirrors: Non-line-of-sight imaging beyond the third bounce.ACM Trans. Graph., 42 (4):1–15, 2023. 3

    2023

  66. [74]

    Structure- from-motion revisited

    Johannes L Schonberger and Jan-Michael Frahm. Structure- from-motion revisited. InProc. CVPR, 2016. 15

    2016

  67. [75]

    Mapping the world in 3D.Nat

    Brent Schwarz. Mapping the world in 3D.Nat. Photonics, 4 (7):429–430, 2010. 1

    2010

  68. [76]

    Non-line-of-sight snapshots and background mapping with an active corner camera.Nat

    Sheila Seidel, Hoover Rueda-Chac ´on, Iris Cusini, Federica Villa, Franco Zappa, Christopher Yu, and Vivek K Goyal. Non-line-of-sight snapshots and background mapping with an active corner camera.Nat. Commun., 14(1):3677, 2023. 3

    2023

  69. [77]

    Neural fields for structured lighting

    Aarrushi Shandilya, Benjamin Attal, Christian Richardt, James Tompkin, and Matthew O’Toole. Neural fields for structured lighting. InProc. ICCV, 2023. 2

    2023

  70. [78]

    Photon-efficient computational 3-d and re- flectivity imaging with single-photon detectors.IEEE Trans

    Dongeek Shin, Ahmed Kirmani, Vivek K Goyal, and Jef- frey H Shapiro. Photon-efficient computational 3-d and re- flectivity imaging with single-photon detectors.IEEE Trans. Comput. Imaging, 1(2):112–125, 2015. 2

    2015

  71. [79]

    NeRV: Neural reflectance and visibility fields for relighting and view synthesis

    Pratul P Srinivasan, Boyang Deng, Xiuming Zhang, Matthew Tancik, Ben Mildenhall, and Jonathan T Barron. NeRV: Neural reflectance and visibility fields for relighting and view synthesis. InProc. CVPR, 2021. 2, 3

    2021

  72. [80]

    V olume rendering digest (for NeRF).arXiv preprint, 2022

    Andrea Tagliasacchi and Ben Mildenhall. V olume rendering digest (for NeRF).arXiv preprint, 2022. 3

    2022

  73. [81]

    Shape and reflectance from two-bounce light transients

    Chia-Yin Tsai, Ashok Veeraraghavan, and Aswin C Sankara- narayanan. Shape and reflectance from two-bounce light transients. InProc. ICCP, pages 1–10. IEEE, 2016. 3

    2016

  74. [82]

    Beyond volumetric albedo–a surface optimiza- tion framework for non-line-of-sight imaging

    Chia-Yin Tsai, Aswin C Sankaranarayanan, and Ioannis Gkioulekas. Beyond volumetric albedo–a surface optimiza- tion framework for non-line-of-sight imaging. InProc. CVPR, 2019. 2

    2019

  75. [83]

    Stanford University, 1998

    Eric Veach.Robust Monte Carlo methods for light transport simulation. Stanford University, 1998. 3

    1998

  76. [84]

    Femto- photography: capturing and visualizing the propagation of light.ACM Trans

    Andreas Velten, Di Wu, Adrian Jarabo, Belen Masia, Christopher Barsi, Chinmaya Joshi, Everett Lawson, Moungi Bawendi, Diego Gutierrez, and Ramesh Raskar. Femto- photography: capturing and visualizing the propagation of light.ACM Trans. Graph., 32(4):1–8, 2013. 2

    2013

  77. [85]

    Barron, and Pratul P

    Dor Verbin, Peter Hedman, Ben Mildenhall, Todd Zickler, Jonathan T. Barron, and Pratul P. Srinivasan. Ref-NeRF: Structured view-dependent appearance for neural radiance fields. InProc. CVPR, 2022. 5, 14

    2022

  78. [86]

    Eclipse: Disam- biguating illumination and materials using unintended shad- ows

    Dor Verbin, Ben Mildenhall, Peter Hedman, Jonathan T Bar- ron, Todd Zickler, and Pratul P Srinivasan. Eclipse: Disam- biguating illumination and materials using unintended shad- ows. InProc. CVPR, 2024. 3

    2024

  79. [87]

    A ray tracing solution for diffuse interreflection

    Gregory J Ward, Francis M Rubinstein, and Robert D Clear. A ray tracing solution for diffuse interreflection. InProc. SIGGRAPH, 1988. 2, 3

    1988

  80. [88]

    Decom- posing global light transport using time of flight imaging.Int

    Di Wu, Andreas Velten, Matthew O’Toole, Belen Masia, Amit Agrawal, Qionghai Dai, and Ramesh Raskar. Decom- posing global light transport using time of flight imaging.Int. J. Comput. Vis., 107:123–138, 2014. 2

    2014

  81. [89]

    Nefii: Inverse rendering for reflectance decomposition with near-field indirect illumina- tion

    Haoqian Wu, Zhipeng Hu, Lincheng Li, Yongqiang Zhang, Changjie Fan, and Xin Yu. Nefii: Inverse rendering for reflectance decomposition with near-field indirect illumina- tion. InProc. CVPR, pages 4295–4304, 2023. 3

    2023

  82. [90]

    GaNI: Global and near field illumination aware neural inverse rendering.arXiv,

    Jiaye Wu, Saeed Hadadan, Geng Lin, Matthias Zwicker, David Jacobs, and Roni Sengupta. GaNI: Global and near field illumination aware neural inverse rendering.arXiv,

  83. [91]

    Differentiable time-gated rendering.ACM Trans

    Lifan Wu, Guangyan Cai, Ravi Ramamoorthi, and Shuang Zhao. Differentiable time-gated rendering.ACM Trans. Graph., 40(6):1–16, 2021. 2, 3

    2021

  84. [92]

    A theory of Fermat paths for non-line- of-sight shape reconstruction

    Shumian Xin, Sotiris Nousias, Kiriakos N Kutulakos, Aswin C Sankaranarayanan, Srinivasa G Narasimhan, and Ioannis Gkioulekas. A theory of Fermat paths for non-line- of-sight shape reconstruction. InProc. CVPR, 2019. 2, 3

    2019

  85. [93]

    NEILF: Neural incident light field for physically-based ma- terial estimation

    Yao Yao, Jingyang Zhang, Jingbo Liu, Yihang Qu, Tian Fang, David McKinnon, Yanghai Tsin, and Long Quan. NEILF: Neural incident light field for physically-based ma- terial estimation. InProc. ECCV, 2022. 3

    2022

  86. [94]

    Shinyoung Yi, Donggun Kim, Kiseok Choi, Adrian Jarabo, Diego Gutierrez, and Min H. Kim. Differentiable transient rendering.ACM Trans. Graph., 40(6), 2021. 2, 3

    2021

  87. [95]

    Inverse global illumination: Recovering reflectance models of real scenes from photographs

    Yizhou Yu, Paul Debevec, Jitendra Malik, and Tim Hawkins. Inverse global illumination: Recovering reflectance models of real scenes from photographs. InProc. SIGGRAPH, 1999. 3

    1999

  88. [96]

    NeRF-lidar: Generating realistic lidar point clouds with neu- ral radiance fields

    Junge Zhang, Feihu Zhang, Shaochen Kuang, and Li Zhang. NeRF-lidar: Generating realistic lidar point clouds with neu- ral radiance fields. InProc. AAAI, 2024. 2

    2024

  89. [97]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InProc. CVPR, 2018. 6

    2018

  90. [98]

    Ner- factor: Neural factorization of shape and reflectance under an unknown illumination.ACM Trans

    Xiuming Zhang, Pratul P Srinivasan, Boyang Deng, Paul De- bevec, William T Freeman, and Jonathan T Barron. Ner- factor: Neural factorization of shape and reflectance under an unknown illumination.ACM Trans. Graph., 40(6):1–18,

  91. [99]

    pro- posal

    Yuanqing Zhang, Jiaming Sun, Xingyi He, Huan Fu, Rongfei Jia, and Xiaowei Zhou. Modeling indirect illumination for inverse rendering. InProc. CVPR, 2022. 2, 3 A. Appendix Overview We provide additional implementation details related to the architecture of our model, optimizati...

    2022

  92. [100]

    For our scenes, we assume our light sources are point sources, calibrated so that their location is known with re- spect to the scene

    with a scene including a checkerboard so that radial camera pose translation can be scaled by matching the re- construction to the board’s known geometry. For our scenes, we assume our light sources are point sources, calibrated so that their location is known with re- spect t...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.