Pith. sign in

REVIEW 3 major objections 5 minor 34 references

Fast Non-Line-of-Sight Transient Data Simulation and an Open Benchmark Dataset

T0 review · 3 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 single depth map accurately simulates hidden-scene transient data for non-line-of-sight imaging.

desk verdict A practical simulator paper whose discrete PSF contradicts its own LCT derivation, so the dataset and benchmark inherit the error. read the letter →

arxiv 2506.03747 v1 pith:TSRNSPOB submitted 2025-06-04 physics.optics

Yingjie Shi , Jinye Miao , Taotao Qin , Fuyao Cai , Yi Wei , Lingfeng Liu , Tongyao Li , Chenyang Wu
show 5 more authors
Huan Liang Yuyang Yin Lianfa Bai Enlai Guo Jing Han
This is my paper
classification physics.optics
keywords non-line-of-sightimagingtransientsimulationlight-conetransformdepthmapbenchmarkdatasetShapeNettime-correlatedsingle-photoncountingcomputational
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 establishes that the transient signals a non-line-of-sight (NLOS) camera would record can be simulated directly from a single depth map and an optional albedo map, using a discretized light-cone transform accelerated by the fast Fourier transform. This avoids expensive path tracing and requires no hardware or complex scene files, only a few tunable parameters for wall size, stand-off distance, time resolution, and noise. The authors validate the approach by reconstructing the hidden geometry from their simulated data with four classical algorithms, and they release a ShapeNet-derived dataset covering seven object classes as a unified benchmark. If the forward model is faithful, this gives the NLOS community a fast, accessible way to generate training data and test reconstruction algorithms across scales from meters to centimeters.

What carries the argument

The central mechanism is the discretized light-cone transform (LCT) expressed as $\tau = R_t^{-1}\{ R_z(p) * \text{PSF} \}$, where $p$ is the albedo-weighted depth map volume, $R_z$ resamples the depth axis via a square-root mapping with amplitude normalization, $R_t^{-1}$ rescales the temporal axis by $v^{3/2}$, and PSF is a Boolean tensor representing the conical wavefront on a uniform grid. A slope parameter ties lateral spatial distances to axial time-depth coordinates, and multilevel binary down-sampling keeps the spatio-temporal coupling scale aligned with resolution, allowing the whole operation to run as a single FFT-accelerated convolution.

What would settle it

Compare the simulator's output against real confocal NLOS measurements taken on a laboratory scene whose geometry and surface reflectance are independently calibrated; if the simulated temporal profile or the shape reconstructed from it diverges beyond the modeled noise, the forward model is not adequate.

Watch

Extended reading notes

Core claim

The paper demonstrates that the confocal NLOS transient measurement can be reduced to a three-dimensional convolution between a volume built from a depth map and a conical point-spread function (PSF), with two resampling operators that fold the depth axis and then unfold the temporal axis while conserving energy. The entire simulation pipeline takes a depth and albedo map as input, maps them to a voxel field, applies the FFT-based convolution, and then adds temporal jitter and Poisson noise, all with a parameter interface for system geometry and detector properties. Systematic experiments on seven ShapeNet categories show that the simulated transients, reconstructed by LCT, phasor field, f-k migration, and filtered backprojection, recover the original hidden shapes with high PSNR and SSIM, thereby validating the physical consistency of the synthetic data.

Load-bearing premise

That a single orthographic depth map with uniform Lambertian albedo, processed through the confocal light-cone transform, adequately represents what a real NLOS detector would record.

Editorial extensions

If this is right

  • Researchers can generate custom NLOS transient datasets on demand for any depth map by adjusting wall size, stand-off distance, time bin, jitter, and noise, eliminating the need to modify shaders or scene files.
  • Supervised deep-learning models for NLOS reconstruction can be trained on a large, physically consistent corpus covering diverse object classes and noise conditions, potentially enabling cross-hardware generalization.
  • The released benchmark tables give a unified, reproducible baseline for comparing new reconstruction algorithms against LCT, phasor field, f-k migration, and filtered backprojection on the same data.
  • System design can be optimized in software by sweeping detector time resolution, scan area, and noise parameters before committing to expensive hardware.
  • Because the framework lowers the storage cost from a full volume to a depth map, large-scale batch processing of synthetic transients becomes practical for training purposes.

Reading between the lines

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

  • The same discretized convolution could be inverted analytically to yield a fast reconstruction algorithm that is directly matched to the forward model, potentially outperforming the four generic baselines on the simulated data.
  • The uniform Lambertian albedo assumption could be relaxed by extending the depth map input to a per-pixel BRDF or a small set of albedo channels, allowing simulation of more realistic surface finishes at little extra cost.
  • The benchmark's synthetic-to-real gap could be tested by fine-tuning a network on these simulated transients and evaluating it on real experimental NLOS measurements; such a transfer test would be a natural next validation.
  • The parameterized noise model (Gaussian jitter plus Poisson corruptions) could be used to study how reconstruction accuracy degrades as a function of time resolution or photon count, yielding design guidance for future SPAD systems.
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

3 major / 5 minor

Summary. The manuscript proposes a fast NLOS transient simulator built on the light-cone transform (LCT). A depth map and optional albedo map are converted into a voxel field, resampled along depth, convolved in the Fourier domain with a discretized point-spread function, and then degraded by temporal Gaussian blur and Poisson noise. The authors release synthetic transients for seven ShapeNet object classes and report reconstruction SSIM/PSNR for LCT, phasor-field (PF), f-k migration (FK), and filtered backprojection (FBP). The central claim is that the simulator generates physically consistent transient measurements and that the released dataset provides a reproducible benchmark for NLOS reconstruction algorithms.

Significance. If the implementation were correct, the paper would offer a useful low-barrier tool: it requires only a depth map, exposes a compact parameter interface, and uses FFT acceleration, which is a practical advantage over path-tracing simulators. The release of a dataset and a comparison of four classical reconstruction algorithms is also potentially valuable for reproducible research. However, the physical-consistency claim currently rests on an internally generated benchmark with no independent validation, and the discrete PSF in the implementation does not match the continuous LCT derivation. The central contribution is therefore not yet established. With the kernel corrected and validation against an independent renderer or real data added, the tool and dataset could be a helpful resource, but the present version overstates its physical fidelity.

major comments (3)
  1. [Section 2 / Algorithm 1] The implemented PSF is internally inconsistent with the continuous LCT model. The continuous derivation uses u=z^2 and v=(ct/2)^2 and yields a kernel delta((x'-x)^2+(y'-y)^2+u-v), i.e. z^2 + (x'-x)^2 + (y'-y)^2 = (ct/2)^2. Algorithm 1, step 1.2, however, constructs PSF(x,y,z)=delta_K(|z - (16w^2/(c^2 T_max^2))(x^2+y^2)|, 0), which is a paraboloid in linear depth. The text's justification, 'Squaring this expression yields the form above', is algebraically incorrect: squaring z=(4/slope)r gives z^2=(4/slope)^2 r^2, not z=(4/slope)^2 r^2. Consequently, off-axis scene points are assigned temporal depths that do not satisfy the physical round-trip relation t=2*sqrt(r^2+z^2)/c. Since the released dataset and the scores in Tables 1-2 are produced with this kernel, the reported reconstructions validate only the self-consistency of the implementation, not the LCT forward model.
  2. [Section 3 / Tables 1-2] The experimental validation is circular. The transient data are generated with the LCT-based forward model, and the compared algorithms—especially LCT but also PF, FK, and FBP—are derived from the same wave-based light-cone model. No real measured transients, no independent Monte Carlo path tracer, and no withheld physical validation are used. Therefore the PSNR and SSIM values in Tables 1-2 demonstrate that the simulators and the inverses are mutually consistent, not that hidden geometries are accurately recovered from physically realistic NLOS measurements. The claim in the abstract and conclusion that reconstructions 'accurately recover hidden geometries' is not supported; an evaluation against an independent transient renderer or real experimental data is required.
  3. [Section 3] The use of a single orthographic depth map to represent the hidden object is a significant modeling limitation that is not discussed. NLOS transients integrate contributions from all surfaces visible from the relay wall, including multiple depth layers and self-occlusions, whereas an orthographic projection captures only one 2.5D view. Combined with the uniform Lambertian albedo of 1 used for the ShapeNet dataset, this restricts the diversity and realism of the benchmark. The claim that the corpus is 'physically consistent, diverse' is therefore overstated; the paper should disclose these simplifications and justify that they do not bias algorithm rankings.
minor comments (5)
  1. [Table 2] The header of Table 2 uses 'BP' for the filtered-backprojection row, while the text and Table 1 use 'FBP'; these labels should be made consistent.
  2. [Section 2 / Section 3] Section 2 defines a grid of (2N)x(2N)x(2M) nodes, while Section 3 describes a 128x128x1024 space-time grid; the relation between these dimensions is not explained and should be clarified.
  3. [Section 7] The Data availability section refers to 'Data Repository' and 'Code Repository' but gives no URLs or persistent identifiers, which prevents readers from accessing the claimed release.
  4. [Section 2] The definitions of the resampling operators R_z and R_t^{-1} are ambiguous: the row/column domains of R_z are not specified, and R_t^{-1}=R_z^T D_v^{3/2} refers to D_v of size M^2 while the preceding grid has 2M depth nodes. A complete index convention and a verification of energy conservation would make the derivation reproducible.
  5. [Section 2] The depth coordinate is mapped to the interval [-1,1], but physical depth is nonnegative; using a signed depth in the PSF and in the depth-to-volume mapping creates potential confusion about where the hidden surface is located.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the LCT-based simulator and benchmark are not equivalent by construction, since independent algorithms (PF, FK, FBP) outperform LCT on the synthetic data.

full rationale

The paper's chain is: (1) adopt the LCT forward model from O'Toole et al. [2]; (2) discretize it via resampling operators and FFT convolution; (3) generate transients from ShapeNet depth maps; (4) benchmark four reconstruction algorithms. The adopted forward model is not an input of the benchmark in a way that forces the results: the dataset is generated by the simulator, and the benchmarked methods include PF, FK, and FBP, which are independent algorithms developed outside this work and cited to external groups [2,17,33,34]. The tables show PF highest and LCT lowest, so the benchmark does not reduce to inverting the generator. The only self-referential aspect is that the same LCT model appears on both sides, but LCT is not the claimed output or the best performer, and no parameter is fitted to make a particular method succeed. The claim that reconstructions recover hidden geometries is an internal consistency check of the simulator, not a derivation of physics from the model. No prediction is renamed from a fit, and no load-bearing premise rests on self-citations. Note: the discrete PSF expression in Section 2/Algorithm 1 may be internally inconsistent with the continuous change-of-variables kernel (squared-depth vs linear-depth paraboloid), but that is a correctness issue and does not make the derivation circular.

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

The central claim rests on modeling assumptions that are standard in the NLOS literature but are not independently validated in this paper: Lambertian single-bounce transport, first-surface depth-map representation, Gaussian jitter, and Poisson statistics. The only free parameters are scenario settings and noise levels chosen by the user; none is fitted to experimental data.

free parameters (4)
  • slope = w/(M c Δt)
    Chosen to map normalized lateral coordinates to the time-depth grid in the PSF; its exact role is inconsistent between the text and Algorithm 1.
  • n_r (λ_poisson) = 2 (dataset default)
    Noise-level parameter controlling Poisson SNR; set to 2 for the public dataset; result tables depend on this choice.
  • temporal jitter FWHM = 70 ps (dataset default)
    Detector temporal jitter; chosen for the dataset; affects reconstruction metrics.
  • uniform albedo = 1.0 (dataset default)
    Albedo assigned to all ShapeNet objects when A is unavailable; affects transient amplitudes and noise.
assumptions (6)
  • domain assumption The confocal LCT forward model adequately represents real NLOS transient measurements
    Adopted in Section 2 as the core model; the paper validates only against this model, never against measured or path-traced transients.
  • domain assumption Hidden objects are Lambertian and light transport is single-bounce (wall-object-wall only)
    Section 2, LCT integral uses isotropic albedo ρ and a single round-trip path; interreflections and non-Lambertian BRDFs are absent.
  • domain assumption An orthographic first-surface depth map and a flat albedo map fully describe the hidden scene for simulation purposes
    Section 3 projects ShapeNet point clouds to 128×128 depth maps; no volumetric scattering, transparency, or concavity self-occlusion beyond the first surface.
  • domain assumption Detector jitter is zero-mean Gaussian and independent per wall position
    Section 2, 'timing error can be well approximated by a zero-mean Gaussian distribution'; no correlated or asymmetric timing errors.
  • standard math Photon counts follow Poisson statistics with variance equal to mean
    Section 2, 'detector output obeys Poisson statistics'; used to inject noise.
  • standard math Change of variables z=√u and the 3D convolution theorem via FFT are valid
    Section 2, used to reduce the transient integral to a convolution; valid for z≥0.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fast Non-Line-of-Sight Transient Data Simulation and an Open Benchmark Dataset." pith.science (2026). https://pith.science/paper/TSRNSPOB

@misc{pith2026250603747,
  author       = {Pith},
  title        = {Pith review of: Fast Non-Line-of-Sight Transient Data Simulation and an Open Benchmark Dataset},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TSRNSPOB}},
  note         = {Machine review of arXiv:2506.03747}
}
read the original abstract

Non-Line-of-Sight (NLOS) imaging reconstructs the shape and depth of hidden objects from picosecond-resolved transient signals, offering potential applications in autonomous driving, security, and medical diagnostics. However, current NLOS experiments rely on expensive hardware and complex system alignment, limiting their scalability. This manuscript presents a simplified simulation method that generates NLOS transient data by modeling light-intensity transport rather than performing conventional path tracing, significantly enhancing computational efficiency. All scene elements, including the relay surface, hidden target, stand-off distance, detector time resolution, and acquisition window are fully parameterized, allowing for rapid configuration of test scenarios. Reconstructions based on the simulated data accurately recover hidden geometries, validating the effectiveness of the approach. The proposed tool reduces the entry barrier for NLOS research and supports the optimization of system design.

Figures

Figures reproduced from arXiv: 2506.03747 by the authors.

Figure 1
Figure 1. Schematic diagram of the non-line-of-sight imaging optical system Within this setting, the light-cone transform (LCT) achieves an effective balance between physical accuracy and computational efficiency [2], and is therefore adopted as the core model in our simulator. For a detection point (𝑥 ′ , 𝑦′ ) on the relay wall, whose axial coordinate is set to (𝑧 ′ = 0) , the three-dimensional transient measurement 𝜏 ∈ R (𝑥… view at source ↗
Figure 2
Figure 2. Comparison of transient data and reconstruction results for targets with different albedos. where parameter 𝐴 is normalised to the interval [0,1], and when 𝐴 is unavailable, an albedo value of 0.5 is applied. In the transient model, the voxel intensity determines the expected photon count. Voxels with low albedo (dark regions) generate fewer photons and therefore exhibit a lower SNR, whereas voxels with high albedo … view at source ↗
Figure 3
Figure 3. Comparison of 1-D transient data and reconstruction results under different temporal jitters.(a) Reconstruction result. (b) One-dimensional transient histogram corresponding to a single spatial position. dominant noise source in the measurement; the simulator explicitly models this noise as well. 𝑇 = Poisson Var(𝜏𝐺) 𝜎𝑛  𝜎𝑛 . the signal power is taken as the variance Var(𝜏𝐺) of the jittered trace, while the noise po… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Transient data and reconstruction results under different poisson noise levels. different research groups vary in scanner design, detector bandwidth, and laser wavelength, their transient measurements display a statistically similar sparse pulse distribution. This simi…
Figure 5
Figure 5. Figure 5: Depth maps of targets from different categories [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: SSIM comparison for targets reconstructed by different methods. 128 × 128 depth map, an operation that both reduces storage and reflects the fact that a real system senses only the first visible surface. We select seven representative categories, such as car, chair, gu…
Figure 7
Figure 7. Figure 7: PSNR comparison for targets reconstructed by different methods. emulated by first applying a Poisson process with intensity factor 𝑛𝑟 = 2 and then convolving each pixel’s time trace with a 70𝑝𝑠 FWHM Gaussian jitter. The resulting transient volumes serve both as benchma…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

34 extracted references · 33 canonical work pages

  1. [1]

    Non-line-of-sight imaging,

    D. Faccio, A. Velten, and G. Wetzstein, “Non-line-of-sight imaging,” Nat. Rev. Phys.2, 318–327 (2020)

    work page 2020

  2. [2]

    Confocalnon-line-of-sightimagingbasedonthelight-conetransform,

    M.O’Toole,D.B.Lindell, andG.Wetzstein,“Confocalnon-line-of-sightimagingbasedonthelight-conetransform,” Nature555, 338–341 (2018)

    work page 2018

  3. [3]

    Non-line-of-sightimagingwithpartial occluders and surface normals,

    F.Heide, M.O’Toole, K.Zang, D.B.Lindell, S.Diamond, andG.Wetzstein, “Non-line-of-sightimagingwithpartial occluders and surface normals,” ACM Trans. on Graph. (ToG)38, 1–10 (2019)

    work page 2019

  4. [4]

    High-resolution non-line-of-sight imaging employing active focusing,

    R. Cao, F. De Goumoëns, B. Blochet, J. Xu, and C. Yang, “High-resolution non-line-of-sight imaging employing active focusing,” Nat. Photonics16, 462–468 (2022)

    work page 2022

  5. [5]

    Non-invasive imaging through scattering medium and around corners beyond 3d memory effect,

    Y. Shi, E. Guo, M. Sun, L. Bai, and J. Han, “Non-invasive imaging through scattering medium and around corners beyond 3d memory effect,” Opt. Lett.47, 4363–4366 (2022)

    work page 2022

  6. [6]

    Imaging consecutive targets through scattering medium and around corners beyond the optical memory effect using untrained network,

    Y. Shi, E. Guo, M. Sun, L. Bai, D. Mu, and J. Han, “Imaging consecutive targets through scattering medium and around corners beyond the optical memory effect using untrained network,” Results Phys.51, 106691 (2023)

    work page 2023

  7. [7]

    Super-resolution non-line-of-sight imaging based on temporal encoding,

    J. Miao, E. Guo, Y. Shi, F. Cai, L. Bai, and J. Han, “Super-resolution non-line-of-sight imaging based on temporal encoding,” Opt. Express31, 40235–40248 (2023)

    work page 2023

  8. [8]

    Deep-inverse correlography: towards real-time high-resolution non-line-of-sight imaging,

    C. A. Metzler, F. Heide, P. Rangarajan, M. M. Balaji, A. Viswanath, A. Veeraraghavan, and R. G. Baraniuk, “Deep-inverse correlography: towards real-time high-resolution non-line-of-sight imaging,” Optica7, 63–71 (2020)

    work page 2020

Show all 34 references
  1. [9]

    Single-shotnon-line-of-sightimagingbasedonchromato-axial differential correlography,

    L.Liu,S.Zhu,W.Zhang,L.Bai,E.Guo,andJ.Han,“Single-shotnon-line-of-sightimagingbasedonchromato-axial differential correlography,” Photonics Res.12, 106–114 (2023)

    2023

  2. [10]

    Corner occluder computational periscopy: Estimating a hidden scene from a single photograph,

    S. W. Seidel, Y. Ma, J. Murray-Bruce, C. Saunders, W. T. Freeman, C. C. Yu, and V. K. Goyal, “Corner occluder computational periscopy: Estimating a hidden scene from a single photograph,” in2019 IEEE International Conference on Computational Photography (ICCP), (IEEE, 2019), pp. 1–9

    2019

  3. [11]

    Inferring light fields from shadows,

    M. Baradad, V. Ye, A. B. Yedidia, F. Durand, W. T. Freeman, G. W. Wornell, and A. Torralba, “Inferring light fields from shadows,” inProceedings of the IEEE conference on computer vision and pattern recognition, (2018), pp. 6267–6275

    2018

  4. [12]

    Using unknown occluders to recover hidden scenes,

    A. B. Yedidia, M. Baradad, C. Thrampoulidis, W. T. Freeman, and G. W. Wornell, “Using unknown occluders to recover hidden scenes,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, (2019), pp. 12231–12239

    2019

  5. [13]

    Looking around the corner using ultrafast transient imaging,

    A. Kirmani, T. Hutchison, J. Davis, and R. Raskar, “Looking around the corner using ultrafast transient imaging,” Int. journal computer vision95, 13–28 (2011)

    2011

  6. [14]

    Occluded imaging with time-of-flight sensors,

    A. Kadambi, H. Zhao, B. Shi, and R. Raskar, “Occluded imaging with time-of-flight sensors,” ACM Trans. on Graph. (ToG)35, 1–12 (2016)

    2016

  7. [15]

    Estimating motion and size of moving non-line-of-sight objects in cluttered environments,

    R. Pandharkar, A. Velten, A. Bardagjy, E. Lawson, M. Bawendi, and R. Raskar, “Estimating motion and size of moving non-line-of-sight objects in cluttered environments,” inCVPR 2011, (IEEE, 2011), pp. 265–272

    2011

  8. [16]

    Revealing hidden scenes by photon-efficient occlusion-based opportunistic active imaging,

    F. Xu, G. Shulkind, C. Thrampoulidis, J. H. Shapiro, A. Torralba, F. N. Wong, and G. W. Wornell, “Revealing hidden scenes by photon-efficient occlusion-based opportunistic active imaging,” Opt. express26, 9945–9962 (2018)

    2018

  9. [17]

    Non-line-of-sight imaging using phasor-field virtual wave optics,

    X. Liu, I. Guillén, M. La Manna, J. H. Nam, S. A. Reza, T. Huu Le, A. Jarabo, D. Gutierrez, and A. Velten, “Non-line-of-sight imaging using phasor-field virtual wave optics,” Nature572, 620–623 (2019)

    2019

  10. [18]

    Non-line-of-sight reconstruction with signal–object collaborative regularization,

    X. Liu, J. Wang, Z. Li, Z. Shi, X. Fu, and L. Qiu, “Non-line-of-sight reconstruction with signal–object collaborative regularization,” Light. Sci. & Appl.10, 198 (2021)

    2021

  11. [19]

    Steady state non-line of sight imaging via unsupervised network,

    Y. Shi, E. Guo, J. Miao, M. Sun, L. Bai, and J. Han, “Steady state non-line of sight imaging via unsupervised network,” inNinth symposium on novel photoelectronic detection technology and applications, vol. 12617 (SPIE, 2023), pp. 559–563

    2023

  12. [20]

    Real-time scan-free non-line-of-sight imaging,

    W. Zhang, E. Guo, S. Zhu, C. Huang, L. Chen, L. Liu, L. Bai, E. Y. Lam, and J. Han, “Real-time scan-free non-line-of-sight imaging,” APL Photonics9 (2024)

    2024

  13. [21]

    Gpu computing,

    J. D. Owens, M. Houston, D. Luebke, S. Green, J. E. Stone, and J. C. Phillips, “Gpu computing,” Proc. IEEE96, 879–899 (2008)

    2008

  14. [22]

    Differentiable transient rendering,

    S. Yi, D. Kim, K. Choi, A. Jarabo, D. Gutierrez, and M. H. Kim, “Differentiable transient rendering,” ACM Trans. on Graph. (TOG)40, 1–11 (2021)

    2021

  15. [23]

    Fastdifferentiabletransientrenderingfornon-line-of-sight reconstruction,

    M.Plack,C.Callenberg,M.Schneider,andM.B.Hullin,“Fastdifferentiabletransientrenderingfornon-line-of-sight reconstruction,” inProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, (2023), pp. 3067–3076

    2023

  16. [24]

    Non-line-of-sight transient rendering,

    D. Royo, J. García, P. Luesia-Lahoz, J. Marco, D. Gutiérrez, A. Muñoz, and A. Jarabo, “Non-line-of-sight transient rendering,” inACM SIGGRAPH 2022 Posters,(2022), pp. 1–2

    2022

  17. [25]

    Non-line-of-sight reconstruction using efficient transient rendering,

    J. Iseringhausen and M. B. Hullin, “Non-line-of-sight reconstruction using efficient transient rendering,” ACM Trans. on Graph. (ToG)39, 1–14 (2020)

    2020

  18. [26]

    Time-of-flight and transient rendering,

    A. K. PEDIREDLA, “Time-of-flight and transient rendering,” Comput. Imaging for Scene Understanding: Transient, Spectral, Polarim. Anal. p. 45 (2024)

    2024

  19. [27]

    Learned feature embeddings for non-line-of-sight imaging and recognition,

    W. Chen, F. Wei, K. N. Kutulakos, S. Rusinkiewicz, and F. Heide, “Learned feature embeddings for non-line-of-sight imaging and recognition,” ACM Trans. on Graph. (ToG)39, 1–18 (2020)

    2020

  20. [28]

    Steady-state non-line-of-sight imaging,

    W. Chen, S. Daneau, F. Mannan, and F. Heide, “Steady-state non-line-of-sight imaging,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, (2019), pp. 6790–6799

    2019

  21. [29]

    A review of new time-to-digital conversion techniques,

    S. Tancock, E. Arabul, and N. Dahnoun, “A review of new time-to-digital conversion techniques,” IEEE transactions on Instrum. Meas.68, 3406–3417 (2019)

    2019

  22. [30]

    A brief introduction to time-to-digital and digital-to-time converters,

    G. W. Roberts and M. Ali-Bakhshian, “A brief introduction to time-to-digital and digital-to-time converters,” IEEE Trans. on Circuits Syst. II: Express Briefs57, 153–157 (2010)

    2010

  23. [31]

    Time-to-digital conversion techniques: a survey of recent developments,

    J. Szyduczyński, D. Kościelnik, and M. Miśkowicz, “Time-to-digital conversion techniques: a survey of recent developments,” Measurement214, 112762 (2023)

    2023

  24. [32]

    Shapenet: An information-rich 3d model repository,

    A. X. Chang, T. Funkhouser, L. Guibas, P. Hanrahan, Q. Huang, Z. Li, S. Savarese, M. Savva, S. Song, H. Suet al., “Shapenet: An information-rich 3d model repository,” arXiv preprint arXiv:1512.03012 (2015)

    2015 arXiv

  25. [33]

    Non-line-of-sight-imaging using dynamic relay surfaces,

    M. La Manna, J.-H. Nam, S. Azer Reza, and A. Velten, “Non-line-of-sight-imaging using dynamic relay surfaces,” Opt. express28, 5331–5339 (2020)

    2020

  26. [34]

    Wave-based non-line-of-sight imaging using fast fk migration,

    D. B. Lindell, G. Wetzstein, and M. O’Toole, “Wave-based non-line-of-sight imaging using fast fk migration,” ACM Trans. on Graph. (ToG)38, 1–13 (2019)

    2019

Pith tools

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