Pith. sign in

REVIEW 4 major objections 5 minor 40 references

Learnable Burst-Encodable Time-of-Flight Imaging for High-Fidelity Long-Distance Depth Sensing

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 Burst-mode ToF measures far and fine with one frequency

desk verdict The burst scheme reduces to a gated iToF slab; Eq. (4) limits unambiguous range to cTm/2, not cTburst/2, so the central long-range claim doesn't follow. read the letter →

arxiv 2505.22025 v1 pith:WLSLNMS2 submitted 2025-05-28 cs.CV eess.IV

Manchao Bao , Shengjiang Fang , Tao Yue , Xuemei Hu This is my paper
classification cs.CVeess.IV
keywords burst-encodabletime-of-flightindirectphasewrappingsingle-frequencymodulationend-to-endlearningdepthreconstructioncodedexposurelong-distancesensing
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 tries to establish that a time-of-flight camera can measure depth both far away and precisely using a single modulation frequency, if the light is emitted as short pulses grouped into a longer burst. Conventional indirect ToF is limited by phase wrapping: distance is read from a phase that repeats every modulation cycle, so long range and fine precision pull against each other. BE-ToF instead reads the delay of the whole burst, so the unambiguous range is set by the long burst period while the precision is set by the short pulse period inside the burst. The authors further claim that the exposure coding functions and a depth-reconstruction network can be learned end-to-end, with constraints that keep the codes binary and hardware-implementable. If correct, this would give autonomous driving and robotics a cheaper long-range depth sensor than direct ToF, which needs picosecond pulses and time-to-digital converters.

What carries the argument

The load-bearing object is the burst-mode modulation identity of Eq. 1 together with the differentiable forward model of Eqs. 2--5. The identity decomposes the measured phase into a coarse delay $\phi_1$ from the burst envelope and a fine phase $\phi_2$ from the short modulation cycle, so the depth formula separates range from precision. The forward model treats the emitted light as a periodic pulse train $M(t)$, attenuated with distance, and integrates it against learnable binary coding functions $D_i(t)$; this is what lets the network optimize codes end-to-end. The double-well loss and first-order difference loss push the learned codes toward binarized, wide-enough pulses that an encodable exposure camera can realize.

What would settle it

Calibrate the prototype on a known target, load the learned coding functions into the intensifier, and compare measured depth error against the simulated MAE at matched SNR; a significant excess beyond the sensor noise floor would refute the ideal-gate forward model. A minimal version is to compare the learned codes with the square coding function on the same hardware: if the learned codes do not beat square on real data, the learnable-coding advantage does not transfer.

Watch

Extended reading notes

Core claim

The central claim is that depth in a burst-mode ToF system is $d = c(\phi_1+\phi_2)T_{\mathrm{burst}}/(4\pi) = c\tau/2 + D(\phi_2)$, with maximum unambiguous range $d_{\mathrm{mur}} = cT_{\mathrm{burst}}/2$ and depth accuracy $\epsilon_d = c\epsilon_\phi T_m/(4\pi)$. Here $\phi_1$ is the coarse phase set by the burst time delay $\tau$, and $\phi_2$ is the fine phase recovered from the demodulation coding functions. Because $T_{\mathrm{burst}}$ and $T_m$ are independent, the system sets its range and its precision separately: the burst period decides how far it can see without wrapping, and the short pulse period decides how finely depth is quantized. The paper argues that this is a new single-frequency paradigm with no phase ambiguity, and validates it with simulated comparisons against single- and dual-frequency baselines and a prototype that images an outdoor scene at about 23 meters.

Load-bearing premise

The simulation assumes the camera's exposure gates respond instantly and linearly to a perfectly periodic pulse train, with constant scene reflectance and ambient light over the burst; the hardware experiment is configured with a 50 ns demodulation period but does not state that the learned codes were loaded, so the long-range precision claim is only demonstrated in simulation.

Editorial extensions

If this is right

  • A single burst-mode frequency reaches about 19 mm MAE at 90--93 m under high SNR in simulation, while a learned single-frequency baseline reports 75 mm and dual-frequency DeepToF reports 43 mm.
  • The maximum unambiguous range becomes $cT_{\mathrm{burst}}/2$; at the prototype's 200 kHz burst rate this is roughly 750 m, so phase wrapping effectively disappears for terrestrial scenes.
  • The sensing window can be aimed at any depth band $[c\tau/2, c(\tau+T_m)/2]$ by adjusting the time delay $\tau$, which also suppresses foreground and background clutter in the captured depth map.
  • The learned coding functions remain binarized and wide enough for hardware implementation, thanks to the double-well and first-order difference losses.
  • The prototype demonstrates the paradigm outdoors at about 23 m, resolving steps, flower beds, trees, and a marble surface.

Reading between the lines

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

  • Editorial inference: the same burst timing trick should transfer to other pulsed-illumination sensors, such as SPAD arrays, wherever the integration window can be shaped in time; the essential requirement is a controllable exposure code, not the specific intensifier hardware.
  • Editorial inference: the long-range precision claim is only shown in simulation, and the hardware section does not state that the learned codes were loaded onto the prototype; a decisive test would be to compare the learned codes against the square coding function on the same hardware.
  • Editorial inference: scene motion becomes a new constraint because the burst period $T_{\mathrm{burst}}$ must stay short enough that the scene does not move across the measurement window; interleaving bursts or sweeping $\tau$ could trade range bandwidth for motion robustness.
  • Editorial inference: since the unambiguous range is tied to $T_{\mathrm{burst}}$, a practical system could switch burst rates per frame to zoom the sensing range, provided the codes are trained across the relevant $T_m$ values.
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. The paper proposes Burst-Encodable Time-of-Flight (BE-ToF), a burst-mode indirect ToF system intended to achieve long-range, high-precision depth sensing with a single modulation frequency. The authors introduce a differentiable forward model, an end-to-end learnable framework that jointly optimizes binarized coding functions and a Restormer-based reconstruction network (RSCF-Net), and a set of losses including double-well and first-order difference regularizers and a Fisher guidance term. They report synthetic evaluations on NYU-V2 at simulated depth ranges from 0-3 m to 90-93 m and a qualitative real-world prototype experiment at about 23 m. The central claim is that the burst geometry simultaneously provides an unambiguous range of cT_burst/2 and depth precision governed by the shorter period T_m, thereby avoiding the phase-wrapping problem of conventional iToF.

Significance. If the central claim were correct, BE-ToF would offer a hardware-friendly alternative to dToF for long-range depth imaging, and the combination of learned binarized coding with an encodable exposure camera could be practically valuable. The paper contains a complete differentiable model, a sensible regularization strategy for hardware implementability, and a real prototype, and the authors are explicit about the physical quantities involved. However, the central claim is undermined by an internal inconsistency between the stated unambiguous range and the paper's own forward model: the measurement equation depends only on the round-trip delay modulo T_m, so a single burst measurement is wrapping-limited in exactly the way the paper claims to avoid. The synthetic evaluation also omits the proposed method from the main quantitative comparison table, and the hardware experiment does not validate the learned-coding component. These issues are load-bearing, so the contribution as presented cannot be accepted.

major comments (4)
  1. [Section 3, Eqs. (1) and (4)] The forward model in Eq. (4) integrates the modulation signal M(t) over exactly one period T_m. Since M(t) is periodic with period T_m, the measurement I_i(s) depends on (t_d - τ) mod T_m, where t_d = 2d/c. Consequently, two depths separated by cT_m/2 produce identical measurements at fixed τ, so the unambiguous range of a single burst measurement is cT_m/2, not cT_burst/2 as claimed in Eq. (1). The term cτ/2 in Eq. (1) is a known gate delay, not an estimated quantity, and the learned coding functions in Eqs. (7)-(8) operate within a single T_m window and cannot recover the coarse pulse index. To support the claimed long unambiguous range, the model would need to include the burst envelope or explicitly scan τ across the burst; as written, the central claim of wrapping-free long-range sensing is not derivable from the paper's own equations.
  2. [Section 4.2, Table 1(a)] The overall performance comparison in Table 1(a) lists Sine+PS, Square+PS, DeepToF, and FisherToF but does not include an 'Ours' row; the 'Ours' row appears only in Table 1(c), which compares reconstruction networks under the same learned coding and is not the overall comparison. The text states that 'Our method achieves the best performance across various distances and SNRs' and refers to Table 1(a) for the detailed comparison, but the quantitative basis for that claim is missing from the table.
  3. [Section 4.2, Table 1 and Section 4.1 dataset description] The NYU-V2 depth maps are originally in the 0-3 m range, yet Table 1 reports results for 30-33 m, 60-63 m, and 90-93 m. The paper does not state how these depth ranges are produced—in particular, the values of τ, T_m, and T_burst, and whether a constant depth offset is added. Because Eq. (4) is periodic in (t_d - τ), adding a constant to d changes only the attenuation factor F_d(s) in Eq. (3) and not the phase information; without knowing the protocol, the long-range simulation results are not reproducible, and they may conflate a known gate offset with an estimated depth.
  4. [Section 5, physical experiment] The hardware experiment is qualitative and does not validate the learned-coding component of the system. It reports a depth map of a scene at about 23 m but provides no ground truth, depth error, or quantitative comparison. The text states that the demodulation function is configured with a 50 ns period and that the delay is adjustable, but it does not state whether the learned binarized coding functions from Section 3.3 were programmed into the encodable image intensifier camera. Consequently, the end-to-end claim that the jointly optimized coding functions and network achieve the reported performance rests entirely on simulation with the same forward model used for training.
minor comments (5)
  1. [Section 3 and Figure 1(b)] The phrase 'the phase shift of the reflected signal spans the full range [0, 2π] within a single burst period' is ambiguous: for a T_m-periodic modulation signal, the phase spans [0, 2π] within each T_m cycle, not within the longer T_burst envelope.
  2. [Section 3.3, Eq. (9)] The Fisher guidance loss uses σ_i(s) = sqrt(E(I_i(s)) + E(n_d) + σ_r^2), while Eq. (5) defines the readout noise as n_r ~ N(0, σ_r^2). The notation should be made consistent so that σ_r is clearly the standard deviation.
  3. [Section 4.1 and Table 1] The paper specifies K=4 and M=1000 but does not report the simulation values of T_m, T_burst, or τ, which are needed to interpret the claimed depth ranges and the relation between the long-range results and Eq. (1).
  4. [Section 6, Limitations] The limitations paragraph states that the operational range is confined between cτ/2 and c(τ+T_m)/2, which is consistent with a single T_m-window measurement but not with the earlier claim that d_mur = cT_burst/2; these statements should be reconciled.
  5. [Table 2] The 'Ours' row in Table 2 reports 8.52, 12.86, 18.20, and 23.51 mm, while Table 1(c) lists different 'Ours' values; the caption says Table 2 values are averages over SNRs, but this averaging procedure should be stated explicitly and the relationship between the two tables clarified.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the self-citation to Fisher guidance is minor and not load-bearing, and the central formula, though internally inconsistent about Tburst versus Tm, is not a fitted-input prediction.

full rationale

The paper's derivation chain is largely self-contained. Equation (1) is a geometric relation for a gated burst: depth is a known gate offset cτ/2 plus a fine demodulation D(φ2); Equation (4) defines the measurements as an integral over one modulation period Tm; and the losses in Eqs. (6)-(11) optimize coding functions and a reconstruction network against this forward model. Training and testing use disjoint splits of NYU-V2, so the synthetic comparisons are not circular: the held-out test scenes measure generalization of the learned coding functions and network. The Fisher guidance loss is credited to the authors' own prior paper [10], but it is a training regularizer, not a load-bearing premise for the depth relation or the prototype demonstration, so this is a minor self-citation rather than circularity. A genuine concern is that the advertised maximum unambiguous range dmur = c·Tburst/2 is not supported by Eq. (4): with a fixed τ and periodic M(t), the measurements depend only on (t_d−τ) mod Tm, giving a per-measurement unambiguous interval of cTm/2; the Limitations section itself restricts the operational range to [cτ/2, c(τ+Tm)/2]. That is an internal-consistency/correctness issue, not a circular one, since no fitted parameter is renamed as a prediction and no uniqueness theorem is imported to forbid alternatives.

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

The central claim rests on the forward model assumptions above; no new physical entities are introduced.

free parameters (4)
  • Loss balance coefficients gamma1, gamma2, gamma3 = gamma1: 5e-4 -> 5e-5, gamma2: 5e-2 -> 1, gamma3: 5
    Hand-chosen and annealed during training; they shape the learned codes and network but do not appear in the core depth equation.
  • Number of coding functions K and sampling points M = K=4, M=1000
    Hyperparameters chosen for the experiments; K>=3 is motivated by unknowns, but K=4 is a specific choice.
  • Learned coding functions Di(t) = Optimized vectors of length M=1000 for each of K=4 measurements
    The central learnable quantities; they are fitted to the training data during end-to-end optimization.
  • Noise parameters E(nd) and sigma_r = Not specified in the text
    Used in Eq. 5 to simulate sensor noise; values are required to reproduce the SNR levels but are not reported.
assumptions (4)
  • domain assumption The reflected signal at each scene point is an ideal scaled and delayed copy of the periodic pulse train M(t), plus constant ambient light (Eqs. 2-3).
    Invoked in the forward model; real pulsed lasers have jitter and intensity drift, and ambient light varies over the burst.
  • domain assumption Light attenuation follows an inverse-square law, F_d proportional to 1/d^2 (Eq. 3).
    Standard approximation from [30]; real long-range scenes include scattering and atmospheric extinction not captured.
  • domain assumption The camera measurement is a linear integral of the reflected signal times the coding function (Eq. 4).
    Ignores image intensifier nonlinearities, phosphor persistence, and gate timing errors.
  • ad hoc to paper Binarized coding functions regularized by the double well and first-order difference losses are implementable on the encodable exposure camera.
    This is an unverified assumption; the prototype does not demonstrate that the learned codes are actually loaded and reproduced at the required temporal resolution.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learnable Burst-Encodable Time-of-Flight Imaging for High-Fidelity Long-Distance Depth Sensing." pith.science (2026). https://pith.science/paper/WLSLNMS2

@misc{pith2026250522025,
  author       = {Pith},
  title        = {Pith review of: Learnable Burst-Encodable Time-of-Flight Imaging for High-Fidelity Long-Distance Depth Sensing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WLSLNMS2}},
  note         = {Machine review of arXiv:2505.22025}
}
read the original abstract

Long-distance depth imaging holds great promise for applications such as autonomous driving and robotics. Direct time-of-flight (dToF) imaging offers high-precision, long-distance depth sensing, yet demands ultra-short pulse light sources and high-resolution time-to-digital converters. In contrast, indirect time-of-flight (iToF) imaging often suffers from phase wrapping and low signal-to-noise ratio (SNR) as the sensing distance increases. In this paper, we introduce a novel ToF imaging paradigm, termed Burst-Encodable Time-of-Flight (BE-ToF), which facilitates high-fidelity, long-distance depth imaging. Specifically, the BE-ToF system emits light pulses in burst mode and estimates the phase delay of the reflected signal over the entire burst period, thereby effectively avoiding the phase wrapping inherent to conventional iToF systems. Moreover, to address the low SNR caused by light attenuation over increasing distances, we propose an end-to-end learnable framework that jointly optimizes the coding functions and the depth reconstruction network. A specialized double well function and first-order difference term are incorporated into the framework to ensure the hardware implementability of the coding functions. The proposed approach is rigorously validated through comprehensive simulations and real-world prototype experiments, demonstrating its effectiveness and practical applicability.

Figures

Figures reproduced from arXiv: 2505.22025 by the authors.

Figure 1
Figure 1. Comparison between iToF and BE-ToF. (a) Principle of iToF imaging, which suffers [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. (a) End-to-end BE-ToF imaging framework for jointly optimizing the demodulation function [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Architecture of the Restormer-based Spatial-Channel Fusion Network(RSCF-Net), (a) [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Demonstration of the dou￾ble well function with two identi￾cal minima located at x = 0 and x = 1. Double Well Function Loss. To enable the optimization of binarized coding functions within the differentiable physical model. We introduce the double well function from qu…
Figure 5
Figure 5. Figure 5: Overall comparisons with traditional iToF methods under various distances and SNRs, [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Comparisons with different coding scheme and reconstruction networks. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Visual ablations on Ldw and L1st [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: (a) System prototype of the BE-ToF, (b) Outdoor Scene, (c) Depth map. [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

40 extracted references · 38 canonical work pages

  1. [1]

    Breaking the temporal and frequency congestion of lidar by parallel chaos

    Ruixuan Chen, Haowen Shu, Bitao Shen, Lin Chang, Weiqiang Xie, Wenchao Liao, Zihan Tao, John E Bowers, and Xingjun Wang. Breaking the temporal and frequency congestion of lidar by parallel chaos. Nature Photonics, 17(4):306–314, 2023

    work page 2023

  2. [2]

    Automotive lidar technology: A survey.IEEE Transactions on Intelligent Transportation Systems, 23(7):6282–6297, 2021

    Ricardo Roriz, Jorge Cabral, and Tiago Gomes. Automotive lidar technology: A survey.IEEE Transactions on Intelligent Transportation Systems, 23(7):6282–6297, 2021

    work page 2021

  3. [3]

    Performance evaluation of the 1st and 2nd generation kinect for multimedia applications

    Simone Zennaro, Matteo Munaro, Simone Milani, Pietro Zanuttigh, Andrea Bernardi, Stefano Ghidoni, and Emanuele Menegatti. Performance evaluation of the 1st and 2nd generation kinect for multimedia applications. InIEEE International Conference on Multimedia and Expo, pages 1–6. IEEE, 2015

    work page 2015

  4. [4]

    Estimation of optical pathlength through tissue from direct time of flight measurement.Physics in Medicine & Biology, 33(12):1433, 1988

    David T Delpy, Mark Cope, Pieter van der Zee, Simon Arridge, Susan Wray, and JS Wyatt. Estimation of optical pathlength through tissue from direct time of flight measurement.Physics in Medicine & Biology, 33(12):1433, 1988

    work page 1988

  5. [5]

    Springer Science & Business Media, 2012

    Miles Hansard, Seungkyu Lee, Ouk Choi, and Radu Patrice Horaud.Time-of-flight cameras: principles, methods and applications. Springer Science & Business Media, 2012

    work page 2012

  6. [6]

    Time-of-flight camera—an introduction.Technical White Paper, (SLOA190B), 2014

    Larry Li et al. Time-of-flight camera—an introduction.Technical White Paper, (SLOA190B), 2014

    work page 2014

  7. [7]

    Theoretical and experimental error analysis of continuous-wave time-of-flight range cameras.Optical Engineering, 48(1):013602–013602, 2009

    Mario Frank, Matthias Plaue, Holger Rapp, Ullrich Köthe, Bernd Jähne, and Fred A Hamprecht. Theoretical and experimental error analysis of continuous-wave time-of-flight range cameras.Optical Engineering, 48(1):013602–013602, 2009

    work page 2009

  8. [8]

    itof-flow-based high frame rate depth imaging

    Yu Meng, Zhou Xue, Xu Chang, Xuemei Hu, and Tao Yue. itof-flow-based high frame rate depth imaging. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4929–4938, 2024

    work page 2024

Show all 40 references
  1. [9]

    An overview of depth cameras and range scanners based on time-of-flight technologies.Machine vision and applications, 27(7):1005– 1020, 2016

    Radu Horaud, Miles Hansard, Georgios Evangelidis, and Clément Ménier. An overview of depth cameras and range scanners based on time-of-flight technologies.Machine vision and applications, 27(7):1005– 1020, 2016

    2016

  2. [10]

    Fisher information guidance for learned time-of-flight imaging

    Jiaqu Li, Tao Yue, Sijie Zhao, and Xuemei Hu. Fisher information guidance for learned time-of-flight imaging. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 16334–16343, 2022

    2022

  3. [11]

    A twofold modulation frequency laser range finder.Journal of Optics A: Pure and Applied Optics, 4(6):S356, 2002

    Stephane Poujouly and Bernard Journet. A twofold modulation frequency laser range finder.Journal of Optics A: Pure and Applied Optics, 4(6):S356, 2002

    2002

  4. [12]

    Analysis of errors in tof range imaging with dual-frequency modulation.IEEE Transactions on Instrumen- tation and Measurement, 60(5):1861–1868, 2011

    Adrian PP Jongenelen, Donald G Bailey, Andrew D Payne, Adrian A Dorrington, and Dale A Carnegie. Analysis of errors in tof range imaging with dual-frequency modulation.IEEE Transactions on Instrumen- tation and Measurement, 60(5):1861–1868, 2011

    2011

  5. [13]

    Fast single-frequency time-of-flight range imaging

    Ryan Crabb and Roberto Manduchi. Fast single-frequency time-of-flight range imaging. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, pages 58–65, 2015

    2015

  6. [14]

    High-resolution long- distance depth imaging lidar with ultra-low timing jitter superconducting nanowire single-photon detectors

    Aongus McCarthy, Gregor G Taylor, Jorge Garcia-Armenta, Boris Korzh, Dmitry V Morozov, Andrew D Beyer, Ryan M Briggs, Jason P Allmaras, Bruce Bumble, Marco Colangelo, et al. High-resolution long- distance depth imaging lidar with ultra-low timing jitter superconducting nanowir...

    2025

  7. [15]

    Lawrence

    Mamadou Diop and Keith St. Lawrence. Improving the depth sensitivity of time-resolved measurements by extracting the distribution of times-of-flight.Biomedical Optics Express, 4(3):447–459, 2013

    2013

  8. [16]

    A 220 m-range direct time-of-flight 688× 384 cmos image sensor with sub-photon signal extraction (spse) pixels using vertical avalanche photo-diodes and 6 khz light pulse counters

    Shinzo Koyama, Motonori Ishii, Shigeru Saito, Masato Takemoto, Yugo Nose, Akito Inoue, Yusuke Sakata, Yuki Sugiura, Manabu Usuda, Tatsuya Kabe, et al. A 220 m-range direct time-of-flight 688× 384 cmos image sensor with sub-photon signal extraction (spse) pixels using vertical ...

    2018

  9. [17]

    Matteo Perenzoni, Daniele Perenzoni, and David Stoppa. A 64×64-pixels digital silicon photomultiplier direct tof sensor with 100-mphotons/s/pixel background rejection and imaging/altimeter mode with 0.14% precision up to 6 km for spacecraft navigation and landing.IEEE Journal ...

    2016

  10. [18]

    Tang Xu, Qianyu Chen, Dajing Bian, and Yue Xu. A near-infrared single-photon detector for direct time-of-flight measurement using time-to-amplitude-digital hybrid conversion method.IEEE Transactions on Instrumentation and Measurement, 73:1–9, 2023. 10

    2023

  11. [19]

    Multi-frequency phase unwrapping for time-of-flight cameras

    David Droeschel, Dirk Holz, and Sven Behnke. Multi-frequency phase unwrapping for time-of-flight cameras. InIEEE International Conference on Intelligent Robots and Systems, pages 1463–1469. IEEE, 2010

    2010

  12. [20]

    Time of flight lidar employing dual-modulation frequencies switching for optimizing unambiguous range extension and high resolution

    Dwi Hanto, Hari Pratomo, Agitta Rianaris, Andi Setiono, Sartika Sartika, Mohamad Syahadi, Eko Joni Pristianto, Dayat Kurniawan, Dwi Bayuwati, Hendra Adinanta, et al. Time of flight lidar employing dual-modulation frequencies switching for optimizing unambiguous range extension...

    2023

  13. [21]

    Deep end-to-end time-of-flight imaging

    Shuochen Su, Felix Heide, Gordon Wetzstein, and Wolfgang Heidrich. Deep end-to-end time-of-flight imaging. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 6383–6392, 2018

    2018

  14. [22]

    Deep optics for single-shot high-dynamic-range imaging

    Christopher A Metzler, Hayato Ikoma, Yifan Peng, and Gordon Wetzstein. Deep optics for single-shot high-dynamic-range imaging. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 1375–1385, 2020

    2020

  15. [23]

    Learning rank-1 diffractive optics for single-shot high dynamic range imaging

    Qilin Sun, Ethan Tseng, Qiang Fu, Wolfgang Heidrich, and Felix Heide. Learning rank-1 diffractive optics for single-shot high dynamic range imaging. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 1386–1396, 2020

    2020

  16. [24]

    Deeply learned filter response functions for hyperspectral reconstruction

    Shijie Nie, Lin Gu, Yinqiang Zheng, Antony Lam, Nobutaka Ono, and Imari Sato. Deeply learned filter response functions for hyperspectral reconstruction. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4767–4776, 2018

    2018

  17. [25]

    Deepstorm3d: dense 3d localization microscopy and psf design by deep learning.Nature Methods, 17(7):734–740, 2020

    Elias Nehme, Daniel Freedman, Racheli Gordon, Boris Ferdman, Lucien E Weiss, Onit Alalouf, Tal Naor, Reut Orange, Tomer Michaeli, and Yoav Shechtman. Deepstorm3d: dense 3d localization microscopy and psf design by deep learning.Nature Methods, 17(7):734–740, 2020

    2020

  18. [26]

    End-to-end optimization of optics and image processing for achromatic extended depth of field and super-resolution imaging.ACM Transactions on Graphics, 37(4):1–13, 2018

    Vincent Sitzmann, Steven Diamond, Yifan Peng, Xiong Dun, Stephen Boyd, Wolfgang Heidrich, Felix Heide, and Gordon Wetzstein. End-to-end optimization of optics and image processing for achromatic extended depth of field and super-resolution imaging.ACM Transactions on Graphics,...

    2018

  19. [27]

    End-to-end fluorescence lifetime imaging with optimized encoding and exposure allocation

    Ziyi Guo, Jiaqu Li, Kanghui Wang, Tao Yue, and Xuemei Hu. End-to-end fluorescence lifetime imaging with optimized encoding and exposure allocation. InIEEE International Conference on Computational Photography, pages 1–12. IEEE, 2024

    2024

  20. [28]

    Mask-tof: Learning microlens masks for flying pixel correction in time-of-flight imaging

    Ilya Chugunov, Seung-Hwan Baek, Qiang Fu, Wolfgang Heidrich, and Felix Heide. Mask-tof: Learning microlens masks for flying pixel correction in time-of-flight imaging. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 9116–9126, 2021

    2021

  21. [29]

    Coded exposure deblurring: Optimized codes for psf estimation and invertibility

    Amit Agrawal and Yi Xu. Coded exposure deblurring: Optimized codes for psf estimation and invertibility. In2009 IEEE Conference on Computer Vision and Pattern Recognition, pages 2066–2073. IEEE, 2009

    2009

  22. [30]

    Light fall-off stereo

    Miao Liao, Liang Wang, Ruigang Yang, and Minglun Gong. Light fall-off stereo. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 1–8. IEEE, 2007

    2007

  23. [31]

    Restormer: Efficient transformer for high-resolution image restoration

    Syed Waqas Zamir, Aditya Arora, Salman Khan, Munawar Hayat, Fahad Shahbaz Khan, and Ming-Hsuan Yang. Restormer: Efficient transformer for high-resolution image restoration. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 5728–5739, 2022

    2022

  24. [32]

    Eca-net: Efficient channel attention for deep convolutional neural networks

    Qilong Wang, Banggu Wu, Pengfei Zhu, Peihua Li, Wangmeng Zuo, and Qinghua Hu. Eca-net: Efficient channel attention for deep convolutional neural networks. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 11534–11542, 2020

    2020

  25. [33]

    The double-well potential in quantum mechanics: a simple, numerically exact formulation.European Journal of Physics, 33(6):1651, 2012

    V Jelic and F Marsiglio. The double-well potential in quantum mechanics: a simple, numerically exact formulation.European Journal of Physics, 33(6):1651, 2012

    2012

  26. [34]

    Indoor segmentation and support inference from rgbd images

    Nathan Silberman, Derek Hoiem, Pushmeet Kohli, and Rob Fergus. Indoor segmentation and support inference from rgbd images. InProceedings of the European Conference on Computer Vision, pages 746–760. Springer, 2012

    2012

  27. [35]

    Intrinsic image decomposition using structure- texture separation and surface normals

    Junho Jeon, Sunghyun Cho, Xin Tong, and Seungyong Lee. Intrinsic image decomposition using structure- texture separation and surface normals. InProceedings of the European Conference on Computer Vision, pages 218–233. Springer, 2014

    2014

  28. [36]

    Deformable kernel networks for joint image filtering

    Beomjun Kim, Jean Ponce, and Bumsub Ham. Deformable kernel networks for joint image filtering. International Journal of Computer Vision, 129(2):579–600, 2021. 11

    2021

  29. [37]

    Towards fast and accurate real-world depth super-resolution: Benchmark dataset and baseline

    Lingzhi He, Hongguang Zhu, Feng Li, Huihui Bai, Runmin Cong, Chunjie Zhang, Chunyu Lin, Meiqin Liu, and Yao Zhao. Towards fast and accurate real-world depth super-resolution: Benchmark dataset and baseline. InProceedings of the IEEE Conference on Computer Vision and Pattern Re...

    2021

  30. [38]

    Curriculum learning

    Yoshua Bengio, Jérôme Louradour, Ronan Collobert, and Jason Weston. Curriculum learning. In Proceedings of the International Conference on Machine Learning, pages 41–48, 2009

    2009

  31. [39]

    Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  32. [40]

    Automatic differentiation in pytorch

    Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in pytorch. 2017. 12

    2017

Pith tools

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