Pith. sign in

REVIEW 3 major objections 4 minor 38 references

MetaH2: A Snapshot Metasurface HDR Hyperspectral Camera

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A metasurface camera can capture a high-dynamic-range image and a hyperspectral datacube in a single snapshot, with a working prototype reporting 60 dB dynamic range and 10 nm spectral resolution from 600 nm to 700 nm.

desk verdict A genuinely new metasurface encoding for joint HDR+hyperspectral snapshot imaging, with a solid simulation study, but the prototype validation is too thin to support the headline numbers. read the letter →

arxiv 2507.08282 v2 pith:J5UICMEM submitted 2025-07-11 eess.IV

Yuxuan Liu , Qi Guo This is my paper · ORCID
classification eess.IV
keywords metasurfacesnapshothyperspectralimaginghighdynamicrangecomputedtomographyspectrometrydeepreconstructionexposurebracketingcomputational
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

MetaH2 is a computational camera that fuses high-dynamic-range (HDR) and hyperspectral imaging into one exposure. The camera uses a metasurface—a flat optical surface patterned at the nanoscale—that splits incoming light into nine copies, each shifted and attenuated by a known power ratio, so a single photosensor frame records both bracketed exposures and spectrally dispersed projections. A deep reconstruction network then inverts these measurements into an HDR image and a hyperspectral datacube. In simulations, the paper reports higher reconstruction accuracy than previous snapshot hyperspectral systems on two benchmark datasets, and a working prototype achieves 60 dB dynamic range with 10 nm spectral resolution over the 600–700 nm band.

What carries the argument

The load-bearing object is the metasurface point spread function h(u,v,λ)=Σ_{i=1}^V α_i h(u−u_i(λ), v−v_i(λ)), with α_i=(1/2)^i. The metasurface is built by interleaving V deflective phase profiles, each designed at the central wavelength with a linear phase ramp whose direction cosines set the spot positions (u_i(λ), v_i(λ)); a multinomial selection of nanocells with probabilities proportional to √α_i distributes energy among the spots. The wavelength-dependent displacement of each spot turns chromatic aberration into a spectral encoder, while the geometric power ratio implements exposure bracketing. Reconstruction combines per-sub-image feature hallucination, a Wiener deconvolution across the nine point spread functions, and a U-Net refinement to output the hyperspectral datacube, each spectral slice of which is an HDR image.

What would settle it

Measure the fabricated metasurface's point spread function with a tunable laser across 600–700 nm and compare spot positions, widths, and power ratios against the designed model; if the mismatch exceeds what the reconstruction network tolerates, the claimed 60 dB dynamic range and 10 nm spectral resolution will not reproduce.

Watch

Extended reading notes

Core claim

The paper's central claim is that one metasurface can encode both scene brightness range and spectral content into a single sensor readout, and that a learned decoder can recover both simultaneously. The encoding is captured by a designed point spread function h(u,v,λ)=Σ α_i h(u−u_i(λ), v−v_i(λ)), where each of the V sub-images carries a power ratio α_i=(1/2)^i and a wavelength-dependent spatial shift, reproducing exposure bracketing and computed-tomography imaging spectrometry (CTIS) at the same time. The authors report that the resulting nine sub-images yield consistently higher hyperspectral reconstruction accuracy than single-disperser CASSI, the dual-camera design, and the split-aperture 2-in-1 camera across four reconstruction algorithms and two benchmark datasets, at the cost of reduced field of view. On the hardware side, a lab prototype with a 1-mm metasurface and a monochrome sensor reconstructs real scenes at 60 dB dynamic range and 10 nm spectral resolution from 600 nm to 700 nm.

Load-bearing premise

The fabricated metasurface's actual point spread function matches the designed nine-spot model closely enough in spot position, width, and power ratio that the trained reconstruction network can invert it, and the paper itself reports residual light below 600 nm that forced a bandpass filter, narrowing the claimed operating range.

Editorial extensions

If this is right

  • A metasurface HDR imager can operate over a roughly 100 nm band instead of the typical 10 nm band of prior metasurface HDR cameras, because chromaticity is treated as signal rather than error.
  • Because HDR and spectral information are captured in one frame, the approach applies to dynamic scenes where sequential captures would misalign.
  • Reconstruction accuracy scales with the number of sub-images, so a user can trade field of view for spectral fidelity by choosing V.
  • The same nine-channel measurement supports multiple off-the-shelf hyperspectral reconstruction networks, not only the specific deep decoder used in the paper.
  • The camera delivers HDR and spectral data from a monochrome sensor without moving parts or multiple exposures, simplifying the optical system.

Reading between the lines

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

  • A natural next test, not performed in the paper, is measuring the fabricated metasurface's PSF and comparing spot positions and power ratios to the designed model; reconstruction accuracy should degrade predictably as that mismatch grows.
  • The exposure-bracketing range could be widened beyond α_i=(1/2)^i, though extreme ratios would compress low-light channels and likely require noise-aware training.
  • Because the encoder is a linear PSF model, the same metasurface could be paired with a task-specific learned reconstruction, such as material classification, rather than generic datacube fidelity.
  • The residual-light failure below 600 nm in the prototype suggests that fabrication-aware design, such as optimizing the metasurface with measured phase errors, could recover the full visible band without a bandpass filter.
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 / 4 minor

Summary. The paper proposes MetaH2, a snapshot metasurface camera that multiplexes nine exposures with distinct power ratios and chromatic dispersions onto a single photosensor, combining exposure bracketing with computed tomography imaging spectrometry (CTIS). A deep Wiener-deconvolution network is used to reconstruct an HDR image and a hyperspectral datacube from the single measurement. The authors report simulation comparisons on the ICVL and Harvard benchmark datasets against SD-CASSI, DCD, the 2-in-1 camera, and four reconstruction algorithms, and they describe a lab prototype with claimed 60 dB dynamic range and 10 nm spectral resolution over a 600-700 nm working band.

Significance. If the prototype claims were supported by adequate measurements, this would be a meaningful advance: joint HDR and hyperspectral imaging in a single snapshot with a compact metasurface is timely, and the simulation study provides a useful head-to-head comparison under controlled assumptions. The authors deserve credit for explicitly acknowledging the field-of-view versus number-of-sub-images tradeoff and for comparing against multiple reconstruction algorithms. However, the real-world quantitative claims currently rest on unverified assumptions about PSF fidelity and on missing calibration measurements, so the significance of the work depends on evidence that the paper does not yet provide.

major comments (3)
  1. [Section 4.2, Fig. 4] The headline prototype claims of 60 dB dynamic range and 10 nm spectral resolution are not accompanied by any measurement protocol. There is no spectral ground truth (e.g., a monochromator scan or a reference spectrometer), no radiometric calibration target with known radiance, no error bars, and no description of how the inset numbers in Fig. 4 were computed. The claim would be circular if the 60 dB figure is derived from the designed power ratios alpha_i=(1/2)^i rather than from a calibrated measurement. Please report a calibration experiment, including photon counts, noise floor, and a statement of how the two headline numbers are measured.
  2. [Section 3.1, Eq. (1); Section 4.2] The reconstruction pipeline assumes the fabricated PSF follows Eq. (1) with the designed weights alpha_i. Section 4.2 reports significant residual light in the central sub-image when the incident wavelength is shorter than 600 nm, leading to a disproportionately high central energy and forcing a bandpass filter to [600,700] nm. This is direct evidence that the fabricated device deviates from the design model, at least outside the operating band. Since no measured wavelength-resolved PSF or measured alpha_i values are reported for the [600,700] nm band, the fidelity of Eq. (1) for the prototype is unverified. Because Eq. (6) uses the designed h_i and the network is trained on the ideal forward model, any unmodeled PSF or energy-ratio error inside the operating band would be absorbed as reconstruction bias. Please provide a measured PSF comparison against Eq. (1) or a perturbation analysis showing reconstruction robustness to PSF and alpha mismatch.
  3. [Section 4.1, Table 1] The claim of universally higher reconstruction accuracy is based on 9 ICVL and 50 Harvard test images, but Table 1 reports only point estimates of PSNR and SSIM. The improvements over the 2-in-1 Camera are occasionally small (e.g., 31.62 dB vs 31.42 dB for DWDN on ICVL), and the comparison is confounded by the number of sub-images (9 vs 1-2) at a fixed sensor size. Report standard deviations or confidence intervals across test images, and clarify how the field-of-view tradeoff is normalized when claiming a universal advantage.
minor comments (4)
  1. [Eq. (4)] The symbol ⊙ is defined as 2D convolution but usually denotes elementwise multiplication; please use a standard convolution symbol, such as ∗, or define the operator consistently with its use in Eq. (5).
  2. [Eq. (3)] The notation 'i ∼ Multinomial(...)' is confusing because i is used both as the index over sub-images and as the random variable; please use a different symbol for the random index, for example j ∼ Multinomial(...).
  3. [Section 4.2] Please specify the type of bandpass filter used in the prototype and state whether the claimed 10 nm spectral resolution is an intrinsic property of the optics, of the reconstruction algorithm, or of the test target; the text currently asserts the number without any derivation or measurement description.
  4. [Fig. 4] The inset numbers in Fig. 4 have no units or axis labels, and the displayed spectra do not indicate wavelength and intensity scales; please label the axes and explain what the inset dynamic-range number represents.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the simulation benchmark is externally grounded and the prototype limitations are empirical validation gaps, not self-referential reductions.

full rationale

The paper's claimed derivation chain is an optical forward model (Eq. 1), an image formation model (Eqs. 4-5), a reconstruction algorithm (Eq. 6), and evaluation on held-out ICVL and Harvard test images. The power ratios alpha_i=(1/2)^i in Eq. 3 are stated design parameters, not quantities fitted to the reconstruction output; the reported simulation accuracy is obtained on test images not used in training and is compared against independently published CASSI, DCD, and 2-in-1 camera models. The prototype claims of 60 dB dynamic range and 10 nm spectral resolution are not accompanied by independent ground-truth spectral or radiometric characterization, and the acknowledged residual-light deviation below 600 nm (Section 4.2) is a real model-device mismatch. However, these are missing-evidence and generalizability concerns, not circularity: nothing in the paper defines the measured dynamic range as the designed exposure ratio, nor does any equation reduce its prediction to its input. The overlapping-author citations (Brookshire et al. [7] for the interleaved metasurface design and D-Flat [25] for simulation) are used as published engineering tools with stated assumptions, and the novel claim of jointly encoding HDR and hyperspectral information is tested on standard benchmarks rather than being justified solely by those citations. Therefore no circular step can be exhibited from the text under the strict reduction test.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim rests on several hand-chosen design parameters (V, alpha_i, working bandwidth) and on assumptions that the fabricated metasurface matches the idealized PSF model and that simulation-trained networks transfer to real hardware. No fundamentally new physical entity is introduced; the metasurface is a fabricated device, not a new particle or force.

free parameters (4)
  • V (number of beam-split directions) = 9
    Chosen by hand; determines the number of sub-images and the field-of-view tradeoff.
  • power ratios alpha_i = alpha_i = (1/2)^i for i = 1,...,9
    Exponentially decreasing power ratios chosen by hand to bracket a wide dynamic range; not fitted to data.
  • working bandwidth = [600 nm, 700 nm]
    Set post-hoc by inserting a bandpass filter after observing residual light below 600 nm (Section 4.2).
  • spectral resolution claim = 10 nm
    Asserted for the prototype in Section 4.2 with no measurement procedure described.
assumptions (3)
  • domain assumption The metasurface can be modeled as a random interleaving of V independent deflective metasurfaces with phase profiles given by Eq. (2), producing the PSF in Eq. (1).
    Inherited from [7]; the fabricated device is assumed to follow this model closely enough for the reconstruction to succeed.
  • domain assumption The image formation follows Eq. (5) with known PSFs and constant photon efficiency over the working bandwidth.
    Used in simulation; the prototype deviates due to residual light and fabrication defects, as acknowledged in Section 4.2.
  • domain assumption The deep reconstruction network trained on simulated data generalizes to real measurements from the prototype.
    No calibration or domain adaptation is described; the network is assumed to transfer across the simulation-to-real gap.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MetaH2: A Snapshot Metasurface HDR Hyperspectral Camera." pith.science (2026). https://pith.science/paper/J5UICMEM

@misc{pith2026250708282,
  author       = {Pith},
  title        = {Pith review of: MetaH2: A Snapshot Metasurface HDR Hyperspectral Camera},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J5UICMEM}},
  note         = {Machine review of arXiv:2507.08282}
}
read the original abstract

We present a metasurface camera that jointly performs high-dynamic range (HDR) and hyperspectral imaging in a snapshot. The system integrates exposure bracketing and computed tomography imaging spectrometry (CTIS) by simultaneously forming multiple spatially multiplexed projections with unique power ratios and chromatic aberrations on a photosensor. The measurements are subsequently processed through a deep reconstruction model to generate an HDR image and a hyperspectral datacube. Our simulation studies show that the proposed system achieves higher reconstruction accuracy than previous snapshot hyperspectral imaging methods on benchmark datasets. We assemble a working prototype and demonstrate snapshot reconstruction of 60 dB dynamic range and 10 nm spectral resolution from 600 nm to 700 nm on real-world scenes from a monochrome photosensor.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 37 canonical work pages

  1. [1]

    INTRODUCTION The high-dynamic range (HDR) and hyperspectral imaging can record an object’s full brightness or spectral profiles, pro- viding critical insights into its intrinsic properties, including material composition, shape curvature, and chemical signa- ture. Traditionally, either approach requires recording multi- ple sequential measurements with va...

    work page Pith review arXiv 2025

  2. [2]

    Past efforts to snapshot HDR imag- ing have mostly focused on engineering the photosensors

    RELA TED WORK Snapshot HDR imaging. Past efforts to snapshot HDR imag- ing have mostly focused on engineering the photosensors. People have demonstrated photodetectors with mosaicked optical density [8], exposure [9], or polarization [10]. Peo- ple have also exploited novel photon transducers, including single-photon avalanche diode [11], modulo camera [1...

    work page

  3. [3]

    Optical Design Our optical design is illustrated in Fig

    IMAGE FORMA TION 3.1. Optical Design Our optical design is illustrated in Fig. 1. It uses an achro- matic refractive lens paired with a polarization-insensitiveV - beamsplitting metasurface as the optical assembly. The work- ing bandwidth of the system is [λlow, λhigh]. Its point spread function (PSF) at wavelength λ is: h(u, v, λ) = VX i=1 αih(u − ui(λ),...

    work page

  4. [4]

    Snapshot Hyperspectral Imaging Comparison First, we want to analyze how the proposed optical modality compares with previous snapshot hyperspectral imaging hard- ware

    EXPERIMENTAL RESULTS 4.1. Snapshot Hyperspectral Imaging Comparison First, we want to analyze how the proposed optical modality compares with previous snapshot hyperspectral imaging hard- ware. We simulate measurements from the single disperser CASSI (SD-CASSI) and its dual camera design (DCD) [28], the split aperture 2-in-1 Camera [26], and our system wi...

    work page

  5. [5]

    Our metasurface encodes both types of information into Fig

    CONCLUSION In this paper, we introduce MetaH2, a metasurface-based camera designed for joint hyperspectral and HDR imag- ing. Our metasurface encodes both types of information into Fig. 3: Visual comparison between MetaH2 and previous snapshot hyperspectral imaging modalities on the ICVL dataset. We visualize the raw measurements, the output image at 550 ...

    work page

  6. [6]

    Recovering high dynamic range radiance maps from photographs,

    Paul E Debevec and Jitendra Malik, “Recovering high dynamic range radiance maps from photographs,” in Seminal Graphics Papers: Push- ing the Boundaries, V olume 2, pp. 643–652. 2023

    work page 2023

  7. [7]

    Design of pushbroom imaging spectrometers for optimum recovery of spectro- scopic and spatial information,

    Pantazis Mouroulis, Robert O Green, and Thomas G Chrien, “Design of pushbroom imaging spectrometers for optimum recovery of spectro- scopic and spatial information,” Applied Optics , vol. 39, no. 13, pp. 2210–2220, 2000

    work page 2000

  8. [8]

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

    Christopher A Metzler, Hayato Ikoma, Yifan Peng, and Gordon Wet- zstein, “Deep optics for single-shot high-dynamic-range imaging,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, 2020, pp. 1375–1385

    work page 2020

Show all 38 references
  1. [9]

    Compact single-shot hyperspectral imaging using a prism,

    Seung-Hwan Baek, Incheol Kim, Diego Gutierrez, and Min H Kim, “Compact single-shot hyperspectral imaging using a prism,” ACM Transactions on Graphics (TOG), vol. 36, no. 6, pp. 1–12, 2017

    2017

  2. [10]

    Simultaneous acquisition of spectral image information,

    Takayuki Okamoto and Ichirou Yamaguchi, “Simultaneous acquisition of spectral image information,”Optics letters, vol. 16, no. 16, pp. 1277– 1279, 1991

    1991

  3. [11]

    Optical metalenses: funda- mentals, dispersion manipulation, and applications,

    Yongli He, Boxiang Song, and Jiang Tang, “Optical metalenses: funda- mentals, dispersion manipulation, and applications,” Frontiers of Op- toelectronics, vol. 15, no. 1, pp. 24, 2022

    2022

  4. [12]

    Metahdr: single shot high-dynamic range imaging and sens- ing using a multifunctional metasurface,

    Charles Brookshire, Yuxuan Liu, Yuanrui Chen, Wei Ting Chen, and Qi Guo, “Metahdr: single shot high-dynamic range imaging and sens- ing using a multifunctional metasurface,” Optics Express, vol. 32, no. 15, pp. 26690–26707, 2024

    2024

  5. [13]

    High dynamic range imaging: Spatially varying pixel exposures,

    Shree K Nayar and Tomoo Mitsunaga, “High dynamic range imaging: Spatially varying pixel exposures,” in Proceedings IEEE Conference on Computer Vision and Pattern Recognition. CVPR 2000 (Cat. No. PR00662). IEEE, 2000, vol. 1, pp. 472–479

    2000

  6. [14]

    Single-shot high dynamic range imaging using coded electronic shutter,

    Hojin Cho, Seon Joo Kim, and Seungyong Lee, “Single-shot high dynamic range imaging using coded electronic shutter,” in Computer Graphics F orum. Wiley Online Library, 2014, vol. 33, pp. 329–338

    2014

  7. [15]

    Hdr reconstruction based on the polarization cam- era,

    Xuesong Wu, Hong Zhang, Xiaoping Hu, Moein Shakeri, Chen Fan, and Juiwen Ting, “Hdr reconstruction based on the polarization cam- era,” IEEE Robotics and Automation Letters , vol. 5, no. 4, pp. 5113– 5119, 2020

    2020

  8. [16]

    A 124-db dynamic-range spad photon-counting image sensor using subframe sampling and extrapo- lating photon count,

    Jun Ogi, Takafumi Takatsuka, Kazuki Hizu, Yutaka Inaoka, Hongbo Zhu, Yasuhisa Tochigi, Yoshiaki Tashiro, Fumiaki Sano, Yusuke Mu- rakawa, Makoto Nakamura, et al., “A 124-db dynamic-range spad photon-counting image sensor using subframe sampling and extrapo- lating photon count...

    2021

  9. [17]

    Mantissacam: Learning snapshot high-dynamic-range imaging with perceptually-based in-pixel irradiance encoding,

    Haley M So, Julien NP Martel, Gordon Wetzstein, and Piotr Dudek, “Mantissacam: Learning snapshot high-dynamic-range imaging with perceptually-based in-pixel irradiance encoding,” in 2022 IEEE In- ternational Conference on Computational Photography (ICCP) . IEEE, 2022, pp. 1–12

    2022

  10. [18]

    Hdr imaging with quanta image sensors: Theoretical limits and optimal reconstruction,

    Abhiram Gnanasambandam and Stanley H Chan, “Hdr imaging with quanta image sensors: Theoretical limits and optimal reconstruction,” IEEE transactions on computational imaging , vol. 6, pp. 1571–1585, 2020

    2020

  11. [19]

    Imaging spectroscopy using tunable filters: a review,

    Nahum Gat, “Imaging spectroscopy using tunable filters: a review,” Wavelet Applications VII, vol. 4056, pp. 50–64, 2000

    2000

  12. [20]

    Single disperser design for coded aperture snapshot spectral imaging,

    Ashwin Wagadarikar, Renu John, Rebecca Willett, and David Brady, “Single disperser design for coded aperture snapshot spectral imaging,” Applied optics, vol. 47, no. 10, pp. B44–B51, 2008

    2008

  13. [21]

    Single-shot hyperspectral-depth imaging with learned diffractive optics,

    Seung-Hwan Baek, Hayato Ikoma, Daniel S Jeon, Yuqi Li, Wolf- gang Heidrich, Gordon Wetzstein, and Min H Kim, “Single-shot hyperspectral-depth imaging with learned diffractive optics,” in Pro- ceedings of the IEEE/CVF International Conference on Computer Vi- sion, 2021, pp. 2651–2660

    2021

  14. [22]

    Miniaturized structured illumination microscopy with diffractive optics,

    Guoxuan Liu, Ning Xu, Huaidong Yang, Qiaofeng Tan, and Guofan Jin, “Miniaturized structured illumination microscopy with diffractive optics,” Photonics Research, vol. 10, no. 5, pp. 1317–1324, 2022

    2022

  15. [23]

    Practical spec- tral photography,

    Ralf Habel, Michael Kudenov, and Michael Wimmer, “Practical spec- tral photography,” in Computer graphics forum. Wiley Online Library, 2012, vol. 31, pp. 449–458

    2012

  16. [24]

    Monocu- lar depth sensing using metalens,

    Fan Yang, Hung-I Lin, Peng Chen, Juejun Hu, and Tian Gu, “Monocu- lar depth sensing using metalens,” Nanophotonics, vol. 12, no. 14, pp. 2987–2996, 2023

    2023

  17. [25]

    Monocular metasurface camera for passive single-shot 4d imaging,

    Zicheng Shen, Feng Zhao, Chunqi Jin, Shuai Wang, Liangcai Cao, and Yuanmu Yang, “Monocular metasurface camera for passive single-shot 4d imaging,” Nature Communications, vol. 14, no. 1, pp. 1035, 2023

    2023

  18. [26]

    Plasmonic lattice lenses for multiwavelength achro- matic focusing,

    Jingtian Hu, Chang-Hua Liu, Xiaochen Ren, Lincoln J Lauhon, and Teri W Odom, “Plasmonic lattice lenses for multiwavelength achro- matic focusing,” ACS nano, vol. 10, no. 11, pp. 10275–10282, 2016

    2016

  19. [27]

    High-efficiency, large-area, topology-optimized metasurfaces,

    Thaibao Phan, David Sell, Evan W Wang, Sage Doshay, Kofi Edee, Jianji Yang, and Jonathan A Fan, “High-efficiency, large-area, topology-optimized metasurfaces,” Light: Science & Applications , vol. 8, no. 1, pp. 48, 2019

    2019

  20. [28]

    Learned multi-aperture color-coded optics for snapshot hyperspectral imaging,

    Zheng Shi, Xiong Dun, Haoyu Wei, Siyu Dong, Zhanshan Wang, Xin- bin Cheng, Felix Heide, and Yifan Peng, “Learned multi-aperture color-coded optics for snapshot hyperspectral imaging,” ACM Trans- actions on Graphics (TOG), vol. 43, no. 6, pp. 1–11, 2024

    2024

  21. [29]

    Grayscale to hyperspectral at any resolution using a phase-only lens,

    Dean Hazineh, Federico Capasso, and Todd Zickler, “Grayscale to hyperspectral at any resolution using a phase-only lens,”arXiv preprint arXiv:2412.02798, 2024

    2024 arXiv

  22. [30]

    D-flat: A differentiable flat-optics frame- work for end-to-end metasurface visual sensor design,

    Dean S Hazineh, Soon Wei Daniel Lim, Zhujun Shi, Federico Capasso, Todd Zickler, and Qi Guo, “D-flat: A differentiable flat-optics frame- work for end-to-end metasurface visual sensor design,” arXiv preprint arXiv:2207.14780, 2022

    2022 arXiv

  23. [31]

    Split- aperture 2-in-1 computational cameras,

    Zheng Shi, Ilya Chugunov, Mario Bijelic, Geoffroi Cˆot´e, Jiwoon Yeom, Qiang Fu, Hadi Amata, Wolfgang Heidrich, and Felix Heide, “Split- aperture 2-in-1 computational cameras,” ACM Transactions on Graph- ics (TOG), vol. 43, no. 4, pp. 1–19, 2024

    2024

  24. [32]

    Deep wiener deconvo- lution: Wiener meets deep learning for image deblurring,

    Jiangxin Dong, Stefan Roth, and Bernt Schiele, “Deep wiener deconvo- lution: Wiener meets deep learning for image deblurring,” Advances in Neural Information Processing Systems, vol. 33, pp. 1048–1059, 2020

    2020

  25. [33]

    Dual-camera design for coded aperture snapshot spectral imaging,

    Lizhi Wang, Zhiwei Xiong, Dahua Gao, Guangming Shi, and Feng Wu, “Dual-camera design for coded aperture snapshot spectral imaging,” Applied optics, vol. 54, no. 4, pp. 848–858, 2015

    2015

  26. [34]

    l-net: Re- construct hyperspectral images from a snapshot measurement,

    Xin Miao, Xin Yuan, Yunchen Pu, and Vassilis Athitsos, “l-net: Re- construct hyperspectral images from a snapshot measurement,” in Pro- ceedings of the IEEE/CVF International Conference on Computer Vi- sion, 2019, pp. 4059–4069

    2019

  27. [35]

    Coded hyperspec- tral image reconstruction using deep external and internal learning,

    Ying Fu, Tao Zhang, Lizhi Wang, and Hua Huang, “Coded hyperspec- tral image reconstruction using deep external and internal learning,” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 44, no. 7, pp. 3404–3420, 2021

    2021

  28. [36]

    End-to-end low cost compres- sive spectral imaging with spatial-spectral self-attention,

    Ziyi Meng, Jiawei Ma, and Xin Yuan, “End-to-end low cost compres- sive spectral imaging with spatial-spectral self-attention,” in European conference on computer vision. Springer, 2020, pp. 187–204

    2020

  29. [37]

    Sparse recovery of hyperspectral signal from natural rgb images,

    Boaz Arad and Ohad Ben-Shahar, “Sparse recovery of hyperspectral signal from natural rgb images,” inComputer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part VII 14. Springer, 2016, pp. 19–34

    2016

  30. [38]

    Statistics of real-world hyper- spectral images,

    Ayan Chakrabarti and Todd Zickler, “Statistics of real-world hyper- spectral images,” in CVPR 2011. IEEE, 2011, pp. 193–200

    2011

Pith tools

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