Pith. sign in

REVIEW 4 major objections 4 minor 49 references

Scene-driven selection of Hadamard patterns—based on a small initial block of measurements—improves single-pixel spectral image reconstruction compared with fixed orderings, in simulation and on a real near-infrared testbed.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

A scene-driven neural network orders Hadamard sensing patterns for single-pixel spectral imaging, improving VIS/NIR reconstructions over fixed orderings at low sampling ratios.

T0 review reviewed 2026-08-02 challenge →

load-bearing objection A solid, incremental adaptive-sensing paper with consistent gains over fixed Hadamard orderings; the untested oracle-support training target is a real but addressable weakness, not a fatal one. the 4 major comments →

arxiv 2607.15045 v1 pith:QPJ2C2EJ submitted 2026-07-16 eess.IV

Deep Scene-Driven Ordering of Hadamard Basis for Single-Pixel Spectral Imaging

Brayan Monroy , Hans Garcia , Henry Arguello , Jorge Bacca This is my paper
classification eess.IV MSC 68U1094A0868T07
keywords Hadamard single-pixel imagingspectral imagingadaptive sensingscene-driven samplingdeep learningcompressed sensingnear-infrared imagingbinary modulation patterns
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 sets out to show that a single-pixel spectral imager does not have to rely on a fixed, scene-independent ordering of Hadamard modulation patterns. Instead, after capturing a small predetermined block of low-frequency Hadamard coefficients, a neural network predicts which of the remaining coefficients carry the most scene energy, and those patterns are measured next. The authors claim this scene-adaptive selection consistently beats conventional orderings (Sequency, Zig-Zag, XY, Cake-Cutting) at sampling ratios from 5% to 30% in terms of PSNR, SSIM, and spectral angle, and that the learned selector transfers from satellite imagery to a real NIR testbed without retraining. If correct, this offers a practical path to faster, higher-quality spectral imaging on existing DMD hardware.

Core claim

The central claim is that an end-to-end trained binary classifier can convert a small set of low-frequency Hadamard measurements into a scene-dependent ordering of the remaining Hadamard basis, and that this ordering improves reconstruction quality. The network P_theta receives the predetermined measurements Y_p and outputs a support mask m_a over unmeasured coefficients; it is trained with binary cross-entropy against an oracle support m_s defined by the top-k_a remaining coefficients ranked by per-band aggregated magnitude |Y|_Sigma. Because the sensing patterns stay strict Hadamard ±1 rows, the method remains DMD-compatible, and because the ordering is a property of the measurement set ra

What carries the argument

The load-bearing mechanism is the two-stage acquisition protocol with a learned support predictor. Stage one selects k_p Hadamard rows via a predetermined ordering (typically Zig-Zag); stage two computes m_a = P_theta(Y_p), masks out already-sensed rows, picks the k_a largest remaining sigmoid scores, and acquires those rows. The training target is the oracle sparse support m_s = top-k of |Y|_Sigma, and the loss is plain binary cross-entropy. This turns 'which patterns next?' into a binary classification problem whose labels come from coefficient magnitude, and it lets the sampling budget be controlled by construction.

Load-bearing premise

The whole training signal rests on the premise that the largest-magnitude remaining Hadamard coefficients, aggregated over spectral bands, are the most informative coefficients to acquire for the reconstruction solvers used—a premise the paper does not prove and that the modest F1 score (about 0.68) suggests is only partially learned.

What would settle it

Take a set of small scenes, fully measure their Hadamard spectra, and for each scene exhaustively search all k_a-subsets of unmeasured coefficients to find the subset that actually minimizes reconstruction error for a given solver. If that optimal subset consistently differs from the top-magnitude oracle support m_s, or if reconstructing from m_s is beaten by an equally-sized alternative subset, the paper's central premise fails and the learned selector is chasing the wrong target.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Fixed Hadamard orderings leave reconstruction quality on the table; a scene-aware selection of the same physical patterns recovers up to roughly 2 dB more PSNR at low sampling ratios.
  • The learned ordering is reconstruction-agnostic: the same selected mask improves a simple transpose backprojection and sophisticated iterative solvers, meaning hardware gains do not depend on a particular algorithm.
  • The selector trained on satellite spectral images transfers to a different near-infrared single-pixel camera without fine-tuning, suggesting the learned criterion is about scene statistics rather than dataset-specific details.
  • Because the method only reorders existing Hadamard patterns, it can be layered onto high-speed cyclic-mask or learned-pattern systems to combine per-pattern speed with compressive efficiency.
  • Mask-level metrics (accuracy ~0.88–0.91, F1 ~0.68–0.70) show the support predictor is informative but imperfect, so reconstruction quality may improve further with better supervision or richer architectures.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The paper's oracle target—largest-magnitude remaining Hadamard coefficients—is a proxy for reconstruction value, not a proof of optimality; a loss that measures downstream reconstruction error directly could yield a different and possibly better support.
  • The two-stage idea extends naturally to multi-stage or video settings: once the first adaptive block is measured, the predictor could be re-applied to the growing set of coefficients to refine the remaining budget.
  • The approach should carry over to other orthogonal bases (e.g., Fourier or Walsh) for which binary or sign-binary patterns are implementable, since the machinery only requires a fast transform and a magnitude-ranked oracle.
  • The reported F1 of ~0.68 suggests a ceiling imposed by the target definition; if the field adopted a reconstruction-aware oracle, the same architecture might show larger apparent gains.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes a scene-driven adaptive ordering of Hadamard patterns for single-pixel spectral imaging. Acquisition is split into a fixed predetermined block (Zig-Zag) and a scene-driven block whose support is predicted by a ResNet-based selector from the initial low-frequency measurements. The selector is trained with binary cross-entropy against an oracle support formed by the largest per-band aggregated Hadamard magnitudes. Reconstruction is performed by transpose, plug-and-play, or consensus-equilibrium solvers. Experiments on EuroSAT and ARAD report consistent PSNR/SSIM/SAM gains over fixed orderings, and an eight-scene NIR testbed with a EuroSAT-trained selector is used to claim cross-domain transferability.

Significance. If the claims hold, the work provides a practical way to make Hadamard single-pixel acquisition content-adaptive while preserving DMD-compatible binary patterns and reconstruction-agnostic sensing. The idea is well motivated and the two-stage formulation is clear. The paper ships code, compares against several fixed orderings, and includes supplementary ablations on loss functions and sampling-ratio splits. The main value is in demonstrating a learned, scene-specific selection of Hadamard coefficients that improves over fixed orderings at low sampling budgets and is transferable to a real NIR setup. However, the central evidence for optimality and for real adaptive acquisition is not yet conclusive: the training target is a magnitude-based proxy that is not validated against reconstruction quality, and the real testbed is an offline emulation rather than an online two-stage acquisition.

major comments (4)
  1. [Sec. IV.B, Eqs. (10)-(11); Sec. V.B] The training signal is the oracle support m_s = ktop(M_r^p |Y|_Sigma, k_a), i.e., the largest aggregated Hadamard magnitudes. No experiment shows that this proxy is aligned with what the nonlinear solvers (PnP, TV, CE) actually need. The reported F1≈0.68 means the selector only partially recovers this proxy. The ablations in Suppl. IX.A compare loss functions but never evaluate final reconstruction PSNR/SSIM/SAM against the oracle support itself or against a support optimized for the final solver. Without such a comparison, the claimed optimality of the learned policy is not established, and the transferability claim rests on an untested assumption. I ask for an experiment where the oracle support is replaced by (i) a random support of the same size and (ii) a support selected by greedy reconstruction-error minimization for the CE solver, then compare final metrics.
  2. [Suppl. XI.A; Sec. VI] The real testbed experiments are described as 'real-time reconstruction' and 'online operation', but the protocol acquires the complete Hadamard set once and then numerically permutes coefficients according to each ordering. This is an emulation of adaptive sensing on real measurements, not a true two-stage scene-driven acquisition. The claim in the Abstract and Conclusions of validation on 'real test-bed acquisitions' is therefore overstated. Please clarify this explicitly in the main text, and, if feasible, implement the actual two-stage DMD acquisition for at least one sampling ratio to demonstrate the hardware compatibility claim.
  3. [Sec. V.A; Tables 2-4; Suppl. X (Fig. 10)] The relative predetermined ratio δ_p = k_p/k used for the main results is not reported. Supplementary Fig. 10 shows that reconstruction quality depends strongly on δ_p, with an optimum around 50-80%. Without specifying δ_p for each table entry, the experiments cannot be reproduced or compared across methods. Also, no error bars or confidence intervals are provided; given that the reported gains over the strongest baselines are about 1-2 dB at some sampling ratios, statistical significance over test splits should be established.
  4. [Sec. V.D, Related Work II.C] The comparison is limited to fixed Hadamard orderings and one static learned-pattern baseline. Existing adaptive/learned Hadamard selection methods (e.g., [12], [13], [37]) are discussed in Related Work but not compared experimentally. Since the paper's contribution is specifically scene-driven ordering, a quantitative comparison against at least the nearest adaptive baseline (e.g., magnitude-sorted selection from [37], or the deep-superpixel method [13]) would strengthen the claim that the proposed learned policy is superior to prior adaptive strategies.
minor comments (4)
  1. [Fig. 8 caption/text] The metric strings in the caption/text are difficult to parse; for example, 'Cake Cutting 42.82dB, 0.969, 0.0225' appears to exceed the reference PSNR in the first group. Please align the per-scene layout with the listed metrics so readers can verify the comparisons.
  2. [Sec. V.A] The phrase 'dropout probability parameter set at 90%' is ambiguous: does it mean keep probability 0.9 or drop probability 0.9? Please clarify.
  3. [Sec. IV.B, Eq. (8)] Although the training is called 'end-to-end', the loss in Eq. (11) is a binary cross-entropy against the oracle support, not a reconstruction loss. The term 'end-to-end' is used in multiple places; consider reserving it for the full sensing-reconstruction pipeline or explicitly defining the end-to-end scope.
  4. [General] The F1-score and accuracy of the mask prediction are reported as single numbers with no variability. Since these are part of the evidence for the learned policy, include means and standard deviations over test folds.

Circularity Check

0 steps flagged

No derivation reduces to its inputs; the scene-driven selector is trained on an oracle support from full data and tested on held-out splits and an out-of-distribution NIR testbed, so the central claim is not circular.

full rationale

The paper's derivation chain is supervised learning, not equation-level circularity. The reference mask m_s is defined in Eq. (10) as the top-k_a remaining Hadamard coefficients by aggregated magnitude, and Eq. (11) trains P_theta to match m_s via BCE. This is a standard oracle-labeling procedure: the network is fitted to full-data labels during training and then evaluated on held-out EuroSAT test images, on ARAD, and on a real NIR testbed without fine-tuning. No fitted parameter is renamed as a prediction, no equation equals another by construction, and the reconstruction metrics (PSNR/SSIM/SAM) are computed from the decoder outputs, not from the training loss. The admitted limitation that the oracle magnitude proxy is not validated against nonlinear solver optimality (F1≈0.68, Sec. V.B) is a correctness/optimality concern, not a circularity concern. Self-citations to the authors' prior work ([13], [27], [38]) are present but are not load-bearing: [38] is described only as a preliminary study, and the main comparisons are against external fixed orderings and external learned-pattern baselines. The real-testbed evaluation compares all orderings under identical photon/noise conditions via numerical permutation of a full Hadamard acquisition, which is a fair emulation rather than circular reasoning. Therefore the paper scores 0 on circularity.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central claim rests on the domain assumptions that spectral images are sparse in the Hadamard basis, that the top-magnitude coefficients are the most informative, and that low-frequency measurements predict them; plus the practical equivalence of full-then-subsample to true adaptive acquisition. The neural network and hyperparameters are free parameters fitted to training data. No new physical entities are introduced.

free parameters (4)
  • Scene-driven selector network parameters θ = ~5.5M parameters learned on EuroSAT/ARAD/CelebA per experiment
    The mapping P_θ from predetermined measurements to support scores is fitted to training data by minimizing BCE (Eq. 11).
  • Training hyperparameters = learning rate 1e-3, 500 epochs, dropout 0.9, Adam
    Chosen manually; affect learned selector and final results.
  • Relative predetermined ratio δ_p = k_p/k = Not explicitly fixed in main text; supplemental explores 1–95% and reports best around 50–80%
    Trade-off parameter tuned empirically on the evaluation datasets; main-table results likely use a favorable value.
  • Choice of predetermined ordering (Zig-Zag) = Zig-Zag [21]
    Selected as the standard predetermined path; other choices would change the input to the selector.
axioms (5)
  • standard math Hadamard matrix orthogonality H^T H = n I
    Used in Eq. (3) for transpose reconstruction.
  • domain assumption Spectral images are compressible in the Hadamard basis; most coefficients are near zero
    Motivates subsampling; stated in Sec. III.
  • domain assumption The top-k aggregated-magnitude Hadamard coefficients are the most informative for reconstruction
    Defines oracle support m_s in Sec. IV.B and Eq. (10).
  • domain assumption The predetermined low-frequency block provides enough information to predict the support of remaining significant coefficients
    Core to the two-stage design; empirically supported by ACC/F1 but not by a theoretical guarantee.
  • domain assumption Acquiring the full Hadamard set and numerically subsampling is equivalent to executing the two-stage adaptive protocol
    Used for real testbed experiments; stated in Supp. XI.A.

reviewed 2026-08-02 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Deep Scene-Driven Ordering of Hadamard Basis for Single-Pixel Spectral Imaging." pith.science (2026). https://pith.science/paper/QPJ2C2EJ

@misc{pith2026260715045,
  author       = {Pith},
  title        = {Pith review of: Deep Scene-Driven Ordering of Hadamard Basis for Single-Pixel Spectral Imaging},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QPJ2C2EJ}},
  note         = {Machine review of arXiv:2607.15045}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Spectral images are highly valuable for various applications, including environmental monitoring and precision agriculture. However, the high cost of specialized sensors limits the wide use of this technology in numerous applications. Current alternatives to acquire high spatial-spectral resolution spectral images, like Single-Pixel Imaging (SPI) enhanced with Deep Optical Coding Design (DOCD), have limitations due to their non-feedback optical designs, leading to limited image quality, with optimal performance achieved only for the specific scenes used during training. This work reformulates the DOCD framework to handle the scene-driven ordering of the Hadamard basis within the SPI architecture for spectral imaging. Taking into account that SPI usually acquires hundreds of snapshots, our approach introduces a scene-driven ordering of the Hadamard matrix for flexible SPI modulation pattern selection based on scene characteristics in an end-to-end optimization. Simulations on spectral datasets and real test-bed acquisitions demonstrate the effectiveness of the proposed method in improving the quality of VIS and NIR spectral images compared to fixed designs.

Figures

Figures reproduced from arXiv: 2607.15045 by Brayan Monroy, Hans Garcia, Henry Arguello, Jorge Bacca.

Figure 1
Figure 1. Figure 1: Generation of DMD Hadamard patterns. Specifically, we select The zig-zag ordering, constructed by permuting the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Hadamard Ordering Strategies. (a) The first row illus [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Hadamard Single Pixel Imaging. The starting spectral [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: scene-driven Hadamard Single Pixel Imaging. (a) Construction of reference sensing mask [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Image reconstruction from our proposed sensing methodology. a) presents reference images and reconstructed images [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparison between the proposed scene-driven ordering and four fixed Hadamard ordering strategies: [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Optical setup for Single Pixel Camera (SPC). Fea￾turing an IT 3900e infrared (IR) lamp as the light source, an IR-optimized Digital Micromirror Device (DMD) for spatial coding, and a NIRQuest spectrometer for measurement acqui￾sition. Additional IR-optimized components such as lenses and mirrors are integrated to enhance performance. in {−1, 1}, whereas the DMD accepts only binary values {0, 1}. Using the … view at source ↗
Figure 8
Figure 8. Figure 8: Real reconstruction results of proposed method and fixed ordering methodologies on NIR spectral images. The first [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Performance of different training cost functions for different split sensing configurations, relative predetermined ratio [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Relative performance on PSNR, SSIM, and SAM scores under different total sampling ratios [PITH_FULL_IMAGE:figures/full_fig_p012_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Visual reconstruction comparison on CelebA test samples under a [PITH_FULL_IMAGE:figures/full_fig_p014_11.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

49 extracted references · 2 linked inside Pith

  1. [1]

    Spectral imaging: principles and applications,

    Y . Garini, I. T. Young, and G. McNamara, “Spectral imaging: principles and applications,”Cytometry Part A: The Journal of the International Society for Analytical Cytology, vol. 69, no. 8, pp. 735–747, 2006

    2006

  2. [2]

    Computational spectral imag- ing: a contemporary overview,

    J. Bacca, E. Martinez, and H. Arguello, “Computational spectral imag- ing: a contemporary overview,”JOSA A, vol. 40, no. 4, pp. C115–C125, 2023

    2023

  3. [3]

    Spectral imaging for remote sensing,

    G. A. Shaw and H. K. Burke, “Spectral imaging for remote sensing,” Lincoln laboratory journal, vol. 14, no. 1, pp. 3–28, 2003. 10

    2003

  4. [4]

    Thermal infrared imaging,

    M. Teena and A. Manickavasagan, “Thermal infrared imaging,” in Imaging with Electromagnetic Spectrum: Applications in Food and Agriculture. Springer, 2014, pp. 147–173

    2014

  5. [5]

    A review: Remote sensing sensors,

    L. Zhu, J. Suomalainen, J. Liu, J. Hyypp ¨a, H. Kaartinen, H. Haggren et al., “A review: Remote sensing sensors,”Multi-purposeful application of geospatial data, pp. 19–42, 2018

    2018

  6. [6]

    Scaling infrared detectors—status and outlook,

    A. Rogalski, “Scaling infrared detectors—status and outlook,”Reports on Progress in Physics, vol. 85, no. 12, p. 126501, 2022

    2022

  7. [7]

    Challenges of small-pixel infrared detectors: a review,

    A. Rogalski, P. Martyniuk, and M. Kopytko, “Challenges of small-pixel infrared detectors: a review,”Reports on Progress in Physics, vol. 79, no. 4, p. 046501, 2016

    2016

  8. [8]

    Single-pixel imaging via compressive sampling,

    M. F. Duarte, M. A. Davenport, D. Takhar, J. N. Laska, T. Sun, K. F. Kelly, and R. G. Baraniuk, “Single-pixel imaging via compressive sampling,”IEEE signal processing magazine, vol. 25, no. 2, pp. 83–91, 2008

    2008

  9. [9]

    Digital micromirror device and its application to projection displays,

    J. B. Sampsell, “Digital micromirror device and its application to projection displays,”Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures Processing, Measurement, and Phenomena, vol. 12, no. 6, pp. 3242–3246, 1994

    1994

  10. [10]

    Infrared imaging and spectroscopy beyond the diffraction limit,

    A. Centrone, “Infrared imaging and spectroscopy beyond the diffraction limit,”Annual review of analytical chemistry, vol. 8, pp. 101–126, 2015

    2015

  11. [11]

    Single-pixel imaging 12 years on: a review,

    G. M. Gibson, S. D. Johnson, and M. J. Padgett, “Single-pixel imaging 12 years on: a review,”Optics express, vol. 28, no. 19, pp. 28 190– 28 208, 2020

    2020

  12. [12]

    Optimized sensing matrix for single pixel multi-resolution compressive spectral imaging,

    H. Garcia, C. V . Correa, and H. Arguello, “Optimized sensing matrix for single pixel multi-resolution compressive spectral imaging,”IEEE Transactions on Image Processing, vol. 29, pp. 4243–4253, 2020

    2020

  13. [13]

    Deep adaptive superpixels for hadamard single pixel imaging in near-infrared spectrum,

    B. Monroy, J. Bacca, and H. Arguello, “Deep adaptive superpixels for hadamard single pixel imaging in near-infrared spectrum,” inICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2023, pp. 1–5

    2023

  14. [14]

    Coupled deep learning coded aperture design for compressive image classification,

    J. Bacca, L. Galvis, and H. Arguello, “Coupled deep learning coded aperture design for compressive image classification,”Optics express, vol. 28, no. 6, pp. 8528–8540, 2020

    2020

  15. [15]

    Image quality of compressive single-pixel imaging using different hadamard orderings,

    P. G. Vaz, D. Amaral, L. R. Ferreira, M. Morgado, and J. Cardoso, “Image quality of compressive single-pixel imaging using different hadamard orderings,”Optics express, vol. 28, no. 8, pp. 11 666–11 681, 2020

    2020

  16. [16]

    Hadamard single-pixel imaging versus fourier single-pixel imaging,

    Z. Zhang, X. Wang, G. Zheng, and J. Zhong, “Hadamard single-pixel imaging versus fourier single-pixel imaging,”Optics Express, vol. 25, no. 16, pp. 19 619–19 639, 2017

    2017

  17. [17]

    Hadamard transforms

    S. S. Agaian, H. Sarukhanyan, K. Egiazarian, and J. Astola, “Hadamard transforms.” SPIE, 2011

    2011

  18. [18]

    Super sub-nyquist single-pixel imaging by means of cake- cutting hadamard basis sort,

    W.-K. Yu, “Super sub-nyquist single-pixel imaging by means of cake- cutting hadamard basis sort,”Sensors, vol. 19, no. 19, p. 4122, 2019

    2019

  19. [19]

    A russian dolls ordering of the hadamard basis for compressive single-pixel imaging,

    M.-J. Sun, L.-T. Meng, M. P. Edgar, M. J. Padgett, and N. Radwell, “A russian dolls ordering of the hadamard basis for compressive single-pixel imaging,”Scientific reports, vol. 7, no. 1, p. 3464, 2017

    2017

  20. [20]

    Super sub-nyquist single-pixel imaging by total variation ascending ordering of the hadamard basis,

    X. Yu, R. I. Stantchev, F. Yang, and E. Pickwell-MacPherson, “Super sub-nyquist single-pixel imaging by total variation ascending ordering of the hadamard basis,”Scientific Reports, vol. 10, no. 1, p. 9338, 2020

    2020

  21. [21]

    Efficient ordering of the hadamard basis for single pixel imaging,

    L. L ´opez-Garc´ıa, W. Cruz-Santos, A. Garc ´ıa-Arellano, P. Filio-Aguilar, J. A. Cisneros-Mart ´ınez, and R. Ramos-Garc ´ıa, “Efficient ordering of the hadamard basis for single pixel imaging,”Optics Express, vol. 30, no. 8, pp. 13 714–13 732, 2022

    2022

  22. [22]

    A detail- enhanced sampling strategy in hadamard single-pixel imaging,

    Y . Cai, S. Li, W. Zhang, H. Wu, X.-r. Yao, and Q. Zhao, “A detail- enhanced sampling strategy in hadamard single-pixel imaging,”arXiv preprint arXiv:2209.04449, 2022

    Pith/arXiv arXiv 2022

  23. [23]

    Spectral imaging with deep learning,

    L. Huang, R. Luo, X. Liu, and X. Hao, “Spectral imaging with deep learning,”Light: Science & Applications, vol. 11, no. 1, p. 61, 2022

    2022

  24. [24]

    High-quality hyperspectral reconstruction using a spectral prior,

    I. Choi, D. S. Jeon, G. Nam, D. Gutierrez, and M. H. Kim, “High-quality hyperspectral reconstruction using a spectral prior,” vol. 36, no. 6, nov

  25. [25]

    Deep plug-and-play priors for spectral snapshot compressive imaging,

    S. Zheng, Y . Liu, Z. Meng, M. Qiao, Z. Tong, X. Yang, S. Han, and X. Yuan, “Deep plug-and-play priors for spectral snapshot compressive imaging,”Photonics Research, vol. 9, no. 2, pp. B18–B29, 2021

    2021

  26. [26]

    Hyperspectral im- age reconstruction using a deep spatial-spectral prior,

    L. Wang, C. Sun, Y . Fu, M. H. Kim, and H. Huang, “Hyperspectral im- age reconstruction using a deep spatial-spectral prior,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 8032–8041

    2019

  27. [27]

    Deep optical coding design in computational imaging: a data- driven framework,

    H. Arguello, J. Bacca, H. Kariyawasam, E. Vargas, M. Marquez, R. Hettiarachchi, H. Garcia, K. Herath, U. Haputhanthri, B. S. Ahluwalia et al., “Deep optical coding design in computational imaging: a data- driven framework,”IEEE Signal Processing Magazine, vol. 40, no. 2, pp. 75–88, 2023

    2023

  28. [28]

    Deep coded aperture design: An end-to-end approach for computational imaging tasks,

    J. Bacca, T. Gelvez-Barrera, and H. Arguello, “Deep coded aperture design: An end-to-end approach for computational imaging tasks,”IEEE Transactions on Computational Imaging, vol. 7, pp. 1148–1160, 2021

    2021

  29. [29]

    Privhar: Recognizing human actions from privacy-preserving lens,

    C. Hinojosa, M. Marquez, H. Arguello, E. Adeli, L. Fei-Fei, and J. C. Niebles, “Privhar: Recognizing human actions from privacy-preserving lens,” inEuropean Conference on Computer Vision. Springer, 2022, pp. 314–332

    2022

  30. [30]

    Optics lens design for privacy-preserving scene captioning,

    P. Arguello, J. Lopez, C. Hinojosa, and H. Arguello, “Optics lens design for privacy-preserving scene captioning,” in2022 IEEE International Conference on Image Processing (ICIP). IEEE, 2022, pp. 3551–3555

    2022

  31. [31]

    A detail-enhanced sampling strategy in hadamard single-pixel imaging,

    Y . Cai, S. Li, W. Zhang, H. Wu, X. Yao, and Q. Zhao, “A detail-enhanced sampling strategy in hadamard single-pixel imaging,”Chinese Optics Letters, vol. 21, no. 7, p. 071101, 2023

    2023

  32. [32]

    Shift-variant color-coded diffractive spectral imaging system,

    H. Arguello, S. Pinilla, Y . Peng, H. Ikoma, J. Bacca, and G. Wetzstein, “Shift-variant color-coded diffractive spectral imaging system,”Optica, vol. 8, no. 11, pp. 1424–1434, 2021

    2021

  33. [33]

    Learning sensor multiplexing design through back- propagation,

    A. Chakrabarti, “Learning sensor multiplexing design through back- propagation,”Advances in Neural Information Processing Systems, vol. 29, 2016

    2016

  34. [34]

    Adaptive filter design via a gradient thresholding algorithm for compressive spectral imaging,

    N. Diaz, H. Rueda, and H. Arguello, “Adaptive filter design via a gradient thresholding algorithm for compressive spectral imaging,” Applied Optics, vol. 57, no. 17, pp. 4890–4900, 2018

    2018

  35. [35]

    Adaptive compressed image sensing based on wavelet-trees,

    S. Dekel, “Adaptive compressed image sensing based on wavelet-trees,” preprint, 2008

    2008

  36. [36]

    Adaptive compressed image sensing using dictionaries,

    A. Averbuch, S. Dekel, and S. Deutsch, “Adaptive compressed image sensing using dictionaries,”SIAM Journal on Imaging Sciences, vol. 5, no. 1, pp. 57–89, 2012

    2012

  37. [37]

    Single-pixel infrared and visible microscope,

    N. Radwell, K. J. Mitchell, G. M. Gibson, M. P. Edgar, R. Bowman, and M. J. Padgett, “Single-pixel infrared and visible microscope,”Optica, vol. 1, no. 5, pp. 285–289, 2014

    2014

  38. [38]

    Predicting the spectrum: Deep adaptive sensing for hadamard single pixel spectral imaging,

    B. Monroy, J. Bacca, and H. Arguello, “Predicting the spectrum: Deep adaptive sensing for hadamard single pixel spectral imaging,” in 2024 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW). IEEE, 2024, pp. 154–158

    2024

  39. [39]

    Mallat,A wavelet tour of signal processing

    S. Mallat,A wavelet tour of signal processing. Elsevier, 1999

    1999

  40. [40]

    Plug-and-play admm for image restoration: Fixed-point convergence and applications,

    S. H. Chan, X. Wang, and O. A. Elgendy, “Plug-and-play admm for image restoration: Fixed-point convergence and applications,”IEEE Transactions on Computational Imaging, vol. 3, no. 1, pp. 84–98, 2016

    2016

  41. [41]

    Hyperspectral image denoising using spatio-spectral total variation,

    H. K. Aggarwal and A. Majumdar, “Hyperspectral image denoising using spatio-spectral total variation,”IEEE Geoscience and Remote Sensing Letters, vol. 13, no. 3, pp. 442–446, 2016

    2016

  42. [42]

    Plug- and-play unplugged: Optimization-free reconstruction using consensus equilibrium,

    G. T. Buzzard, S. H. Chan, S. Sreehari, and C. A. Bouman, “Plug- and-play unplugged: Optimization-free reconstruction using consensus equilibrium,”SIAM Journal on Imaging Sciences, vol. 11, no. 3, pp. 2001–2020, 2018

    2001

  43. [43]

    Plug-and-play priors for model based reconstruction,

    S. V . Venkatakrishnan, C. A. Bouman, and B. Wohlberg, “Plug-and-play priors for model based reconstruction,” in2013 IEEE global conference on signal and information processing. IEEE, 2013, pp. 945–948

    2013

  44. [44]

    Image denoising: The deep learning revolution and beyond—a survey paper,

    M. Elad, B. Kawar, and G. Vaksman, “Image denoising: The deep learning revolution and beyond—a survey paper,”SIAM Journal on Imaging Sciences, vol. 16, no. 3, pp. 1594–1654, 2023

    2023

  45. [45]

    Deepinverse: A python package for solving imaging inverse problems with deep learning,

    J. Tachella, M. Terris, S. Hurault, A. Wang, D. Chen, M.-H. Nguyen, M. Song, T. Davies, L. Davy, J. Donget al., “Deepinverse: A python package for solving imaging inverse problems with deep learning,”arXiv preprint arXiv:2505.20160, 2025

    Pith/arXiv arXiv 2025

  46. [46]

    Single-pixel imaging using physics enhanced deep learning,

    F. Wang, C. Wang, C. Deng, S. Han, and G. Situ, “Single-pixel imaging using physics enhanced deep learning,”Photonics Research, vol. 10, no. 1, pp. 104–110, 2021

    2021

  47. [47]

    Deep learning for real-time single-pixel video,

    C. F. Higham, R. Murray-Smith, M. J. Padgett, and M. P. Edgar, “Deep learning for real-time single-pixel video,”Scientific reports, vol. 8, no. 1, p. 2369, 2018

    2018

  48. [48]

    Single pixel imaging at megahertz switching rates via cyclic hadamard masks,

    E. Hahamovich, S. Monin, Y . Hazan, and A. Rosenthal, “Single pixel imaging at megahertz switching rates via cyclic hadamard masks,” Nature communications, vol. 12, no. 1, p. 4516, 2021. 11 SUPPLEMENTALMATERIALS IX. ADDITIONALSIMULATIONS. A. Evaluation of loss functions In this section, we comprehensively evaluate the proposed method under different train...

    arXiv 2021

  49. [2017]

    Available: https://doi.org/10.1145/3130800.3130810

    [Online]. Available: https://doi.org/10.1145/3130800.3130810

    arXiv

This paper was first reviewed by deepseek-v4-flash on August 2, 2026.