Pith. sign in

REVIEW 4 major objections 5 minor 59 references

FRN: Fractal-Based Recursive Spectral Reconstruction Network

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 FRN reconstructs 31-band hyperspectral images from RGB by recursively predicting two new spectral bands at a time from neighboring bands plus the RGB image, claiming state-of-the-art accuracy with only 0.30M parameters.

desk verdict Interesting recursive architecture with a plausible SOTA claim, but the baseline protocol is unverified and the theory overreaches; send to review with a requirement for code and matched splits. read the letter →

arxiv 2505.15439 v1 pith:KIL5MNPY submitted 2025-05-21 cs.CV eess.IV

Ge Meng , Zhongnan Cai , Ruizhe Chen , Jingyan Tu , Yingying Wang , Yue Huang , Xinghao Ding This is my paper
classification cs.CVeess.IV
keywords spectralreconstructionhyperspectralimagingrecursivegenerationfractalnetworksstatespacemodelband-awaremasklow-rankpriorRGB-to-HSI
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 claims that spectral reconstruction from RGB need not be done in a single one-shot pass over all spectral bands. Instead, a small atomic module can be invoked recursively, each time predicting two new spectral bands from the bands already generated plus the RGB image, following the low-rank structure of hyperspectral data. On the CAVE and Harvard datasets this recursive 'fractal' scheme reaches higher PSNR, SSIM, and UIQI than one-shot transformer and CNN baselines while using fewer than 0.31 million parameters. If true, this shows a progressive band-by-band generation strategy is a parameter-efficient alternative to direct full-spectrum reconstruction.

What carries the argument

The central objects are the fractal generator and the band-aware SSM block. The fractal generator gi is an atomic module that maps level input xi to xi+1 producing n output channels, so that K = n^m channels are generated over m levels; this realizes the divide-and-conquer factorization of p(y1,...,yK) into conditional distributions, where each generator models one conditional p(next subset | previous subsets). The atomic module is BAMamba, a VMamba-style state space model whose hidden-state update uses a mask M = 1 if ΔA ≥ ε else 0, suppressing low-correlation pixels before cross-scanning. The recursion is what carries the argument: it replaces one ill-posed RGB-to-HSI map with m easier conditional maps, each conditioned on neighboring bands and the RGB image.

What would settle it

Measure whether the chain-truncation holds: train FRN with bands fed in a random order, or with a band skipped between the conditioning set and the target band. If reconstruction quality collapses when the conditioning bands are not the immediate spectral neighbors, that shows short-context conditioning rather than general low-rank structure is carrying the result. A second check: replace the ΔA-threshold mask with a mask built directly from spatial correlation between predicted and ground-truth bands; if performance is essentially unchanged, the mask mechanism's explanation is wrong.

Watch

Extended reading notes

Core claim

FRN reconstructs a 31-band HSI by factorizing the joint distribution over wavelengths into a chain of conditional generations, with each recursive level generating two new bands conditioned on previously generated bands and the RGB input. The atomic generator is a U-Net style state space model (BAMamba) with linear complexity; it thresholds the discretized SSM coefficient ΔA to build a band-aware mask that suppresses pixels with low spatial correlation before scanning. With five recursive levels and two channels per level, FRN reports CAVE PSNR 41.05 dB, SSIM 0.9900, and Harvard PSNR 42.88 dB, exceeding the strongest baselines (MSFN: 39.84/42.65 dB) with fewer parameters (0.30M vs 2.48M). The ablation tables attribute the gains to the recursive levels and to the band-aware mask at the right threshold.

Load-bearing premise

The argument assumes that each new spectral band can be generated accurately from only the immediately available previous bands plus the RGB image, without needing information from distant bands; and it assumes that thresholding the state space coefficient ΔA produces a mask of spatially low-correlation pixels.

Editorial extensions

If this is right

  • Since K = n^m, a 31-band HSI needs only five recursive levels; increasing the number of levels monotonically improved CAVE PSNR in the ablation (from 39.86 dB one-step to 41.05 dB at M=5), suggesting that spectra with more bands could be handled by adding levels rather than enlarging the network.
  • The band-aware mask threshold must be tuned carefully: α=0.5 gives 41.05 dB, while α=0.8 degrades to 37.48 dB, so the mask is not a free lunch and requires per-dataset adjustment.
  • Excluding the RGB image from each recursive level drops CAVE PSNR from 41.05 to 38.62 dB, indicating that the RGB image remains a necessary conditioning signal at every level, not just at the first.
  • With 0.30M parameters and linear-complexity SSM scanning, the recursive design is compatible with deployment on limited hardware, provided the multiple recursive invocations are acceptable in the runtime budget.

Reading between the lines

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

  • The same recursive generator could be applied to other ill-posed inverse problems with a natural sequential output structure, such as multi-spectral to hyper-spectral fusion or video frame interpolation, where each step adds a small number of output channels.
  • A direct measurement of inter-band correlation matrices on CAVE and Harvard could test whether the Markov-like truncation is empirically justified; if distant bands correlate strongly but the model only conditions on immediate neighbors, the reported gains may be dataset-specific rather than universal.
  • The band-aware mask derived from ΔA could be compared against masks built directly from spatial correlation or attention maps; if performance is unchanged, the authors' explanation of what the mask selects would need revision.
  • The recursive approach might also be combined with a learned stopping rule that skips levels when spectral variation is small, further reducing computation for smooth spectra.
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 FRN, a recursive spectral reconstruction network that reconstructs a 31-band hyperspectral image from an RGB input by repeatedly invoking a shared atomic generator, moving from broad to narrow spectral bands. The atomic generator is BAMamba, a VMamba-style state space model augmented with a band-aware mask that suppresses pixels whose SSM transition coefficients fall below a threshold. On the CAVE and Harvard datasets the method reports PSNR values of 41.05 dB and 42.88 dB with 0.30M parameters, outperforming seven baselines, and ablations are provided for the mask threshold, the number of recursive levels, and the number of reference spectral channels. The paper frames the recursive decomposition as a divide-and-conquer factorization of the joint distribution of spectral bands and attributes the gains to the low-rank structure of HSIs and the suppression of low-correlation regions.

Significance. If the reported results hold under a fair comparison protocol, the central claim is valuable: FRN demonstrates that a parameter-efficient recursive paradigm, rather than a one-shot full-spectrum network, can achieve state-of-the-art RGB-to-HSI reconstruction. The paper explicitly reports a very small parameter count (0.30M), provides quantitative and qualitative comparisons on two public datasets, and includes ablations of the main design choices. The connection to fractal generative models is a useful framing, and the BAMamba module addresses a real computational concern in recursive architectures. However, the significance is currently conditional on resolving two load-bearing issues: whether the baseline numbers were obtained under identical data splits, and whether the theoretical justifications in Eqs. (4) and (8) actually describe what the implemented network does.

major comments (4)
  1. [§4.1 and Table 1] The paper reports its own train/validation/test splits (20/6/6 on CAVE, 30/10/10 on Harvard) but never states whether the seven baseline numbers in Table 1 were obtained by retraining under these same splits or quoted from their original papers. Published spectral reconstruction papers commonly use different splits, such as 20 train / 12 test on CAVE and 30 train / 20 test on Harvard, so if the baseline numbers come from larger test sets, the claimed improvements of 1.21 dB (CAVE) and 0.23 dB (Harvard) could be artifacts of an easier test protocol rather than of the FRN design. The authors must specify the exact protocol for every baseline, provide per-image results, or release code so that the comparison can be reproduced under a single protocol.
  2. [§3.2, Eq. (4)] Equation (4) is an exact chain-rule factorization in which each conditional distribution depends on all earlier generated variables, but the implemented architecture conditions each atomic generator only on the previous level's output plus the RGB image, i.e., a short-context Markov-like truncation. No spectral-correlation measurement, conditional-independence test, or theorem is provided to show that this truncation is sufficient for HSI bands. As a result, Eq. (4) does not actually justify the architecture; it provides only an analogy. The authors should either supply empirical evidence that neighboring-band conditioning captures the relevant dependence, or explicitly reframe Eq. (4) as motivation rather than derivation.
  3. [§3.3, Eqs. (8)-(9)] The band-aware mask is generated by thresholding the SSM coefficient ΔA, but ΔA is the discretized state-transition matrix from Eq. (6), a learned parameter that is not shown to encode per-pixel spatial correlation. The paper provides no derivation connecting the magnitude of ΔA to the reflectance-based spatial correlation discussed around Fig. 4. Furthermore, Eq. (9) changes the SSM output from Eq. (7) by replacing C h_t with (C⊙M) h_t and dropping the D x_t term, and this modification is not derived from the SSM formulation. The authors should provide a derivation or a focused experiment showing that thresholding ΔA actually isolates low-correlation regions rather than some other artifact.
  4. [§4.5, Tables 2-4] The ablation tables select the hyperparameters α, M, and S using the CAVE dataset, but the text does not state whether these selections were made on the validation split or on the test split. If the latter, the reported test-set numbers are optimistically biased and the comparison to baselines becomes unfair. The paper should explicitly state that all hyperparameters were chosen on the validation split, or, if not, should re-run the final evaluation on held-out test data after fixing hyperparameters.
minor comments (5)
  1. [§3.2] The paper says K = n^m and uses m=5, n=2, giving 32 channels, but the datasets contain 31 spectral bands; the treatment of the extra channel is not explained.
  2. [§3.3] Equation (10) writes CS(LN(feat))⊙SiLU(LN(feat)) + feat, but the text says CS is the sequence DWConv→SiLU→SSM→LN; it is unclear whether the first LN is inside or outside the CS operation.
  3. [Figure 4] The residual maps in Figure 4 are not described: the reader is not told how the residuals are computed or which RGB channel each panel corresponds to.
  4. [General] There are several typos and minor wording issues, including 'band-awared' in the caption of Figure 3 and 'an band-aware mask' in Section 3.3; a careful proofread is needed.
  5. [General] No code availability statement or link is provided; releasing code would substantially strengthen the reproducibility of the headline results.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: FRN's central claim is validated against external baselines on public datasets, and its architectural derivations are exact identities or explicit heuristics rather than reductions to inputs.

full rationale

The paper's load-bearing derivation is the chain-rule factorization in Eq. (4), which is an exact probability identity and therefore cannot be circular; the subsequent claim that only neighboring bands are needed is presented as an explicit design assumption, not as a consequence of Eq. (4). The recursive generator is defined as x_{i+1}=g_i(x_i), but the evaluation measures reconstruction against ground-truth HSIs rather than against the generator's own input, so no quantity is being predicted from itself. The band-aware mask in Eq. (8) is a thresholded SSM parameter interpreted as a correlation filter; this is an empirical heuristic rather than a formal reduction of the output to the input. The paper contains no load-bearing self-citations: the fractal-generator reference [31] is external, and the authors' prior works [36,37] are not used to justify the central claim. Hyperparameters such as α and M are ablated on the CAVE dataset, but they are not fitted values that are then renamed as predictions. The potential concern that Table 1 baselines may come from different train/test splits is a fairness or correctness issue, not a circularity of the kind defined here. Overall, the central SOTA claim is checked against external baselines on public datasets, making the contribution empirically self-contained rather than definitionally circular.

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

The central method rests on a small number of architectural choices (n=2, M=5, alpha=0.5, S=4) tuned on the evaluation datasets, and on three unproven domain assumptions about spectral self-similarity, context truncation, and the masking of Delta A. No new physical entities are introduced.

free parameters (4)
  • Suppression threshold alpha = 0.5
    Selected by the CAVE ablation sweep (Table 2); it controls the threshold epsilon in the band-aware mask (Eq. 8).
  • Recursive levels M = 5
    Selected by CAVE ablation (Table 3); with branching factor 2, five levels yield 32 bands, matching the 31-band datasets.
  • Reference channels S = 4
    Selected by CAVE ablation (Table 4); the number of neighboring wavelengths plus RGB input to each atomic generator.
  • Branching factor n = 2
    Each atomic generator outputs two channels per invocation; with M=5 this determines the 32-band output. This is a design choice and is not independently ablated.
assumptions (4)
  • domain assumption Spectral bands are locally self-similar and low-rank, so a new band can be predicted from neighboring bands.
    Invoked in Section 3.2 to motivate recursive generation; no spectral-correlation analysis or low-rank test is presented.
  • ad hoc to paper The exact chain-rule factorization in Eq. (4) remains a faithful model when each factor conditions only on the previous level's output plus the RGB image.
    The architecture feeds only parent bands and RGB into each atomic generator, while Eq. (4) factors over all previously generated variables; the equivalence is assumed without proof.
  • ad hoc to paper Thresholding the SSM coefficient Delta A produces a mask that suppresses pixels with low cross-band spatial correlation.
    Eqs. (8)-(9) in Section 3.3 assert this connection; no derivation or measurement links Delta A values to spatial correlation.
  • standard math The standard Mamba discretization (zero-order hold) in Section 3.3 is valid for the band-generation setting.
    Eqs. (5)-(7) are standard SSM discretization formulas, assumed without further proof.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FRN: Fractal-Based Recursive Spectral Reconstruction Network." pith.science (2026). https://pith.science/paper/KIL5MNPY

@misc{pith2026250515439,
  author       = {Pith},
  title        = {Pith review of: FRN: Fractal-Based Recursive Spectral Reconstruction Network},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KIL5MNPY}},
  note         = {Machine review of arXiv:2505.15439}
}
read the original abstract

Generating hyperspectral images (HSIs) from RGB images through spectral reconstruction can significantly reduce the cost of HSI acquisition. In this paper, we propose a Fractal-Based Recursive Spectral Reconstruction Network (FRN), which differs from existing paradigms that attempt to directly integrate the full-spectrum information from the R, G, and B channels in a one-shot manner. Instead, it treats spectral reconstruction as a progressive process, predicting from broad to narrow bands or employing a coarse-to-fine approach for predicting the next wavelength. Inspired by fractals in mathematics, FRN establishes a novel spectral reconstruction paradigm by recursively invoking an atomic reconstruction module. In each invocation, only the spectral information from neighboring bands is used to provide clues for the generation of the image at the next wavelength, which follows the low-rank property of spectral data. Moreover, we design a band-aware state space model that employs a pixel-differentiated scanning strategy at different stages of the generation process, further suppressing interference from low-correlation regions caused by reflectance differences. Through extensive experimentation across different datasets, FRN achieves superior reconstruction performance compared to state-of-the-art methods in both quantitative and qualitative evaluations.

Figures

Figures reproduced from arXiv: 2505.15439 by the authors.

Figure 2
Figure 2. Overview of the Fractal-Based Progressive Spectral Reconstruction Paradigm: (a) illustrates [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 1
Figure 1. PSNR-Parameters comparisons of FRN and SOTA methods. Fractals are common patterns observed in neural net￾works [31]. Numerous studies have demonstrated the effectiveness of fractal or scale-invariant small-world network structures in the brain and its functional net￾works [4, 48, 6]. This inspires the idea that a larger network can be recursively constructed from smaller atomic modules. In addition to neural network… view at source ↗
Figure 3
Figure 3. The details of BAMamba. BAMamba is a U-Net style network built with state space [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Residual maps across the R, G, and B channels from a CAVE dataset sample. Different objects may exhibit substantial differ￾ences in emissivity at the same wavelength, which implies that the energy intensity in HSI can un￾dergo significant spatial variations. Furthermor…
Figure 5
Figure 5. Figure 5: Comparison of the reconstruction results of different methods on one scene from the CAVE [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Comparison of the reconstruction results of different methods on one scene from the [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Comparison of the spectral curves among different methods on CAVE (left) and Harvard [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

59 extracted references · 51 canonical work pages

  1. [31]

    T. Li, Q. Sun, L. Fan, and K. He. Fractal generative models.arXiv preprint arXiv:2502.17437, 2025

    arXiv 2025

  2. [1]

    Akhtar and A

    N. Akhtar and A. Mian. Hyperspectral recovery from rgb images using gaussian processes.IEEE transactions on pattern analysis and machine intelligence, 42(1):100–113, 2018

    work page 2018

  3. [2]

    Arad and O

    B. Arad and O. 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, pages 19–34. Springer, 2016

    work page 2016

  4. [3]

    Backman, M

    V . Backman, M. B. Wallace, L. Perelman, J. Arendt, R. Gurjar, M. Müller, Q. Zhang, G. Zonios, E. Kline, T. McGillican, et al. Detection of preinvasive cancer cells.nature, 406(6791):35–36, 2000

    work page 2000

  5. [4]

    D. S. Bassett, A. Meyer-Lindenberg, S. Achard, T. Duke, and E. Bullmore. Adaptive reconfiguration of fractal small-world human brain functional networks.Proceedings of the National Academy of Sciences, 103(51):19518–19523, 2006

    work page 2006

  6. [5]

    Breuer and J

    M. Breuer and J. Albertz. Geometric correction of airborne whiskbroom scanner imagery using hybrid auxiliary data.International Archives of Photogrammetry and Remote Sensing, 33(B3/1; PART 3):93–100, 2000

    work page 2000

  7. [6]

    Bullmore and O

    E. Bullmore and O. Sporns. Complex brain networks: graph theoretical analysis of structural and functional systems.Nature reviews neuroscience, 10(3):186–198, 2009

    work page 2009

  8. [7]

    Y . Cai, X. Hu, H. Wang, Y . Zhang, H. Pfister, and D. Wei. Learning to generate realistic noisy images via pixel-level noise-aware adversarial training.Advances in Neural Information Processing Systems, 34:3259–3270, 2021

    work page 2021

Show all 59 references
  1. [8]

    Y . Cai, J. Lin, X. Hu, H. Wang, X. Yuan, Y . Zhang, R. Timofte, and L. Van Gool. Coarse-to-fine sparse transformer for hyperspectral image reconstruction. InEuropean conference on computer vision, pages 686–704. Springer, 2022

    2022

  2. [9]

    Y . Cai, J. Lin, X. Hu, H. Wang, X. Yuan, Y . Zhang, R. Timofte, and L. Van Gool. Mask-guided spectral-wise transformer for efficient hyperspectral image reconstruction. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 17502–17511, 2022

    2022

  3. [10]

    Y . Cai, J. Lin, Z. Lin, H. Wang, Y . Zhang, H. Pfister, R. Timofte, and L. Van Gool. Mst++: Multi-stage spectral-wise transformer for efficient spectral reconstruction. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 745–755, 2022

    2022

  4. [11]

    Y . Cai, J. Lin, H. Wang, X. Yuan, H. Ding, Y . Zhang, R. Timofte, and L. V . Gool. Degradation-aware unfolding half-shuffle transformer for spectral compressive imaging.Advances in Neural Information Processing Systems, 35:37749–37761, 2022

    2022

  5. [12]

    X. Cao, T. Yue, X. Lin, S. Lin, X. Yuan, Q. Dai, L. Carin, and D. J. Brady. Computational snapshot multispectral cameras: Toward dynamic capture of the spectral world.IEEE Signal Processing Magazine, 33(5):95–108, 2016

    2016

  6. [13]

    Y . Cao, C. Liu, Z. Wu, W. Yao, L. Xiong, J. Chen, and Z. Huang. Remote sensing image segmentation using vision mamba and multi-scale multi-frequency feature fusion.arXiv preprint arXiv:2410.05624, 2024

    2024 arXiv

  7. [14]

    Chakrabarti and T

    A. Chakrabarti and T. Zickler. Statistics of real-world hyperspectral images. InCVPR 2011, pages 193–200. IEEE, 2011

    2011

  8. [15]

    R. Dian, Y . Liu, and S. Li. Spectral super-resolution via deep low-rank tensor representation.IEEE Transactions on Neural Networks and Learning Systems, 2024

    2024

  9. [16]

    R. Dian, T. Shan, W. He, and H. Liu. Spectral super-resolution via model-guided cross-fusion network. IEEE Transactions on Neural Networks and Learning Systems, 2023

    2023

  10. [17]

    P. K. Diederik. Adam: A method for stochastic optimization.(No Title), 2014

    2014

  11. [18]

    H. Du, X. Tong, X. Cao, and S. Lin. A prism-based system for multispectral video acquisition. In2009 IEEE 12th International Conference on Computer Vision, pages 175–182. IEEE, 2009

    2009

  12. [19]

    Galliani, C

    S. Galliani, C. Lanaras, D. Marmanis, E. Baltsavias, and K. Schindler. Learned spectral super-resolution. arXiv preprint arXiv:1703.09470, 2017

    2017 arXiv

  13. [20]

    A. F. Goetz, G. Vane, J. E. Solomon, and B. N. Rock. Imaging spectrometry for earth remote sensing. science, 228(4704):1147–1153, 1985. 10

    1985

  14. [21]

    J. He, J. Li, Q. Yuan, H. Shen, and L. Zhang. Spectral response function-guided deep optimization-driven network for spectral super-resolution.IEEE Transactions on Neural Networks and Learning Systems, 33(9):4213–4227, 2021

    2021

  15. [22]

    W. He, N. Yokoya, and X. Yuan. Fast hyperspectral image recovery of dual-camera compressive hy- perspectral imaging via non-iterative subspace-based fusion.IEEE Transactions on Image Processing, 30:7170–7183, 2021

    2021

  16. [23]

    J. Hu, L. Shen, and G. Sun. Squeeze-and-excitation networks. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 7132–7141, 2018

    2018

  17. [24]

    X. Hu, Y . Cai, J. Lin, H. Wang, X. Yuan, Y . Zhang, R. Timofte, and L. Van Gool. Hdnet: High-resolution dual-domain learning for spectral compressive imaging. in 2022 ieee. InCVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 17521–17530, 2022

    2022

  18. [25]

    Huang, W

    T. Huang, W. Dong, X. Yuan, J. Wu, and G. Shi. Deep gaussian scale mixture prior for spectral compressive imaging. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 16216–16225, 2021

    2021

  19. [26]

    N. Keshava. Distance metrics and band selection in hyperspectral processing with applications to material identification and spectral libraries.IEEE Transactions on Geoscience and remote sensing, 42(7):1552– 1565, 2004

    2004

  20. [27]

    M. J. Khan, H. S. Khan, A. Yousaf, K. Khurshid, and A. Abbas. Modern trends in hyperspectral image analysis: A review.Ieee Access, 6:14118–14129, 2018

    2018

  21. [28]

    J. Li, Y . Leng, R. Song, W. Liu, Y . Li, and Q. Du. Mformer: Taming masked transformer for unsupervised spectral reconstruction.IEEE Transactions on Geoscience and Remote Sensing, 61:1–12, 2023

    2023

  22. [29]

    J. Li, C. Wu, R. Song, Y . Li, and F. Liu. Adaptive weighted attention network with camera spectral sensitivity prior for spectral reconstruction from rgb images. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, pages 462–463, 2020

    2020

  23. [30]

    J. Li, C. Wu, R. Song, Y . Li, W. Xie, L. He, and X. Gao. Deep hybrid 2-d–3-d cnn based on dual second- order attention with camera spectral sensitivity prior for spectral super-resolution.IEEE Transactions on Neural Networks and Learning Systems, 34(2):623–634, 2021

    2021

  24. [32]

    Y . Li, Y . Shi, K. Wang, B. Xi, J. Li, and P. Gamba. Target detection with unconstrained linear mixture model and hierarchical denoising autoencoder in hyperspectral imagery.IEEE Transactions on Image Processing, 31:1418–1432, 2022

    2022

  25. [33]

    Llull, X

    P. Llull, X. Liao, X. Yuan, J. Yang, D. Kittle, L. Carin, G. Sapiro, and D. J. Brady. Coded aperture compressive temporal imaging.Optics express, 21(9):10526–10545, 2013

    2013

  26. [34]

    Lu and B

    G. Lu and B. Fei. Medical hyperspectral imaging: a review.Journal of biomedical optics, 19(1):010901– 010901, 2014

    2014

  27. [35]

    Melgani and L

    F. Melgani and L. Bruzzone. Classification of hyperspectral remote sensing images with support vector machines.IEEE Transactions on geoscience and remote sensing, 42(8):1778–1790, 2004

    2004

  28. [36]

    Z. Meng, M. Qiao, J. Ma, Z. Yu, K. Xu, and X. Yuan. Snapshot multispectral endomicroscopy.Optics Letters, 45(14):3897–3900, 2020

    2020

  29. [37]

    Z. Meng, Z. Yu, K. Xu, and X. Yuan. Self-supervised neural networks for spectral snapshot compressive imaging. InProceedings of the IEEE/CVF international conference on computer vision, pages 2622–2631, 2021

    2021

  30. [38]

    R. M. Nguyen, D. K. Prasad, and M. S. Brown. Training-based spectral reconstruction from a single rgb image. InComputer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part VII 13, pages 186–201. Springer, 2014

    2014

  31. [39]

    Parmar, S

    M. Parmar, S. Lansel, and B. A. Wandell. Spatio-spectral reconstruction of the multispectral datacube using sparse recovery. In2008 15th IEEE International Conference on Image Processing, pages 473–476. IEEE, 2008

    2008

  32. [40]

    H. Peng, X. Chen, and J. Zhao. Residual pixel attention network for spectral reconstruction from rgb images. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, pages 486–487, 2020. 11

    2020

  33. [41]

    Poli and T

    D. Poli and T. Toutin. Review of developments in geometric modelling for high resolution satellite pushbroom sensors.The Photogrammetric Record, 27(137):58–73, 2012

    2012

  34. [42]

    J. Qiao, J. Liao, W. Li, Y . Zhang, Y . Guo, Y . Wen, Z. Qiu, J. Xie, J. Hu, and S. Lin. Hi-mamba: Hierarchical mamba for efficient image super-resolution.arXiv preprint arXiv:2410.10140, 2024

    2024 arXiv

  35. [43]

    K. Ren, X. Wu, L. Xu, and L. Wang. Remotedet-mamba: A hybrid mamba-cnn network for multi-modal object detection in remote sensing images.arXiv preprint arXiv:2410.13532, 2024

    2024

  36. [44]

    Y . Ren, X. Li, M. Guo, B. Li, S. Zhao, and Z. Chen. Mambacsr: Dual-interleaved scanning for compressed image super-resolution with ssms.arXiv preprint arXiv:2408.11758, 2024

    2024 arXiv

  37. [45]

    Robles-Kelly

    A. Robles-Kelly. Single image spectral reconstruction for multimedia applications. InProceedings of the 23rd ACM international conference on Multimedia, pages 251–260, 2015

    2015

  38. [46]

    Z. Shi, C. Chen, Z. Xiong, D. Liu, and F. Wu. Hscnn+: Advanced cnn-based hyperspectral recovery from rgb images. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, pages 939–947, 2018

    2018

  39. [47]

    Shimoni, R

    M. Shimoni, R. Haelterman, and C. Perneel. Hypersectral imaging for military and security applications: Combining myriad processing and sensing techniques.IEEE Geoscience and Remote Sensing Magazine, 7(2):101–117, 2019

    2019

  40. [48]

    O. Sporns. Small-world connectivity, motif composition, and complexity of fractal neuronal connections. Biosystems, 85(1):55–64, 2006

    2006

  41. [49]

    Stiebel, S

    T. Stiebel, S. Koppers, P. Seltsam, and D. Merhof. Reconstructing spectral images from rgb-images using a convolutional neural network. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, pages 948–953, 2018

    2018

  42. [50]

    Uzkent, A

    B. Uzkent, A. Rangnekar, and M. Hoffman. Aerial vehicle tracking by adaptive fusion of hyperspectral likelihood maps. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, pages 39–48, 2017

    2017

  43. [51]

    A. A. Wagadarikar, N. P. Pitsianis, X. Sun, and D. J. Brady. Video rate spectral imaging using a coded aperture snapshot spectral imager.Optics express, 17(8):6368–6388, 2009

    2009

  44. [52]

    Wang and A

    Z. Wang and A. C. Bovik. A universal image quality index.IEEE signal processing letters, 9(3):81–84, 2002

    2002

  45. [53]

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli. Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4):600–612, 2004

    2004

  46. [54]

    Y . Wu, R. Dian, and S. Li. Multistage spatial–spectral fusion network for spectral super-resolution.IEEE Transactions on Neural Networks and Learning Systems, 2024

    2024

  47. [55]

    Xiong, Z

    Z. Xiong, Z. Shi, H. Li, L. Wang, D. Liu, and F. Wu. Hscnn: Cnn-based hyperspectral image recovery from spectrally undersampled projections. InProceedings of the IEEE international conference on computer vision workshops, pages 518–525, 2017

    2017

  48. [56]

    Yasuma, T

    F. Yasuma, T. Mitsunaga, D. Iso, and S. K. Nayar. Generalized assorted pixel camera: postcapture control of resolution, dynamic range, and spectrum.IEEE transactions on image processing, 19(9):2241–2253, 2010

    2010

  49. [57]

    Y . Yuan, X. Zheng, and X. Lu. Hyperspectral image superresolution by transfer learning.IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 10(5):1963–1974, 2017

    1963

  50. [58]

    Zhang, Z

    L. Zhang, Z. Lang, P. Wang, W. Wei, S. Liao, L. Shao, and Y . Zhang. Pixel-aware deep function-mixture network for spectral super-resolution. InProceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 12821–12828, 2020

    2020

  51. [59]

    Q. Zhu, Y . Cai, Y . Fang, Y . Yang, C. Chen, L. Fan, and A. Nguyen. Samba: Semantic segmentation of remotely sensed images with state space model.Heliyon, 10(19), 2024. 12

    2024

Pith tools

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