Pith. sign in

REVIEW 4 major objections 5 minor 42 references

HSRMamba: Contextual Spatial-Spectral State Space Model for Single Image Hyperspectral Super-Resolution

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read A state-space model designed for hyperspectral images, HSRMamba, reports top scores over eight super-resolution baselines on the Chikusei and Houston datasets at ×4 and ×8 magnification.

desk verdict Solid Mamba-for-HSISR architecture paper, but the SOTA margin is the same size as the ablation effect, and with 4-image test sets and no error bars, the attribution to LSSP/GSRM is not pinned down. read the letter →

arxiv 2501.18500 v2 pith:IODRP22B submitted 2025-01-30 cs.CV eess.IV

Shi Chen , Lefei Zhang , Liangpei Zhang This is my paper · ORCID
classification cs.CVeess.IV
keywords hyperspectralimagesuper-resolutionstatespacemodelMambaspectralreorderinglocalspatial-spectralpartitioningbidirectionalscanningrestoration
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that the Mamba state-space architecture, normally built for 1D sequences, can be made effective for single hyperspectral image super-resolution if the input is ordered to respect both local and global spatial-spectral structure. It proposes HSRMamba, which scans small 3D windows of the hyperspectral cube bidirectionally and, in a separate path, reorders spectral bands by their global correlation before scanning. The authors report that the model outperforms eight existing super-resolution methods on the Chikusei and Houston datasets at both $\times 4$ and $\times 8$ magnification on all reported quality metrics. The intended upshot is that state-space models can carry long-range spatial-spectral modeling for hyperspectral restoration at linear cost, without the quadratic attention overhead of Transformers.

What carries the argument

The load-bearing machinery is a bidirectional state-space scanning unit (BSSM) fed by two ordering strategies. LSSP divides the 3D feature cube into local $h \times w \times c$ windows so that the scan sees spatially and spectrally adjacent pixels together, while GSRM computes a spectral correlation matrix and rearranges bands so that highly similar spectra become neighbors in the scan sequence. These orderings are what the paper credits with mitigating local pixel forgetting and reducing sensitivity to input order, respectively.

What would settle it

Train HSRMamba twice with the only difference being the global spectral reordering: once with the similarity-based ordering and once with a fixed random permutation of the same bands, keeping every other hyper-parameter identical. If the random-order model does not lose PSNR relative to the no-reordering baseline, the similarity ordering is not the active ingredient behind the reported gains.

Watch

Extended reading notes

Core claim

The central claim is that HSRMamba, a state-space model built from a local spatial-spectral partitioning mechanism (LSSP) and a global spectral reordering mechanism (GSRM), beats the current best single hyperspectral image super-resolution methods in both quantitative and visual quality. The reported numbers are best on all six metrics on Chikusei at $\times 4$ and $\times 8$ and on Houston at $\times 4$ and $\times 8$, with the largest margin on Houston $\times 4$, where PSNR rises from 46.2015 dB (MSDformer) to 46.9653 dB. The paper argues that the two ordering mechanisms address the two known Mamba weaknesses for hyperspectral data: flattening the cube makes locally adjacent pixels lose their causal connection, and the scan's output depends heavily on input order.

Load-bearing premise

The paper credits its gains to the local 3D windows (LSSP) and the spectral reordering (GSRM), but the ablation experiments toggle these together with other components, so the measured improvement is not proven to come from those two mechanisms alone.

Editorial extensions

If this is right

  • State-space models with carefully chosen scan orders can match or exceed Transformer-based models on hyperspectral image reconstruction while keeping sequence cost linear.
  • The same local-partitioning and global-reordering design could be applied to other SSM-based restoration tasks, such as denoising or fusion, where spatial-spectral adjacency matters.
  • On Chikusei and Houston, the reported margins at ×4 and ×8 show the method works across different band counts (128 and 48) and different spatial resolutions.
  • The ablation results indicate that adding more Mamba groups beyond four hurts accuracy, suggesting the architecture has an optimal depth for this task.

Reading between the lines

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

  • The paper does not report the window sizes h, w, c used in LSSP; a natural extension is to test sensitivity to these choices, since they directly control the local-global trade-off.
  • The spectral-reordering step is not tied to the state-space scan: it could be inserted into a windowed Transformer or a convolutional network to see whether similarity-based ordering is a general inductive bias or specifically beneficial to causal scans.
  • A cleaner ablation, replacing GSRM with a random permutation of bands while holding the rest fixed, would isolate whether the gains come from the similarity ordering itself or simply from any reordering of the input sequence.
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 HSRMamba, a state-space-model (Mamba) architecture for single hyperspectral image super-resolution. The method introduces a local spatial-spectral partitioning (LSSP) mechanism that feeds 3D patches to a bidirectional SSM, and a global spectral reordering mechanism (GSRM) that permutes spectral bands by global correlation before SSM processing. The network is evaluated on the Chikusei and Houston 2018 datasets at ×4 and ×8 scales with six quantitative metrics, plus additional results in the supplementary material, and it reports the best scores on all metrics in Table 1. Ablation studies in Table 2 and Table 3 isolate the effects of LSSP, GSRM, and the number of Mamba groups.

Significance. If the reported results hold, HSRMamba would be a useful contribution to single hyperspectral image super-resolution: it demonstrates that a linear-complexity Mamba backbone can be adapted to spatial-spectral structure through local partitioning and global spectral reordering, and it ships with public code, standard metrics, and comparisons to eight baselines. The empirical claim is currently plausible but not yet fully supported because the headline state-of-the-art margin is the same size as the uncontrolled ablation effect, and no variance estimates are provided.

major comments (4)
  1. [§4.3, Table 2; §3.3] The ablation does not isolate the causal role of LSSP and GSRM. The LSSP window sizes (h, w, c) are never reported, the ✗ rows do not state what replaces the removed modules, and the claim that CA/MLP are held fixed is only asserted. Quantitatively, on Chikusei ×4 HSRMamba's margin over the best baseline MSDformer is 40.2781 − 40.0902 = 0.188 dB, while Table 2 attributes 40.2781 − 40.0714 = 0.207 dB to enabling both LSSP and GSRM. With single runs and only four test images, the entire headline gain is the same size as the uncontrolled ablation effect. Please report the window sizes, the replacement operations for the ✗ rows, and per-seed or confidence-interval statistics.
  2. [§4.1, Table 1] No error bars, repeated runs, or seed information are given for any method. The Chikusei test set consists of 4 images and the Houston test set of 8 images, and the ablation differences in Table 2 are 0.09–0.12 dB in PSNR; without variance estimates these cannot be distinguished from run-to-run noise. Additionally, the protocol for the eight baselines is not described: it is unclear whether they were retrained on the same LR/HR patches or run from official checkpoints, which is necessary to interpret the comparisons.
  3. [§3.5, Eq. (9)] The loss weights λs = 0.3 and λg = 0.1 are described as chosen empirically, but no sensitivity analysis is provided. Because these weights directly trade off L1, SAM, and gradient terms, the reported PSNR/SAM/ERGAS numbers could depend on this choice. Please include a sensitivity study (e.g., a small grid over λs and λg) or otherwise justify that the results are robust to the weighting.
  4. [§3.3, Implementation Details in §4.1] The LSSP partition sizes h, w, and c are central to the proposed 'local spatial-spectral partitioning' mechanism, but they are never specified, not even in Implementation Details. Without these values, the local module is not reproducible and the claim that the improvement comes from local patch-wise causality rather than a tuned sequence-length hyperparameter is untestable. Please report the exact partition sizes used for each dataset and scale.
minor comments (5)
  1. [Throughout] There are multiple typos, including 'Spaital' in the Section 3.3 title, 'Datesets' in Section 4.1, 'partitionin' in the introduction, 'glboal' in Section 3.4, 'Sensin' in two references, and 'respresents' near Eq. (4).
  2. [Figure 4] The caption of Figure 4 contains a garbled string ('LinearLinearFlipConv1DSiLuFlipConv1DSSMSiLuSiLuSSMReverse78563412...') and should be rewritten as a readable description of the bidirectional scanning scheme.
  3. [§2.1] The sentence 'we propose the Mamba network' should read 'we propose a Mamba-based network' to avoid the impression that the paper introduces the general Mamba architecture.
  4. [§4.1] The text lists RMSE among the six evaluation metrics, but Table 1 does not report RMSE. Either report it or remove it from the list.
  5. [Abstract and Introduction] The claim of being 'the first SSM tailored for SHSR' should be softened or supported by a more systematic comparison with recently published Mamba-based hyperspectral restoration methods, since the related-work discussion only lists general Mamba models.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: results are measured on held-out test images against independent baselines.

full rationale

HSRMamba is an empirical supervised super-resolution pipeline. The claimed predictions are PSNR/SSIM/SAM/CC/ERGAS values computed on held-out test crops (Chikusei: 4 non-overlapping 512x512x128 images from the top region; Houston: 8 non-overlapping 256x256x48 images), while training uses separately cropped HR patches with LR patches generated by bicubic downsampling. No target quantity is defined in terms of a fitted parameter, and no 'prediction' reduces to a training value. The architecture is described by Equations 5-10, and component contributions are assessed by ablations in Table 2; these are standard empirical attribution statements, not derivations from premises. Self-citations (e.g., MSDformer, Chen et al. 2023b) are used as a comparison baseline and as prior architecture context, not as load-bearing justification of the present result. The central SOTA claim therefore has independent empirical content and is not circular. Concerns about missing error bars, unreported LSSP window sizes, and uncontrolled ablation variants are robustness or attribution issues, not circularity.

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

The central claim is an empirical performance comparison, so it rests on standard ML hyperparameters and domain assumptions rather than on fitted physical constants. The main free parameters are loss weights and architectural choices; the window sizes for LSSP are notably absent.

free parameters (4)
  • λ_s (SAM loss weight) = 0.3
    Chosen empirically in Eq. 9 with no sensitivity analysis; changes the loss surface and therefore the final performance.
  • λ_g (gradient loss weight) = 0.1
    Chosen empirically in Eq. 9 with no sensitivity analysis.
  • Number of CSMG groups N = 4
    Selected via ablation in Table 3; N=6 overfits and N=2 underperforms, so N=4 is a data-driven choice.
  • LSSP window sizes (h, w, c) = not specified
    The LSSP mechanism divides features into H/h × W/w × C/c blocks (Section 3.3), but the values of h, w, c are never given. These sizes control the local context size and are likely tuned, but the paper omits them.
assumptions (4)
  • domain assumption Mamba's input-order sensitivity and local forgetting are the key limitations for HSISR.
    Stated as motivation in the Introduction and abstract; not independently measured outside the proposed solutions.
  • domain assumption Bicubic downsampling is an adequate degradation model for training and evaluating HSI SR.
    All LR patches are generated by bicubic downsampling (Section 4.1), so the method may not transfer to real low-resolution HSI acquisition.
  • standard math The standard SSM discretization (zero-order hold, Eq. 2) is correct.
    Used as background from prior work without derivation in this paper.
  • domain assumption The bidirectional SSM (BSSM) from SSUMamba is a suitable base unit.
    HSRMamba builds directly on SSUMamba's BSSM as its core scanner (Section 3.3), trusting that this prior module is effective and correctly implemented.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HSRMamba: Contextual Spatial-Spectral State Space Model for Single Image Hyperspectral Super-Resolution." pith.science (2026). https://pith.science/paper/IODRP22B

@misc{pith2026250118500,
  author       = {Pith},
  title        = {Pith review of: HSRMamba: Contextual Spatial-Spectral State Space Model for Single Image Hyperspectral Super-Resolution},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IODRP22B}},
  note         = {Machine review of arXiv:2501.18500}
}
read the original abstract

Mamba has demonstrated exceptional performance in visual tasks due to its powerful global modeling capabilities and linear computational complexity, offering considerable potential in hyperspectral image super-resolution (HSISR). However, in HSISR, Mamba faces challenges as transforming images into 1D sequences neglects the spatial-spectral structural relationships between locally adjacent pixels, and its performance is highly sensitive to input order, which affects the restoration of both spatial and spectral details. In this paper, we propose HSRMamba, a contextual spatial-spectral modeling state space model for HSISR, to address these issues both locally and globally. Specifically, a local spatial-spectral partitioning mechanism is designed to establish patch-wise causal relationships among adjacent pixels in 3D features, mitigating the local forgetting issue. Furthermore, a global spectral reordering strategy based on spectral similarity is employed to enhance the causal representation of similar pixels across both spatial and spectral dimensions. Finally, experimental results demonstrate our HSRMamba outperforms the state-of-the-art methods in quantitative quality and visual results. Code is available at: https://github.com/Tomchenshi/HSRMamba.

Figures

Figures reproduced from arXiv: 2501.18500 by the authors.

Figure 1
Figure 1. Comparison of trade-offs between model performance and [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The overview of the proposed HSRMamba. HSRMamba primarily comprises multiple Contextual Spatial-Spectral Mamba Groups [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The pipeline of the designed LSSM (left) and GSCM (right). [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: The pipeline of the proposed BSSM. In this paper, we introduce a BSSM [Fu et al., 2024] as the foundational unit for global spatial-spectral feature ex￾traction. As shown in [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Visual results on the Chikusei dataset with spectral bands 70-100-36 as R-G-B at scale factor [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Visual results on the Houston dataset with spectral bands 26-20-10 as R-G-B at scale factor [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Mean spectral difference curves of different methods on [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 6
Figure 6. Figure 6: We can also observe that HSRMamba provides more [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 39 canonical work pages

  1. [1]

    Mask-guided spectral-wise transformer for efficient hyperspectral image reconstruction

    [Cai et al., 2022] Yuanhao Cai, Jing Lin, Xiaowan Hu, Hao- qian Wang, Xin Yuan, Yulun Zhang, Radu Timofte, and Luc Van Gool. Mask-guided spectral-wise transformer for efficient hyperspectral image reconstruction. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17481–17490,

    work page 2022

  2. [4]

    Bidirectional dilation trans- former for multispectral and hyperspectral image fusion

    [Deng et al., 2023] Shangqi Deng, Liang-Jian Deng, Xiao Wu, Ran Ran, and Rui Wen. Bidirectional dilation trans- former for multispectral and hyperspectral image fusion. In Proceedings of the International Joint Conference on Artificial Intelligence, pages 3633–3641,

    work page 2023

  3. [5]

    Hyper- spectral image super-resolution via subspace-based low tensor multi-rank regularization

    [Dian and Li, 2019] Renwei Dian and Shutao Li. Hyper- spectral image super-resolution via subspace-based low tensor multi-rank regularization. IEEE Transactions on Image Processing, 28(10):5135–5146,

    work page 2019

  4. [8]

    Bidirectional 3d quasi-recurrent neural network for hyper- spectral image super-resolution

    [Fu et al., 2021] Ying Fu, Zhiyuan Liang, and Shaodi You. Bidirectional 3d quasi-recurrent neural network for hyper- spectral image super-resolution. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sens- ing, 14:2674–2688,

    work page 2021

  5. [10]

    Mamba: Linear- time sequence modeling with selective state spaces

    [Gu and Dao, 2023] Albert Gu and Tri Dao. Mamba: Linear- time sequence modeling with selective state spaces. arXiv preprint arXiv: 2312.00752,

    arXiv 2023

  6. [11]

    Efficiently modeling long sequences with structured state spaces

    [Gu et al., 2022] Albert Gu, Karan Goel, and Christopher R´e. Efficiently modeling long sequences with structured state spaces. In Proceedings of the International Confer- ence on Learning Representations,

    work page 2022

  7. [13]

    Mambair: A sim- ple baseline for image restoration with state-space model

    [Guo et al., 2024] Hang Guo, Jinmin Li, Tao Dai, Zhihao Ouyang, Xudong Ren, and Shu-Tao Xia. Mambair: A sim- ple baseline for image restoration with state-space model. In Proceedings of the European Conference on Computer Vision, volume 15076, pages 222–241,

    work page 2024

  8. [14]

    A self-supervised remote sensing image fusion framework with dual-stage self-learning and spectral super-resolution injection

    [He et al., 2023] Jiang He, Qiangqiang Yuan, Jie Li, Yi Xiao, and Liangpei Zhang. A self-supervised remote sensing image fusion framework with dual-stage self-learning and spectral super-resolution injection. ISPRS Journal of Pho- togrammetry and Remote Sensing, 204:131–144,

    work page 2023

Show all 42 references
  1. [15]

    Exploring the spectral prior for hyperspectral image super-resolution

    [Hu et al., 2024] Qian Hu, Xinya Wang, Junjun Jiang, Xiao- Ping Zhang, and Jiayi Ma. Exploring the spectral prior for hyperspectral image super-resolution. IEEE Transactions on Image Processing, 33:5260–5272,

    2024

  2. [16]

    A comparative study of spatial approaches for urban map- ping using hyperspectral rosis images over pavia city, northern italy

    [Huang and Zhang, 2009] Xin Huang and Liangpei Zhang. A comparative study of spatial approaches for urban map- ping using hyperspectral rosis images over pavia city, northern italy. International Journal of Remote Sensing , 30(12):3205–3221,

    2009

  3. [18]

    Learning spatial-spectral prior for super- resolution of hyperspectral imagery

    [Jiang et al., 2020] Junjun Jiang, He Sun, Xianming Liu, and Jiayi Ma. Learning spatial-spectral prior for super- resolution of hyperspectral imagery. IEEE Trans. Compu- tational Imaging, 6:1082–1096,

    2020

  4. [21]

    Ex- ploring the relationship between 2d/3d convolution for hy- perspectral image super-resolution

    [Li et al., 2021] Qiang Li, Qi Wang, and Xuelong Li. Ex- ploring the relationship between 2d/3d convolution for hy- perspectral image super-resolution. IEEE Transactions on Geoscience and Remote Sensing , 59(10):8693–8703,

    2021

  5. [22]

    Dual-stage approach toward hyperspectral im- age super-resolution

    [Li et al., 2022] Qiang Li, Yuan Yuan, Xiuping Jia, and Qi Wang. Dual-stage approach toward hyperspectral im- age super-resolution. IEEE Transactions on Image Pro- cessing, 31:7252–7263,

    2022

  6. [23]

    Learning attention from attention: Efficient self-refinement transformer for face super-resolution

    [Li et al., 2023] Guanxin Li, Jingang Shi, Yuan Zong, Fei Wang, Tian Wang, and Yihong Gong. Learning attention from attention: Efficient self-refinement transformer for face super-resolution. In Proceedings of the International Joint Conference on Artificial Intelligence , pag...

    2023

  7. [24]

    Swinir: Image restoration using swin transformer

    [Liang et al., 2021] Jingyun Liang, Jiezhang Cao, Guolei Sun, Kai Zhang, Luc Van Gool, and Radu Timofte. Swinir: Image restoration using swin transformer. In Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops, pages 1833–1844,

    2021

  8. [25]

    Interactformer: Interac- tive transformer and CNN for hyperspectral image super- resolution

    [Liu et al., 2022] Yaoting Liu, Jianwen Hu, Xudong Kang, Jing Luo, and Shaosheng Fan. Interactformer: Interac- tive transformer and CNN for hyperspectral image super- resolution. IEEE Transactions on Geoscience and Remote Sensin, 60:1–15,

    2022

  9. [26]

    Medical hy- perspectral imaging: a review

    [Lu and Fei, 2014] Guolan Lu and Baowei Fei. Medical hy- perspectral imaging: a review. Journal of biomedical op- tics, 19(1):010901–010901,

    2014

  10. [28]

    Hyperspectral image spatial super-resolution via 3d full convolutional neural network

    [Mei et al., 2017] Shaohui Mei, Xin Yuan, Jingyu Ji, Yifan Zhang, Shuai Wan, and Qian Du. Hyperspectral image spatial super-resolution via 3d full convolutional neural network. Remote Sensing, 9(11):1139,

    2017

  11. [29]

    Hi-mamba: Hierarchical mamba for efficient image super-resolution.arXiv preprint arXiv:2410.10140,

    [Qiao et al., 2024] Junbo Qiao, Jincheng Liao, Wei Li, Yu- lun Zhang, Yong Guo, Yi Wen, Zhangxizi Qiu, Jiao Xie, Jie Hu, and Shaohui Lin. Hi-mamba: Hierarchical mamba for efficient image super-resolution.arXiv preprint arXiv:2410.10140,

    2024 arXiv

  12. [30]

    Aitken, Rob Bishop, Daniel Rueckert, and Zehan Wang

    [Shi et al., 2016] Wenzhe Shi, Jose Caballero, Ferenc Huszar, Johannes Totz, Andrew P. Aitken, Rob Bishop, Daniel Rueckert, and Zehan Wang. Real-time single im- age and video super-resolution using an efficient sub-pixel convolutional neural network. In Proceedings of the IEEE...

    2016

  13. [32]

    Hyperspectral image super-resolution meets deep learning: A survey and perspective

    [Wang et al., 2023] Xinya Wang, Qian Hu, Yingsong Cheng, and Jiayi Ma. Hyperspectral image super-resolution meets deep learning: A survey and perspective. IEEE/CAA Jour- nal of Automatica Sinica, 10(8):1668–1691,

    2023

  14. [33]

    Combining global receptive field and spatial spectral information for single-image hyperspectral super-resolution

    [Wu et al., 2023] Yiming Wu, Ronghui Cao, Yikun Hu, Jin Wang, and Kenli Li. Combining global receptive field and spatial spectral information for single-image hyperspectral super-resolution. Neurocomputing, 542:126277,

    2023

  15. [34]

    Hyper- spectral image denoising using uncertainty-aware adjus- tor

    [Xiao and Wei, 2023] Jiahua Xiao and Xing Wei. Hyper- spectral image denoising using uncertainty-aware adjus- tor. In Proceedings of the International Joint Conference on Artificial Intelligence, pages 1560–1568,

    2023

  16. [35]

    Frequency-assisted mamba for remote sensing image super-resolution

    [Xiao et al., 2024] Yi Xiao, Qiangqiang Yuan, Kui Jiang, Yuzeng Chen, Qiang Zhang, and Chia-Wen Lin. Frequency-assisted mamba for remote sensing image super-resolution. IEEE Transactions on Multimedia , 27:1783–1796,

    2024

  17. [36]

    As 3itransunet: Spatial-spectral interactive transformer u-net with alternating sampling for hyperspec- tral image super-resolution

    [Xu et al., 2023] Qin Xu, Shiji Liu, Jiahui Wang, Bo Jiang, and Jin Tang. As 3itransunet: Spatial-spectral interactive transformer u-net with alternating sampling for hyperspec- tral image super-resolution. IEEE Transactions on Geo- science and Remote Sensing, 61:1–13,

    2023

  18. [37]

    Airborne hyperspectral data over chikusei

    [Yokoya and Iwasaki, 2016] Naoto Yokoya and Akira Iwasaki. Airborne hyperspectral data over chikusei. Space Appl. Lab., Univ. Tokyo, Tokyo, Japan, Tech. Rep. SAL-2016-05-27,

    2016

  19. [38]

    Image super- resolution using very deep residual channel attention net- works

    [Zhang et al., 2018] Yulun Zhang, Kunpeng Li, Kai Li, Lichen Wang, Bineng Zhong, and Yun Fu. Image super- resolution using very deep residual channel attention net- works. In Proceedings of the European Conference on Computer Vision, volume 11211, pages 294–310,

    2018

  20. [39]

    Es- saformer: Efficient transformer for hyperspectral image super-resolution

    [Zhang et al., 2023] Mingjin Zhang, Chi Zhang, Qiming Zhang, Jie Guo, Xinbo Gao, and Jing Zhang. Es- saformer: Efficient transformer for hyperspectral image super-resolution. In Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision,

    2023

  21. [40]

    Perceive-ir: Learn- ing to perceive degradation better for all-in-one image restoration

    [Zhang et al., 2025] Xu Zhang, Jiaqi Ma, Guoli Wang, Qian Zhang, Huan Zhang, and Lefei Zhang. Perceive-ir: Learn- ing to perceive degradation better for all-in-one image restoration. IEEE Transactions on Image Processing ,

    2025

  22. [42]

    Vision mamba: Efficient visual representation learning with bidirectional state space model

    [Zhu et al., 2024] Lianghui Zhu, Bencheng Liao, Qian Zhang, Xinlong Wang, Wenyu Liu, and Xinggang Wang. Vision mamba: Efficient visual representation learning with bidirectional state space model. In Proceedings of the International Conference on Machine Learning, 2024

    2024

  23. [2009]

    Localmamba: Visual state space model with windowed selective scan

    [Huang et al., 2024] Tao Huang, Xiaohuan Pei, Shan You, Fei Wang, Chen Qian, and Chang Xu. Localmamba: Visual state space model with windowed selective scan. arXiv preprint arXiv:2403.09338,

    2024 arXiv

  24. [2014]

    Recent advances of hyper- spectral imaging technology and applications in agricul- ture

    [Lu et al., 2020] Bing Lu, Phuong D Dao, Jiangui Liu, Yuhong He, and Jiali Shang. Recent advances of hyper- spectral imaging technology and applications in agricul- ture. Remote Sensing, 12(16):2659,

    2020

  25. [2016]

    Panchromatic and hyperspectral image fusion: Outcome of the 2022 whispers hyperspectral pansharpening challenge

    [Vivone et al., 2022] Gemine Vivone, Andrea Garzelli, Yang Xu, Wenzhi Liao, and Jocelyn Chanussot. Panchromatic and hyperspectral image fusion: Outcome of the 2022 whispers hyperspectral pansharpening challenge. IEEE Journal of Selected Topics in Applied Earth Observations and...

    2022

  26. [2017]

    Noise prior knowledge informed bayesian inference network for hyperspectral super-resolution

    [Dong et al., 2023] Wenqian Dong, Jiahui Qu, Song Xiao, Tongzhen Zhang, Yunsong Li, and Xiuping Jia. Noise prior knowledge informed bayesian inference network for hyperspectral super-resolution. IEEE Transactions on Im- age Processing, 32:3121–3135,

    2023

  27. [2018]

    Mixed 2d/3d convolutional network for hyperspectral image super-resolution

    [Li et al., 2020] Qiang Li, Qi Wang, and Xuelong Li. Mixed 2d/3d convolutional network for hyperspectral image super-resolution. Remote Sensing, 12(10):1660,

    2020

  28. [2019]

    Hyperspectral image super-resolution via non-local sparse tensor factorization

    [Dian et al., 2017] Renwei Dian, Leyuan Fang, and Shutao Li. Hyperspectral image super-resolution via non-local sparse tensor factorization. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 3862–3871,

    2017

  29. [2020]

    Single hyperspectral image super- resolution with grouped deep recursive residual network

    [Li et al., 2018] Yong Li, Lei Zhang, Chen Ding, Wei Wei, and Yanning Zhang. Single hyperspectral image super- resolution with grouped deep recursive residual network. In Proceedings of the International Conference on Multi- media Big Data, pages 1–4,

    2018

  30. [2021]

    Ssumamba: Spatial-spectral selective state space model for hyperspectral image denoising.IEEE Transactions on Geoscience and Remote Sensing,

    [Fu et al., 2024] Guanyiman Fu, Fengchao Xiong, Jianfeng Lu, and Jun Zhou. Ssumamba: Spatial-spectral selective state space model for hyperspectral image denoising.IEEE Transactions on Geoscience and Remote Sensing,

    2024

  31. [2022]

    Toward sta- ble, interpretable, and lightweight hyperspectral super- resolution

    [Guo et al., 2023] Wen-jin Guo, Weiying Xie, Kai Jiang, Yunsong Li, Jie Lei, and Leyuan Fang. Toward sta- ble, interpretable, and lightweight hyperspectral super- resolution. In Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition , pages 22272–22281,

    2023

  32. [2023]

    Cross-scope spatial-spectral information aggrega- tion for hyperspectral image super-resolution.IEEE Trans- actions on Image Processing, 33:5878–5891,

    [Chen et al., 2024] Shi Chen, Lefei Zhang, and Liangpei Zhang. Cross-scope spatial-spectral information aggrega- tion for hyperspectral image super-resolution.IEEE Trans- actions on Image Processing, 33:5878–5891,

    2024

  33. [2024]

    Freqformer: Frequency-aware transformer for lightweight image super- resolution

    [Dai et al., 2024] Tao Dai, Jianping Wang, Hang Guo, Jin- min Li, Jinbao Wang, and Zexuan Zhu. Freqformer: Frequency-aware transformer for lightweight image super- resolution. In Proceedings of the International Joint Con- ference on Artificial Intelligence, pages 731–739,

    2024

  34. [2025]

    Mambaformersr: A lightweight model for remote-sensing image super-resolution

    [Zhi et al., 2024] Ruicong Zhi, Xiaopei Fan, and Jingye Shi. Mambaformersr: A lightweight model for remote-sensing image super-resolution. IEEE Geoscience and Remote Sensing Letters, 21:1–5,

    2024

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.