Pith. sign in

REVIEW 3 major objections 4 minor 51 references

PCMamba: Physics-Informed Cross-Modal State Space Model for Dual-Camera Compressive Hyperspectral Imaging

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read PCMamba reconstructs dual-camera hyperspectral images by factoring them into temperature, emissivity, and texture, reaching 44.47 dB average PSNR.

desk verdict The engineering in PCMamba is solid and the reconstruction numbers are strong, but the physics-informed core is a thermal model applied to visible-light data, so the T/e maps are free latents and the central claim does not hold. read the letter →

arxiv 2505.16373 v1 pith:CY4KLNFN submitted 2025-05-22 eess.IV cs.CV

classification eess.IVcs.CV
keywords hyperspectralimagingcompressivesensingdual-camerastatespacemodelMambaphysics-informeddeeplearningthermalemissionimagereconstruction
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

PCMamba is a deep network for reconstructing hyperspectral images from the compressed measurement and the panchromatic (PAN) image of a dual-camera snapshot system. Its central claim is that reconstruction improves when the network is built around the physical way the signal was produced: a spectral sample is modeled as the object's thermal emission, set by temperature and emissivity through Planck's law, plus reflected environmental radiation that carries texture. The network decodes temperature, emissivity, and texture from the two inputs and resynthesizes the hyperspectral cube with that same forward equation. On simulated and real benchmarks it reports 44.47 dB average PSNR and 0.994 SSIM, about 0.66 dB above the next-best method, with a Mamba backbone that keeps sequence processing linear. If the physics-informed factorization is what delivers the gain, then the network yields interpretable physical intermediates instead of a pure black-box mapping.

What carries the argument

Two mechanisms carry the argument. The TeX (temperature-emissivity-texture) decomposition is the forward physical synthesis rule of Eq. (12), which writes the measured signal as emissivity-weighted blackbody radiation $e_\lambda B_\lambda(T)$ plus reflected environmental radiation $(1-e_\lambda)X_\lambda$; it turns the reconstruction problem into one of disentangling the latent physical factors $T$, $e$, and $X$. The Cross-Modal Scanning Mamba Block (CSMB) is the network module that performs non-overlapping pixel-wise cross-scanning of backbone features and PAN features, running through the sequence depthwise convolution, SiLU, state space model, and layer normalization; scanning each pixel position only once across the two modalities halves the sequence length and adds a positional inductive bias. The loss $L_{\text{total}} = L_{\text{Rec}} + L_M$ closes the loop, with $L_M$ re-coding the synthesized cube through the coded aperture so the reconstruction is checked against the original measurement.

What would settle it

Retrain PCMamba on the visible-range CAVE/KAIST scenes with the blackbody term $e_\lambda B_\lambda(T)$ in Eq. (12) forced to zero. If average PSNR stays near 44.47 dB, the Planck-law path is not carrying the improvement. A complementary check is to inspect the learned temperature maps on a scene with known room-temperature objects: physically meaningful temperatures would cluster near 300 K, whereas a free latent field would spread arbitrarily across the scene.

Watch

Extended reading notes

Core claim

On its own terms, the discovery is that the limiting factor in dual-camera compressive hyperspectral imaging is not only spectral-spatial modeling but the neglect of the physical imaging process. The paper asserts that if temperature $T$, emissivity $e$, and texture $X$ are accurately recovered, the HSI can be synthesized by the forward rule $T = e_\lambda B_\lambda(T) + (1-e_\lambda)X_\lambda$ (Eq. 12), where $B_\lambda(T)$ is Planck's blackbody radiation. PCMamba instantiates this by giving its U-Net three decoders that output $T$, $e$, and $X$ and combining them according to that rule, and it adds a measurement-consistency loss that re-compresses the reconstruction through the same coded mask and compares with the input measurement. The ablations attribute 0.82 dB of the average PSNR to the TeX decomposition and 1.91 dB to the cross-modal scan, supporting the claim that the factorization, not just the backbone, drives the improvement.

Load-bearing premise

The load-bearing premise is that the measured hyperspectral signal is mainly the scene's own heat radiation (written as Planck's-law blackbody emission) plus reflected light from the surroundings. On the visible-range CAVE and KAIST test sets used in the paper, room-temperature objects emit essentially no visible blackbody radiation, so if that term is negligible the physics equation is a weak constraint and the reported gain would have to come from the network architecture instead.

Editorial extensions

If this is right

  • If the TeX path is correct, the temperature, emissivity, and texture maps are interpretable byproducts, so one network returns material and thermal estimates alongside the reconstructed hyperspectral cube.
  • The non-overlapping cross-scan halves the state-space-model sequence length, which is why PCMamba reports a higher average PSNR than 9-stage transformer baselines while using less than half their computational cost.
  • The measurement-consistency loss $L_M$ provides a check that needs no ground truth: any candidate reconstruction can be re-compressed through the coded mask and compared with the captured measurement.
  • Removing $L_M$ costs 0.39 dB and removing $L_{\text{Rec}}$ collapses PSNR to 17.29 dB, so the final quality depends on both the physics-informed decoder structure and the joint loss, not on either alone.

Reading between the lines

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

  • Because the CAVE and KAIST benchmarks are visible-range scenes, ambient-temperature blackbody radiation is negligible there; Eq. (12) effectively reduces to $T \approx (1-e_\lambda)X_\lambda$, so on these datasets the learned temperature is most likely a latent code, not a physical temperature. A thermal-range (LWIR) test would be needed for the physics to be active.
  • The same forward-model-plus-consistency-loss scheme could in principle be transferred to single-camera CASSI or snapshot video by replacing the PAN branch with any auxiliary image source; the paper does not test those settings.
  • If emissivity is truly a material fingerprint, the emissivity decoder should give stable, similar maps across different scenes containing the same material; the paper reports no such cross-scene consistency check.
  • The 0.82 dB ablation gain of TeX decomposition conflates two changes: the three-branch decoder and the physics synthesis rule; a controlled test that keeps the three branches but replaces Eq. (12) with a learned combination would separate the physics from the extra capacity.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes PCMamba, a state-space-model network for dual-camera compressive hyperspectral imaging. The method couples a U-Net-like Mamba architecture with a 'TeX decomposition' that models the captured hyperspectral signal as a physics-based synthesis of temperature T, emissivity e, and texture X, culminating in Eq. (12): T = eλ Bλ(T) + (1−eλ)Xλ. The network is trained with an L1 reconstruction loss and a measurement-consistency loss. On the standard CAVE/KAIST simulated benchmarks and a real dataset, the paper reports an average PSNR of 44.47 dB and SSIM of 0.994, outperforming the compared methods, and ablation studies attribute gains to the TeX decomposition, the Cross-Modal Scanning Mamba Block, and the consistency loss.

Significance. If the physical interpretation were supported, the paper would make a timely contribution: it is among the first attempts to inject a thermal-emission forward model into deep HSI reconstruction, and it demonstrates a computationally efficient cross-modal scanning mechanism. The empirical side is carefully executed within the paper's own protocol: Table 1 gives per-scene comparisons against multiple recent methods, and Table 2 reports controlled ablations for the proposed modules and losses. These results alone support a useful 'physics-inspired' reparameterization. However, the central claim that the network performs a forward physical synthesis of visible-range images is not supported by the evidence: the thermal model in Eq. (12) is negligible in the spectral range of the test data, and the learned temperature/emissivity maps are never validated against physical measurements. The physics narrative therefore currently outruns the data, and this needs to be addressed before publication.

major comments (3)
  1. [Section 3.1, Eqs. (10)-(12)] The forward model is a thermal-emission model, but the evaluation data are in the visible range. CAVE and KAIST are visible-range hyperspectral datasets, and Fig. 6 shows reconstructed bands at 486.9 nm, 543.8 nm, 584.3 nm, and 641.1 nm. At a scene temperature of roughly 300 K, Planck's law in Eq. (7) gives a blackbody radiance at 500 nm on the order of 10^-25 W m^-3 sr^-1, which is negligible compared with reflected visible radiance. Consequently, Eq. (12) degenerates to T ≈ (1−eλ)Xλ, and the temperature map T and emissivity map eλ are not identifiable from the measured visible radiance: infinitely many (T, eλ) pairs produce essentially the same signal. The manuscript never specifies the spectral regime of the test data or the assumed scene temperature. To support the 'forward physical process' claim, the authors should either validate the model on thermal-range data where Planckian emission is dominant, or substantially reframe Section 3.1 as a physics-inspired latent factorization rather than a physical forward model.
  2. [Section 4.5, Table 2(a)] The three decoders output temperature T, emissivity e, and texture X, but these maps are never compared with any physical ground truth or independent measurement. They are free latent maps optimized only by the L1 reconstruction loss (Eq. 20) and the measurement-consistency loss (Eq. 21), and the latter checks the final HSI against the input measurement, not the physical correctness of T or e. The ablation in Table 2(a) shows a 0.82 dB improvement when the three decoders are added, but this only establishes that this particular learned reparameterization helps reconstruction; it does not establish that Planckian emission is being modeled. The paper needs either physical validation of the T/e maps, or a clear statement that these are latent variables with a physics-inspired structure rather than estimated physical quantities.
  3. [Section 3.1, Eq. (14)] The assumption that emissivity is approximately constant across the working wavelength range is asserted with a reference to [3] and is not justified for the visible-range datasets used in Section 4.1. Even if this assumption were adopted, it does not resolve the identifiability problem: with a negligible blackbody term, Eq. (14) still reduces to a product of a constant e and texture X, so T and e remain free parameters. The authors should either supply a physical justification for the constant-emissivity assumption in the relevant regime or remove it from the core derivation.
minor comments (4)
  1. [Section 3.1, Eqs. (10)-(12)] The symbol T is used both for total radiance on the left-hand side of Eqs. (10)-(12) and for thermodynamic temperature inside Bλ(T). This conflation makes the derivation difficult to follow; use distinct symbols such as L and T_phys.
  2. [Section 4.1] Please state explicitly the spectral ranges of the CAVE, KAIST, and real datasets, and clarify whether the real dataset has ground truth or is used only for visual evaluation. This information is essential for assessing whether Eq. (12) is applicable to the reported experiments.
  3. [Table 1] The header 'PFsion-RGB' appears to be a typo for 'PFusion-RGB'; also, GFLOPs are reported for some methods but not for PFusion and PIDS. Please make the table consistent.
  4. [Section 3.3, Eq. (19)] The description of the cross-scanning operation would benefit from a precise definition of how the two feature maps are interleaved without 'pixel position repetition' and how the positional inductive bias arises; the current text and Fig. 4 leave this to the reader's inference.

Circularity Check

2 steps flagged · score 6.0 of 10

The physics-informed core of PCMamba reduces, on the visible CAVE/KAIST test bands, to a learned multiplicative reparameterization: Planck emission is negligible, so Eq. 12 degenerates and the temperature/emissivity outputs are unvalidated fitted latents.

  1. renaming known result [Section 3.1, Eqs. (6)-(12); Section 4.5, Table 2(a)]
    "T=γ(eλBλ(T)+(1−eλ)Xλ)+(1−γ)Λ, (11) ... T=eλBλ(T)+(1−eλ)Xλ. (12) This implies that the HSI can be synthesized through a forward physical process if the temperature T, emissivity e, and texture X can be accurately obtained."

    At the wavelengths of the evaluation datasets (CAVE/KAIST; Fig. 6 shows 486.9-641.1 nm), a 300 K object has Bλ(T) essentially zero, so Eq. 12 degenerates to T ≈ (1−eλ)Xλ. Consequently the Planck/emissivity term carries no information: T and eλ are unidentifiable free latent maps, and any HSI can be represented by the trivial choice eλ=0, Xλ=HSI. The 'forward physical synthesis' is therefore only a learned multiplicative reparameterization of decoder outputs, and the +0.82 dB ablation gain for TeX does not demonstrate that Planckian emission is modeled. Since no physical ground truth for T or e is reported, the physical vocabulary is a renaming of the network's fitted latents rather than a derivation.

  2. fitted input called prediction [Section 3.3, decoder paragraph; Section 3.4, Eqs. (20)-(22)]
    "Finally, three decoders are applied to the output features of the U-net to generate the desired temperature T, emissivity e, and texture X. ... Ltotal = LRec + LM."

    The three quantities called temperature, emissivity, and texture are produced by decoders at the end of the U-net and are supervised only by LRec (Eq. 20) and LM (Eq. 21). These losses enforce agreement with the target HSI and the input measurement, not with any physical temperature or emissivity measurement. The maps are therefore fitted degrees of freedom: whatever values minimize the reconstruction loss are labeled 'physical properties'. In the visible regime where the Planck term vanishes, the fitting is almost unconstrained, so the claimed prediction of T and e is a relabeling of the optimized latent variables, not an independently derived physical result.

full rationale

The reconstruction numbers themselves are not circular: PCMamba is trained and evaluated against external CAVE, KAIST, and real HSI ground truths, and the PSNR/SSIM improvements over SOTA are genuine empirical results. The circularity lies in the paper's central physics claim. The forward model Eq. 12 is presented as Planck-law-based synthesis, but the test data are visible-range HSIs where ambient-temperature blackbody radiation is negligible. The equation then reduces to a product/selection of two unconstrained latent fields, and the 'temperature' and 'emissivity' maps are fit by the same reconstruction and consistency losses whose outputs they are supposed to explain. No physical measurement validates T or e, and the ablation only shows that adding the TeX parameterization helps as a learned transform. Thus the physics-informed interpretation is a renaming/fitted labeling rather than an independent constraint, warranting a partial circularity score.

Assumptions & free parameters 1 free parameters · 6 assumptions · 1 invented entities

The central mechanism relies on a thermal-radiative decomposition (Eqs. 5-12) imported from Ref. [3]. That decomposition presupposes thermal emission, while the CAVE and KAIST data are visible-light images. The network then adds three decoder maps named temperature, emissivity, and texture; these maps are unmeasured latent fields fit by the reconstruction loss, so they carry no independent evidence. The BFR hyperparameter (0.7) is tuned by the ablation in Table 2(b). No formal verification, code, or independent measurement is supplied to ground the physics claim.

free parameters (1)
  • BFR (backbone feature ratio) = 0.7
    Tuned by ablation in Table 2(b); values 0.3 to 0.9 were tested and 0.7 was chosen. It controls the proportion of backbone features in the CSMB cross-scanning and directly affects the reported PSNR.
assumptions (6)
  • standard math Planck's law (Eq. 7) gives the blackbody radiation term Bλ(T).
    Governing equation for blackbody radiation; a standard physics input, not derived in the paper.
  • domain assumption The CASSI and PAN measurements are linear operators plus Gaussian noise (Eqs. 1-2).
    Assumes a linear measurement model, standard in compressive imaging but not derived or verified for the real system.
  • domain assumption Environmental emission is (1-eλ)Xλ, where Xλ aggregates reflected radiation from surrounding objects (Eq. 8).
    A Kirchhoff-style radiative model borrowed from Ref. [3]; not validated for the test scenes.
  • ad hoc to paper Emissivity is approximately constant for a material across the working wavelength range (Eq. 14).
    The paper assumes eλ is constant so it can factor it out of the integral. This simplification is taken from Ref. [3] and is not tested on CAVE or KAIST data.
  • domain assumption Atmospheric transmissivity γ is close to 1, so atmospheric radiation can be dropped (Eq. 12).
    The paper states γ is typically close to 1 due to water vapor and CO2 absorption, making the approximation plausible, but no measurement is provided.
  • domain assumption The evaluated hyperspectral scenes are in a regime where thermal emission is a significant part of the signal (Eq. 12).
    This is the strongest hidden assumption. The CAVE and KAIST datasets are visible-range images, where ambient-temperature blackbody radiation is negligible, so the first term of Eq. 12 cannot be physically significant.
invented entities (1)
  • Learned temperature, emissivity, and texture maps (T, e, X)
    purpose: To synthesize the HSI via Eq. 12 and justify the physics-informed reconstruction narrative.
    The paper never compares these maps to any measured temperature or emissivity, so they are free latent variables with no independent falsifiable handle. They are fit by the reconstruction loss rather than grounded in physical measurement.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PCMamba: Physics-Informed Cross-Modal State Space Model for Dual-Camera Compressive Hyperspectral Imaging." pith.science (2026). https://pith.science/paper/CY4KLNFN

@misc{pith2026250516373,
  author       = {Pith},
  title        = {Pith review of: PCMamba: Physics-Informed Cross-Modal State Space Model for Dual-Camera Compressive Hyperspectral Imaging},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CY4KLNFN}},
  note         = {Machine review of arXiv:2505.16373}
}
read the original abstract

Panchromatic (PAN) -assisted Dual-Camera Compressive Hyperspectral Imaging (DCCHI) is a key technology in snapshot hyperspectral imaging. Existing research primarily focuses on exploring spectral information from 2D compressive measurements and spatial information from PAN images in an explicit manner, leading to a bottleneck in HSI reconstruction. Various physical factors, such as temperature, emissivity, and multiple reflections between objects, play a critical role in the process of a sensor acquiring hyperspectral thermal signals. Inspired by this, we attempt to investigate the interrelationships between physical properties to provide deeper theoretical insights for HSI reconstruction. In this paper, we propose a Physics-Informed Cross-Modal State Space Model Network (PCMamba) for DCCHI, which incorporates the forward physical imaging process of HSI into the linear complexity of Mamba to facilitate lightweight and high-quality HSI reconstruction. Specifically, we analyze the imaging process of hyperspectral thermal signals to enable the network to disentangle the three key physical properties-temperature, emissivity, and texture. By fully exploiting the potential information embedded in 2D measurements and PAN images, the HSIs are reconstructed through a physics-driven synthesis process. Furthermore, we design a Cross-Modal Scanning Mamba Block (CSMB) that introduces inter-modal pixel-wise interaction with positional inductive bias by cross-scanning the backbone features and PAN features. Extensive experiments conducted on both real and simulated datasets demonstrate that our method significantly outperforms SOTA methods in both quantitative and qualitative metrics.

Figures

Figures reproduced from arXiv: 2505.16373 by the authors.

Figure 1
Figure 1. Our physics-informed HSI reconstruc￾tion method. Traditional methods utilize hand-designed pri￾ors for reconstruction, such as sparsity [26, 42], non-local similarity [14, 16], low-rank [28], and total variation [23, 39]. However, these methods require manual adjustment of the parameters, which often leads to mismatches between the prior assumptions and the actual problem. In deep learning-based methods, end-to-end … view at source ↗
Figure 2
Figure 2. The dual-camera compressive hy￾perspectral imaging system. The principle of DCCHI is shown in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 1
Figure 1. Therefore, to fully describe the hyperspectral signal, it is necessary to consider the emission [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figures from the paper (4 more)
Figure 3
Figure 3. Figure 3: Overview of PCMamba. PCMamba consists of a state space model network with a U-net [PITH_FULL_IMAGE:figures/full_fig_p005_3.png]
Figure 4
Figure 4. Figure 4: Illustration of Cross-Modal Scanning Mamba Block (CSMB). CSMB performs non [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 6
Figure 6. Figure 6: Comparison of the reconstruction results of different methods on two scenes from the [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 5
Figure 5. Figure 5: The visual comparisons be￾tween our method and SOTA methods on real dataset. Numerical Results. The metrics of different methods on ten simulated scenes are shown in Tab. 1. Our method achieves superior performance in most scenes. The average PSNR and average SSIM of o…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

51 extracted references · 43 canonical work pages

  1. [3]

    Heat-assisted detection and ranging.Nature, 619(7971):743–748, 2023

    Fanglin Bao, Xueji Wang, Shree Hari Sureshbabu, Gautam Sreekumar, Liping Yang, Vaneet Aggarwal, Vishnu N Boddeti, and Zubin Jacob. Heat-assisted detection and ranging.Nature, 619(7971):743–748, 2023

  2. [1]

    Compressive coded aperture spectral imaging: An introduction.IEEE Signal Processing Magazine, 31(1):105–115, 2013

    Gonzalo R Arce, David J Brady, Lawrence Carin, Henry Arguello, and David S Kittle. Compressive coded aperture spectral imaging: An introduction.IEEE Signal Processing Magazine, 31(1):105–115, 2013

  3. [2]

    Higher-order computational model for coded aperture spectral imaging.Applied optics, 52(10):D12–D21, 2013

    Henry Arguello, Hoover Rueda, Yuehao Wu, Dennis W Prather, and Gonzalo R Arce. Higher-order computational model for coded aperture spectral imaging.Applied optics, 52(10):D12–D21, 2013

  4. [4]

    A new twist: Two-step iterative shrinkage/thresholding algorithms for image restoration.IEEE Transactions on Image processing, 16(12):2992–3004, 2007

    José M Bioucas-Dias and Mário AT Figueiredo. A new twist: Two-step iterative shrinkage/thresholding algorithms for image restoration.IEEE Transactions on Image processing, 16(12):2992–3004, 2007

  5. [5]

    Coarse-to-fine sparse transformer for hyperspectral image reconstruction

    Yuanhao Cai, Jing Lin, Xiaowan Hu, Haoqian Wang, Xin Yuan, Yulun Zhang, Radu Timofte, and Luc Van Gool. Coarse-to-fine sparse transformer for hyperspectral image reconstruction. InEuropean conference on computer vision, pages 686–704. Springer, 2022

  6. [6]

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

    Yuanhao Cai, Jing Lin, Xiaowan Hu, Haoqian Wang, Xin Yuan, Yulun Zhang, Radu Timofte, and Luc Van Gool. Mask-guided spectral-wise transformer for efficient hyperspectral image reconstruction. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 17502–17511, 2022

  7. [7]

    Mst++: Multi-stage spectral-wise transformer for efficient spectral reconstruction

    Yuanhao Cai, Jing Lin, Zudi Lin, Haoqian Wang, Yulun Zhang, Hanspeter Pfister, Radu Timofte, and Luc Van Gool. Mst++: Multi-stage spectral-wise transformer for efficient spectral reconstruction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 745–755, 2022

  8. [8]

    Degradation-aware unfolding half-shuffle transformer for spectral compressive imaging.Advances in Neural Information Processing Systems, 35:37749–37761, 2022

    Yuanhao Cai, Jing Lin, Haoqian Wang, Xin Yuan, Henghui Ding, Yulun Zhang, Radu Timofte, and Luc V Gool. Degradation-aware unfolding half-shuffle transformer for spectral compressive imaging.Advances in Neural Information Processing Systems, 35:37749–37761, 2022

Show all 51 references
  1. [9]

    Binarized spectral compressive imaging.Advances in Neural Information Processing Systems, 36, 2024

    Yuanhao Cai, Yuxin Zheng, Jing Lin, Xin Yuan, Yulun Zhang, and Haoqian Wang. Binarized spectral compressive imaging.Advances in Neural Information Processing Systems, 36, 2024

  2. [10]

    Remote sensing image segmentation using vision mamba and multi-scale multi-frequency feature fusion.arXiv preprint arXiv:2410.05624, 2024

    Yice Cao, Chenchen Liu, Zhenhua Wu, Wenxin Yao, Liu Xiong, Jie Chen, and Zhixiang Huang. Remote sensing image segmentation using vision mamba and multi-scale multi-frequency feature fusion.arXiv preprint arXiv:2410.05624, 2024

  3. [11]

    Prior image guided snapshot compressive spectral imaging

    Yurong Chen, Yaonan Wang, and Hui Zhang. Prior image guided snapshot compressive spectral imaging. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(9):11096–11107, 2023

  4. [12]

    High-quality hyperspectral reconstruction using a spectral prior

    Inchang Choi, MH Kim, D Gutierrez, DS Jeon, and G Nam. High-quality hyperspectral reconstruction using a spectral prior. Technical report, 2017

  5. [13]

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

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

  6. [14]

    Exploiting spectral-spatial correlation for coded hyperspectral image restoration

    Ying Fu, Yinqiang Zheng, Imari Sato, and Yoichi Sato. Exploiting spectral-spatial correlation for coded hyperspectral image restoration. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 3727–3736, 2016

  7. [15]

    Enhanced facial recognition for thermal imagery using polarimetric imaging.Optics letters, 39(13):3857–3859, 2014

    Kristan P Gurton, Alex J Yuffa, and Gorden W Videen. Enhanced facial recognition for thermal imagery using polarimetric imaging.Optics letters, 39(13):3857–3859, 2014

  8. [16]

    Non-local meets global: An iterative paradigm for hyperspectral image restoration.IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(4):2089–2107, 2020

    Wei He, Quanming Yao, Chao Li, Naoto Yokoya, Qibin Zhao, Hongyan Zhang, and Liangpei Zhang. Non-local meets global: An iterative paradigm for hyperspectral image restoration.IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(4):2089–2107, 2020

  9. [17]

    Fast hyperspectral image recovery of dual-camera compressive hyperspectral imaging via non-iterative subspace-based fusion.IEEE Transactions on Image Processing, 30:7170–7183, 2021

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

  10. [18]

    Hdnet: High-resolution dual-domain learning for spectral compressive imaging

    Xiaowan Hu, Yuanhao Cai, Jing Lin, Haoqian Wang, Xin Yuan, Yulun Zhang, Radu Timofte, and Luc Van Gool. Hdnet: High-resolution dual-domain learning for spectral compressive imaging. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1754...

  11. [19]

    Deep gaussian scale mixture prior for spectral compressive imaging

    Tao Huang, Weisheng Dong, Xin Yuan, Jinjian Wu, and Guangming 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. 10

  12. [20]

    Incropera, DP

    F.P. Incropera, DP. DeWitt, T.L. Bergman, and AdrienneS. Lavine. Principles of heat and mass transfer. Jun 2018

  13. [21]

    Nirmal 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

  14. [22]

    Modern trends in hyperspectral image analysis: A review.Ieee Access, 6:14118–14129, 2018

    Muhammad Jaleed Khan, Hamid Saeed Khan, Adeel Yousaf, Khurram Khurshid, and Asad Abbas. Modern trends in hyperspectral image analysis: A review.Ieee Access, 6:14118–14129, 2018

  15. [23]

    Multiframe image estimation for coded aperture snapshot spectral imagers.Applied optics, 49(36):6824–6833, 2010

    David Kittle, Kerkil Choi, Ashwin Wagadarikar, and David J Brady. Multiframe image estimation for coded aperture snapshot spectral imagers.Applied optics, 49(36):6824–6833, 2010

  16. [24]

    Pixel adaptive deep unfolding transformer for hyperspectral image reconstruction

    Miaoyu Li, Ying Fu, Ji Liu, and Yulun Zhang. Pixel adaptive deep unfolding transformer for hyperspectral image reconstruction. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 12959–12968, 2023

  17. [25]

    Target detection with unconstrained linear mixture model and hierarchical denoising autoencoder in hyperspectral imagery.IEEE Transactions on Image Processing, 31:1418–1432, 2022

    Yunsong Li, Yanzi Shi, Keyan Wang, Bobo Xi, Jiaojiao Li, and Paolo Gamba. Target detection with unconstrained linear mixture model and hierarchical denoising autoencoder in hyperspectral imagery.IEEE Transactions on Image Processing, 31:1418–1432, 2022

  18. [26]

    Spatial-spectral encoded compressive hyperspectral imaging.ACM Transactions on Graphics (TOG), 33(6):1–11, 2014

    Xing Lin, Yebin Liu, Jiamin Wu, and Qionghai Dai. Spatial-spectral encoded compressive hyperspectral imaging.ACM Transactions on Graphics (TOG), 33(6):1–11, 2014

  19. [27]

    Flexible ftir spectral imaging enhancement for industrial robot infrared vision sensing.IEEE Transactions on Industrial Informatics, 16(1):544–554, 2019

    Tingting Liu, Hai Liu, You-Fu Li, Zengzhao Chen, Zhaoli Zhang, and Sannyuya Liu. Flexible ftir spectral imaging enhancement for industrial robot infrared vision sensing.IEEE Transactions on Industrial Informatics, 16(1):544–554, 2019

  20. [28]

    Rank minimization for snapshot compressive imaging.IEEE transactions on pattern analysis and machine intelligence, 41(12):2990–3006, 2018

    Yang Liu, Xin Yuan, Jinli Suo, David J Brady, and Qionghai Dai. Rank minimization for snapshot compressive imaging.IEEE transactions on pattern analysis and machine intelligence, 41(12):2990–3006, 2018

  21. [29]

    Vmamba: Visual state space model.Advances in neural information processing systems, 37:103031–103063, 2025

    Yue Liu, Yunjie Tian, Yuzhong Zhao, Hongtian Yu, Lingxi Xie, Yaowei Wang, Qixiang Ye, Jianbin Jiao, and Yunfan Liu. Vmamba: Visual state space model.Advances in neural information processing systems, 37:103031–103063, 2025

  22. [30]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. InProceedings of the IEEE/CVF international conference on computer vision, pages 10012–10022, 2021

  23. [31]

    Coded aperture compressive temporal imaging.Optics express, 21(9):10526–10545, 2013

    Patrick Llull, Xuejun Liao, Xin Yuan, Jianbo Yang, David Kittle, Lawrence Carin, Guillermo Sapiro, and David J Brady. Coded aperture compressive temporal imaging.Optics express, 21(9):10526–10545, 2013

  24. [32]

    End-to-end low cost compressive spectral imaging with spatial- spectral self-attention

    Ziyi Meng, Jiawei Ma, and Xin Yuan. End-to-end low cost compressive spectral imaging with spatial- spectral self-attention. InEuropean conference on computer vision, pages 187–204. Springer, 2020

  25. [33]

    Snapshot multispectral endomi- croscopy.Optics Letters, 45(14):3897–3900, 2020

    Ziyi Meng, Mu Qiao, Jiawei Ma, Zhenming Yu, Kun Xu, and Xin Yuan. Snapshot multispectral endomi- croscopy.Optics Letters, 45(14):3897–3900, 2020

  26. [34]

    Hi-mamba: Hierarchical mamba for efficient image super-resolution.arXiv preprint arXiv:2410.10140, 2024

    Junbo Qiao, Jincheng Liao, Wei Li, Yulun 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

  27. [35]

    Remotedet-mamba: A hybrid mamba-cnn network for multi-modal object detection in remote sensing images.arXiv preprint arXiv:2410.13532, 2024

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

  28. [36]

    Mambacsr: Dual-interleaved scanning for compressed image super-resolution with ssms.arXiv preprint arXiv:2408.11758, 2024

    Yulin Ren, Xin Li, Mengxi Guo, Bingchen Li, Shijie Zhao, and Zhibo Chen. Mambacsr: Dual-interleaved scanning for compressed image super-resolution with ssms.arXiv preprint arXiv:2408.11758, 2024

  29. [37]

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

    Michal Shimoni, Rob Haelterman, and Christiaan 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

  30. [38]

    Aerial vehicle tracking by adaptive fusion of hyperspectral likelihood maps

    Burak Uzkent, Aneesh Rangnekar, and Matthew 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

  31. [39]

    Dual-camera design for coded aperture snapshot spectral imaging.Applied optics, 54(4):848–858, 2015

    Lizhi Wang, Zhiwei Xiong, Dahua Gao, Guangming Shi, and Feng Wu. Dual-camera design for coded aperture snapshot spectral imaging.Applied optics, 54(4):848–858, 2015. 11

  32. [40]

    High-speed hyperspectral video acquisition by combining nyquist and compressive sampling.IEEE transactions on pattern analysis and machine intelligence, 41(4):857–870, 2018

    Lizhi Wang, Zhiwei Xiong, Hua Huang, Guangming Shi, Feng Wu, and Wenjun Zeng. High-speed hyperspectral video acquisition by combining nyquist and compressive sampling.IEEE transactions on pattern analysis and machine intelligence, 41(4):857–870, 2018

  33. [41]

    Adaptive nonlocal sparse representation for dual-camera compressive hyperspectral imaging.IEEE transactions on pattern analysis and machine intelligence, 39(10):2104–2111, 2016

    Lizhi Wang, Zhiwei Xiong, Guangming Shi, Feng Wu, and Wenjun Zeng. Adaptive nonlocal sparse representation for dual-camera compressive hyperspectral imaging.IEEE transactions on pattern analysis and machine intelligence, 39(10):2104–2111, 2016

  34. [42]

    Simultaneous depth and spectral imaging with a cross-modal stereo system.IEEE Transactions on Circuits and Systems for Video Technology, 28(3):812–817, 2016

    Lizhi Wang, Zhiwei Xiong, Guangming Shi, Wenjun Zeng, and Feng Wu. Simultaneous depth and spectral imaging with a cross-modal stereo system.IEEE Transactions on Circuits and Systems for Video Technology, 28(3):812–817, 2016

  35. [43]

    In2set: Intra-inter similarity exploiting transformer for dual-camera compressive hyperspectral imaging

    Xin Wang, Lizhi Wang, Xiangtian Ma, Maoqing Zhang, Lin Zhu, and Hua Huang. In2set: Intra-inter similarity exploiting transformer for dual-camera compressive hyperspectral imaging. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages...

  36. [44]

    Dual camera snapshot hyperspectral imaging system via physics-informed learning.Optics and Lasers in Engineering, 154:107023, 2022

    Hui Xie, Zhuang Zhao, Jing Han, Yi Zhang, Lianfa Bai, and Jun Lu. Dual camera snapshot hyperspectral imaging system via physics-informed learning.Optics and Lasers in Engineering, 154:107023, 2022

  37. [45]

    Compressive sensing by learning a gaussian mixture model from measurements.IEEE Transactions on Image Processing, 24(1):106–119, 2014

    Jianbo Yang, Xuejun Liao, Xin Yuan, Patrick Llull, David J Brady, Guillermo Sapiro, and Lawrence Carin. Compressive sensing by learning a gaussian mixture model from measurements.IEEE Transactions on Image Processing, 24(1):106–119, 2014

  38. [46]

    Specat: Spatial-spectral cumulative-attention transformer for high-resolution hyperspectral image reconstruction

    Zhiyang Yao, Shuyang Liu, Xiaoyun Yuan, and Lu Fang. Specat: Spatial-spectral cumulative-attention transformer for high-resolution hyperspectral image reconstruction. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 25368–25377, 2024

  39. [47]

    Generalized assorted pixel camera: postcapture control of resolution, dynamic range, and spectrum.IEEE transactions on image processing, 19(9):2241–2253, 2010

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

  40. [48]

    Hyperspectral image denoising employing a spatial–spectral deep residual convolutional neural network.IEEE Transactions on Geoscience and Remote Sensing, 57(2):1205–1218, 2018

    Qiangqiang Yuan, Qiang Zhang, Jie Li, Huanfeng Shen, and Liangpei Zhang. Hyperspectral image denoising employing a spatial–spectral deep residual convolutional neural network.IEEE Transactions on Geoscience and Remote Sensing, 57(2):1205–1218, 2018

  41. [49]

    Generalized alternating projection based total variation minimization for compressive sensing

    Xin Yuan. Generalized alternating projection based total variation minimization for compressive sensing. In2016 IEEE International conference on image processing (ICIP), pages 2539–2543. IEEE, 2016

  42. [50]

    Hyperspectral image superresolution by transfer learning

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

  43. [51]

    Samba: Semantic segmentation of remotely sensed images with state space model.Heliyon, 10(19), 2024

    Qinfeng Zhu, Yuanzhi Cai, Yuan Fang, Yihan Yang, Cheng Chen, Lei Fan, and Anh Nguyen. Samba: Semantic segmentation of remotely sensed images with state space model.Heliyon, 10(19), 2024. 12

Pith tools

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