Pith. sign in

REVIEW 4 major objections 3 minor 67 references

PHASE: Physiology-Aware Hyperspectral Reconstruction via Object-to-Human Domain Adaptation

T0 review · 4 major / 3 minor · reviewed 2026-08-03 · deepseek-v4-flash

Pith's one-line read SpectralAdapt reconstructs human hyperspectral images from RGB with only 1.5% labeled target data, nearly matching fully supervised performance.

desk verdict SpectralAdapt's core idea is reasonable, but its own Table I defeats the 'highest SAM' claim and the SDM module is never specified for the unlabeled RGB inputs it masks. read the letter →

arxiv 2511.13020 v2 pith:25OL7NDZ submitted 2025-11-17 cs.CV cs.AI

classification cs.CVcs.AI
keywords semi-superviseddomainadaptationhyperspectralreconstructionmeanteacherspectraldensitymaskingendmemberalignmentRGB-to-HSIhuman-centeredimagingmedicalsegmentation
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 argues that semi-supervised domain adaptation, previously used for classification, can be repurposed for hyperspectral reconstruction of human tissue. It proposes SpectralAdapt, a mean-teacher framework that transfers knowledge from large object-scene datasets (NTIRE2020/2022) to the small Hyper-Skin dataset. With only three labeled target images (1.5%), the method reports SSIM/SAM/PSNR near fully supervised levels and outperforms six SSDA baselines. If true, this would make clinical hyperspectral imaging practical without large human datasets.

What carries the argument

Mean Teacher paradigm with EMA-updated teacher; Spectral Density Masking (SDM), which derives per-channel masking ratios from a Spectral Angle Mapper deviation measure between a hyperspectral cube and a spatially averaged perturbation; and Spectral Endmember Representation Alignment (SERA), which initializes an endmember bank via the Automated Target Generation Process on labeled spectra and updates it with momentum, enforcing a max-cosine alignment loss on globally pooled predictions.

What would settle it

Take the Hyper-Skin test set and compute SDM masking ratios three ways: from the fixed 3 labeled target cubes, from the fixed source cubes, and per-image from teacher predictions on each unlabeled target RGB. If the performance differences are negligible, the per-sample adaptivity claim collapses, and SDM reduces to a fixed masking scheme.

Watch

Extended reading notes

Core claim

SpectralAdapt's central claim is that a student–teacher consistency framework, augmented with spectral priors, closes the domain shift between object-centric and human-centric hyperspectral reconstruction. Two modules carry the argument: Spectral Density Masking (SDM) adaptively masks RGB channels according to their spectral complexity, and Spectral Endmember Representation Alignment (SERA) pulls global-pooled predictions toward a momentum-updated bank of ATGP-derived endmembers. On NTIRE2020→Hyper-Skin and NTIRE2022→Hyper-Skin, the method reports up to +2.20 SSIM and −3.06 SAM over baselines with 1.5% labeled supervision, and its reconstructed HSI improves downstream medical segmentation ov

Load-bearing premise

The Spectral Density Masking ratios are computed from a hyperspectral cube, yet the method applies them to unlabeled RGB images for which no hyperspectral cube exists—the paper never says where that cube comes from, so the claimed per-sample adaptivity may in fact depend on a fixed prior.

Editorial extensions

If this is right

  • If correct, RGB-to-HSI reconstruction for medical use becomes deployable with a handful of labeled patient images, bypassing costly bulk-hyperspectral acquisition.
  • The success would open a general recipe: use physical priors (spectral densities, endmembers) to adapt dense regression models across domains, not just classifiers.
  • Reconstructed HSI approaching raw-HSI segmentation accuracy would make hyperspectral-based diagnosis accessible from ordinary camera hardware.
  • The near-match with fully supervised training suggests label scarcity is not the binding constraint in human HSI reconstruction; domain-shift engineering is.

Reading between the lines

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

  • The SDM density computation is unspecified for unlabeled RGB inputs—the paper defines D_b on a hyperspectral cube but never states whether the ratios come from labeled target HSI, teacher predictions, source HSI, or a fixed prior; if fixed, the claimed per-sample adaptivity is not actually per-sample.
  • SERA's endmember bank may converge toward a small set of dominant skin melanin/hemoglobin spectra, which would explain its power on facial skin but could limit transfer to organs with diverse chromophores; the HeiPorSPECTRAL results are a first test of that limit.
  • One could test the framework's core hypothesis directly by ablating SDM with masking ratios computed from source vs target labeled cubes; identical results would indicate the 'adaptivity' is decorative.
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 / 3 minor

Summary. The full text of the manuscript presents SpectralAdapt, a semi-supervised domain adaptation (SSDA) framework for hyperspectral image reconstruction from RGB. It builds on a Mean Teacher structure and introduces Spectral Density Masking (SDM) and Spectral Endmember Representation Alignment (SERA). The method is evaluated by transferring NTIRE2020 or NTIRE2022 to the human-centric Hyper-Skin dataset at 0.5% and 1.5% labeled target samples, with additional downstream segmentation experiments on Choledoch and HeiPorSPECTRAL. The paper claims to be the first SSDA method for hyperspectral reconstruction and to consistently outperform state-of-the-art SSDA approaches. However, the manuscript as provided has serious internal inconsistencies: the title/abstract describe a different method (PHASE), the reported SAM numbers contradict the headline claim of best performance, and the SDM mechanism is underspecified for RGB inputs.

Significance. If the claims were correct, the work would be significant: it addresses a practical low-label regime for human-centered hyperspectral reconstruction, proposes physically motivated spectral priors, and reports extensive ablations and downstream validation. The paper also attempts to adapt several classification-oriented SSDA baselines to a regression setting, which is useful for the community. However, the current version cannot be used as a reliable basis for these conclusions because the central comparison claim is contradicted by the paper's own Table I and because a key component (SDM) is not specified in a way that is reproducible for RGB inputs. The significance of the approach therefore remains to be demonstrated in a coherent, corrected manuscript.

major comments (4)
  1. [Section V-D, Table I] The text states that 'our method achieves the highest SSIM, SAM, and PSNR scores on both the NTIRE2020 and NTIRE2022 datasets.' This is directly contradicted by Table I for SAM at the emphasized 3-labeled (1.5%) setting: on NTIRE2020→Hyper-Skin, SpectralAdapt reports SAM 17.11±4.56, while IDM-SDA reports 14.97±3.17 and Unmix-SDA 15.73±3.18; on NTIRE2022→Hyper-Skin, SpectralAdapt reports 17.76±2.07 while IDM-SDA reports 14.88±3.50. Lower SAM is better, so SpectralAdapt is not the best on SAM in either transfer. The accompanying claim that with 3 labels the method 'nearly matches' fully supervised performance is also overstated: SAM is 17.11/17.76 vs. 14.00 fully supervised, with non-trivial SSIM and PSNR gaps. The central comparison claim must be corrected or supported with new results.
  2. [Section IV-B, Eqs. (2)-(5), Fig. 2] The spectral density D_b is defined for a hyperspectral cube X∈R^{H×W×C} by perturbing channels in region b, but SDM is applied to unlabeled target RGB images, which have no hyperspectral cube. The paper never states whether D_b is computed from labeled target HSI, source HSI, teacher predictions, or some RGB approximation. This is load-bearing for the SDM contribution and for the ablation results in Table V: if D_b is a fixed constant derived from labeled data, the claimed per-sample adaptivity is false; if D_b is computed from teacher predictions, the mask depends on the very model being trained. Please specify the exact source of D_b and whether it is per-sample or global, and justify the chosen value.
  3. [Title/abstract vs. full text] The manuscript is identified as 'PHASE: Physiology-Aware Hyperspectral Reconstruction via Object-to-Human Domain Adaptation' with an abstract describing Physiological Channel Reinterpretation and Physiologically Constrained Alignment. The full text, however, is 'SpectralAdapt: Semi-Supervised Domain Adaptation with Spectral Priors for Human-Centered Hyperspectral Image Reconstruction' and contains none of the PHASE contributions. No passage reconciles these two identities. A reader cannot determine which method is actually being proposed, and the claimed PHASE contribution is not present in the full text. This fundamental coherence defect must be resolved before further review.
  4. [Table I, Source-Only row] The Source-Only row reports identical SSIM/SAM/PSNR values (59.87, 67.57, 17.86) for both NTIRE2020→Hyper-Skin and NTIRE2022→Hyper-Skin. Since these are two different source datasets, identical values to two decimal places are implausible unless the same pretrained source-only model is used for both, which is not stated. Please clarify whether this is a data-reporting error or whether the same source-only checkpoint was used in both protocols, and correct the table accordingly.
minor comments (3)
  1. [Table I] Several table entries run together without spacing, e.g. '20.76±1.7622.82±1.98' and '19.32±1.4522.04±1.93'. Please fix the formatting for readability.
  2. [Section IV-C, Eq. (14)-(15)] The assignment rule a_i = argmax_j (z_i^T e_j^{t-1}) is used to define B_t^k before it is actually stated. Move the definition before Equation (14) or restructure the text so the notation is introduced in order.
  3. [Section IV-B] The symbol C is used both for the number of spectral channels in the HSI cube and for the constructed channel-index sets I_b (B, G, R). Use distinct notation to avoid confusion, especially since Eq. (2) seems to refer to spectral channels while the RGB masking operates on three input channels.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the central results are empirical benchmark claims, not derived quantities; SDM/SERA are inductive biases with a reproducibility gap rather than a circular reduction.

full rationale

This is an empirical method paper, and its central claims are benchmark results (SSIM/SAM/PSNR on NTIRE-to-Hyper-Skin transfers) rather than quantities derived from first principles. The loss equations (L_sup, L_con, L_SERA, L_total) combine supervised reconstruction, consistency regularization, and endmember-guided alignment; none of them is equivalent by construction to the reported metrics. SERA initializes an endmember bank from labeled source/target pixel spectra and then aligns predictions to it via a contrastive-style loss (Eqs. 10-15); this is a semi-supervised inductive bias, not a fitted parameter renamed as a prediction. SDM's spectral density D_b is defined on a hyperspectral cube (Eqs. 2-5) but applied to RGB inputs, and the paper never states where D_b comes from for unlabeled RGB targets; that is a genuine specification/reproducibility gap, but it is not a circular step because no final metric is defined in terms of D_b, and the masking ratios are not claimed to be the output being evaluated. The only self-citation is ref. [54], which includes co-author J. Chen; it is used to motivate masking-based learning and is not load-bearing for the main contribution. The claim in Section V-D that the method 'achieves the highest SSIM, SAM, and PSNR scores' is contradicted by Table I on SAM at the 1.5% label ratio (e.g., IDM-SDA reports 14.97 vs. 17.11 for SpectralAdapt on NTIRE2020→Hyper-Skin), but that is a factual/correctness issue, not circularity. Overall, the paper does not derive its predictions from its inputs in a way that would warrant a circularity score above 1.

Assumptions & free parameters 6 free parameters · 3 assumptions · 0 invented entities

No new physical entities are posited. The free parameters are training hyperparameters and unstated SDM/endmember settings; the central axioms are the covariate-shift assumption, the transferability of HSI-computed spectral density to RGB masking, and the claim that ATGP endmembers carry physiological meaning.

free parameters (6)
  • lambda_sup = 0.4
    Weight balancing L1 and SSIM in the supervised loss; chosen from sensitivity analysis in Fig. 7 on the target benchmark.
  • lambda_un = 0.3
    Weight balancing consistency loss and SERA loss; selected from sensitivity analysis in Fig. 7.
  • SDM masking rate r = 70% (optimal in Table VI)
    The masking ratio is tuned via Table VI on the same target datasets; r_min and r_max are not reported.
  • endmember bank size K = not reported
    ATGP endmember count K is introduced in Sec. IV-C but its value is never stated.
  • mask block size s = not reported
    Channel-wise block masking uses blocks of size s x s, but s is not specified.
  • teacher EMA momentum m_ema and endmember momentum m_end = 0.99 and 0.9
    Two momentum coefficients chosen by hand in Sec. V-C; standard but still tunable hyperparameters.
assumptions (3)
  • domain assumption Source and target share similar conditional distributions P_S(y|x) ≈ P_T(y|x) (covariate shift).
    Invoked in Sec. I and Fig. 1 to justify SSDA. If false, aligning source and target features can hurt reconstruction.
  • domain assumption Spectral density D_b computed from a hyperspectral cube can be transferred to RGB masking decisions.
    Eq. (5) defines D_b from ground-truth HSI, but unlabeled target data has no HSI; the paper does not specify the substitute input.
  • domain assumption ATGP-selected endmembers are physically meaningful, domain-invariant anchors for human skin spectra.
    Sec. IV-C assumes extreme pixels selected by projection residual are the right spectral anchors; no physiological validation is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PHASE: Physiology-Aware Hyperspectral Reconstruction via Object-to-Human Domain Adaptation." pith.science (2026). https://pith.science/paper/25OL7NDZ

@misc{pith2026251113020,
  author       = {Pith},
  title        = {Pith review of: PHASE: Physiology-Aware Hyperspectral Reconstruction via Object-to-Human Domain Adaptation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/25OL7NDZ}},
  note         = {Machine review of arXiv:2511.13020}
}
read the original abstract

Although hyperspectral imaging offers unparalleled non-invasive physiological insight, its bulky hardware, slow acquisition, and regulatory burden severely limit its clinical availability. A natural workaround is to reconstruct hyperspectral information from ubiquitous RGB or CASSI measurements. However, existing paradigms, developed for object-centric scenes, rely on reflectance-based feature alignment, assuming that spectral similarity preserves semantic meaning. This assumption breaks down in physiological imaging, where visually similar RGB responses may arise from distinct and entangled physiological states. This mismatch motivates a shift from reflectance alignment to physiology-aware representation learning, grounded in shared light-matter interaction principles -- a shift that introduces fundamental challenges from cross-channel semantic shifts (C1) and irreversible information loss in RGB-based acquisition (C2). We therefore design PHASE, a physiology-aware hyperspectral reconstruction paradigm that fundamentally redefines object-to-human transfer by disentangling cross-channel physiological semantics via Physiological Channel Reinterpretation and restricting reconstruction to physiologically plausible solutions through Physiologically Constrained Alignment. Under two source-to-target transfer protocols, PHASE consistently outperforms state-of-the-art methods by up to +2.20 SSIM and -3.06 in SAM with merely 1.5% labeled supervision.

Figures

Figures reproduced from arXiv: 2511.13020 by the authors.

Figure 1
Figure 1. Illustration of the domain adaptation scenario in HSI reconstruction under the SSDA paradigm. (a) Illustration of distribution shift between the source [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of our proposed method. ζ(weak) and ζ ′ (strong) denote two stochastic augmentations applied to the same input before the student and teacher, θ, θ ′ represents student and teacher model parameters respectively, where student and teacher model share the same architecture and θ ′ is updated through the Exponential Moving Average (EMA) of θ. Spectral density masking (SDM) adaptively masks RGB regions based on… view at source ↗
Figure 3
Figure 3. Spectral Density Masking Illustration. We visualize three sampled regions from an RGB image (top left) and their corresponding reflectance spectra from the ground truth hyperspectral image (bottom left). The spectral bands are across the 400–700 nm range and color-coded into blue (400–500 nm), green (510–580 nm), and red (600–700 nm). Notably, higher angular deviations in the red region indicate greater spectral com… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Visualization of the downstream evaluation pipeline, which comprises [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Qualitative comparison of segmentation results on the Choledoch and HeiPorSPECTRAL datasets using three input modalities: RGB, reconstructed [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Error map visual comparisons of predictions using different modules [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

67 extracted references

  1. [1]

    Modern trends in hyperspectral image analysis: A review,

    M. J. Khan, H. S. Khan, A. Yousaf, K. Khurshid, and A. Abbas, “Modern trends in hyperspectral image analysis: A review,”Ieee Access, vol. 6, pp. 14 118–14 129, 2018. 12

  2. [2]

    Hyperspectral remote sensing in litho- logical mapping, mineral exploration, and environmental geology: an updated review,

    S. Peyghambari and Y . Zhang, “Hyperspectral remote sensing in litho- logical mapping, mineral exploration, and environmental geology: an updated review,”Journal of Applied Remote Sensing, vol. 15, no. 3, pp. 031 501–031 501, 2021

  3. [3]

    Current advances in hyperspectral remote sensing in urban planning,

    A. Nisha and A. Anitha, “Current advances in hyperspectral remote sensing in urban planning,” in2022 Third International Conference on intelligent computing instrumentation and control technologies (ICI- CICT). IEEE, 2022, pp. 94–98

  4. [4]

    Intra- operative guidance using hyperspectral imaging: a review for surgeons,

    M. Barberio, S. Benedicenti, M. Pizzicannella, E. Felli, T. Collins, B. Jansen-Winkeln, J. Marescaux, M. G. Viola, and M. Diana, “Intra- operative guidance using hyperspectral imaging: a review for surgeons,” Diagnostics, vol. 11, no. 11, p. 2066, 2021

  5. [5]

    Non-invasive skin cancer diagnosis using hyperspectral imaging for in-situ clinical support,

    R. Leon, B. Martinez-Vega, H. Fabelo, S. Ortega, V . Melian, I. Casta ˜no, G. Carretero, P. Almeida, A. Garcia, E. Quevedoet al., “Non-invasive skin cancer diagnosis using hyperspectral imaging for in-situ clinical support,”Journal of clinical medicine, vol. 9, no. 6, p. 1662, 2020

  6. [6]

    Hyperspectral imaging for diagnosis and detection of ex-vivo breast cancer,

    I. H. Aboughaleb, M. H. Aref, and Y . H. El-Sharkawy, “Hyperspectral imaging for diagnosis and detection of ex-vivo breast cancer,”Photodi- agnosis and Photodynamic Therapy, vol. 31, p. 101922, 2020

  7. [7]

    Detection of blood in fish muscle by constrained spectral unmixing of hyperspectral images,

    M. H. Skjelvareid, K. Heia, S. H. Olsen, and S. K. Stormo, “Detection of blood in fish muscle by constrained spectral unmixing of hyperspectral images,”Journal of Food Engineering, vol. 212, pp. 252–261, 2017

  8. [8]

    Hyperspectral imaging as a possible tool for visualization of changes in hemoglobin oxygenation in patients with deficient hemodynamics–proof of concept,

    C. Sicher, R. Rutkowski, S. Lutze, S. von Podewils, T. Wild, M. Kretch- ing, and G. Daeschlein, “Hyperspectral imaging as a possible tool for visualization of changes in hemoglobin oxygenation in patients with deficient hemodynamics–proof of concept,”Biomedical Engineer- ing/Biomedizinische Technik, vol. 63, no. 5, pp. 609–616, 2018

Show all 67 references
  1. [9]

    Ntire 2022 spectral recovery challenge and data set,

    B. Arad, R. Timofte, R. Yahel, N. Morag, A. Bernat, Y . Cai, J. Lin, Z. Lin, H. Wang, Y . Zhanget al., “Ntire 2022 spectral recovery challenge and data set,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 863–881

  2. [10]

    Hyper-skin: a hyperspectral dataset for reconstructing facial skin-spectra from rgb images,

    P. C. Ng, Z. Chi, Y . Verdie, J. Lu, and K. N. Plataniotis, “Hyper-skin: a hyperspectral dataset for reconstructing facial skin-spectra from rgb images,”Advances in Neural Information Processing Systems, vol. 36, 2024

  3. [11]

    Essaformer: Efficient transformer for hyperspectral image super- resolution,

    M. Zhang, C. Zhang, Q. Zhang, J. Guo, X. Gao, and J. Zhang, “Essaformer: Efficient transformer for hyperspectral image super- resolution,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 23 073–23 084

  4. [12]

    Reconstruction of cassi- raman images with machine-learning,

    A. Brorsson, M. Nordberg, and D. Gustafsson, “Reconstruction of cassi- raman images with machine-learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 4388–4395

  5. [13]

    Re- constructing and segmenting hyperspectral images from compressed measurements,

    Q. Zhang, R. Plemmons, D. Kittle, D. Brady, and S. Prasad, “Re- constructing and segmenting hyperspectral images from compressed measurements,” in2011 3rd Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS). IEEE, 2011, pp. 1–4

  6. [14]

    Hyperspectral image reconstruction by deep convolutional neural network for classification,

    Y . Li, W. Xie, and H. Li, “Hyperspectral image reconstruction by deep convolutional neural network for classification,”Pattern Recognition, vol. 63, pp. 371–383, 2017

  7. [15]

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

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

  8. [16]

    Hsi-denet: Hyperspectral image restoration via convolutional neural network,

    Y . Chang, L. Yan, H. Fang, S. Zhong, and W. Liao, “Hsi-denet: Hyperspectral image restoration via convolutional neural network,”IEEE Transactions on Geoscience and Remote Sensing, vol. 57, no. 2, pp. 667–682, 2018

  9. [17]

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

    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, 2022, pp. 17 502–17 511

  10. [18]

    Coarse-to-fine sparse transformer for hyperspectral image recon- struction,

    ——, “Coarse-to-fine sparse transformer for hyperspectral image recon- struction,” inEuropean conference on computer vision. Springer, 2022, pp. 686–704

  11. [19]

    Statistics of real-world hyperspectral images,

    A. Chakrabarti and T. Zickler, “Statistics of real-world hyperspectral images,” inCVPR 2011. IEEE, 2011, pp. 193–200

  12. [20]

    Generalized assorted pixel camera: postcapture control of resolution, dynamic range, and spectrum,

    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, vol. 19, no. 9, pp. 2241–2253, 2010

  13. [21]

    Hyperspec- tral pansharpening based on improved deep image prior and residual reconstruction,

    W. G. C. Bandara, J. M. J. Valanarasu, and V . M. Patel, “Hyperspec- tral pansharpening based on improved deep image prior and residual reconstruction,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–16, 2021

  14. [22]

    Cross- domain hyperspectral image classification based on bi-directional do- main adaptation,

    Y . Zhang, W. Li, W. Jia, M. Zhang, R. Tao, and S. Liang, “Cross- domain hyperspectral image classification based on bi-directional do- main adaptation,”IEEE Transactions on Circuits and Systems for Video Technology, 2025

  15. [23]

    Ssda3d: Semi-supervised domain adaptation for 3d object detection from point cloud,

    Y . Wang, J. Yin, W. Li, P. Frossard, R. Yang, and J. Shen, “Ssda3d: Semi-supervised domain adaptation for 3d object detection from point cloud,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 3, 2023, pp. 2707–2715

  16. [24]

    Cross-domain few-shot learning method based on fractional domain information for hyperspectral image multi-class change detection,

    S. Feng, J. Zhang, Y . Fan, X. Liu, C. Zhao, W. Li, and R. Tao, “Cross-domain few-shot learning method based on fractional domain information for hyperspectral image multi-class change detection,”IEEE Transactions on Circuits and Systems for Video Technology, 2025

  17. [25]

    Semi-supervised student-teacher learning for single image super-resolution,

    L. Wang and K.-J. Yoon, “Semi-supervised student-teacher learning for single image super-resolution,”Pattern Recognition, vol. 121, p. 108206, 2022

  18. [26]

    Semi-supervised heterogeneous domain adaptation: Theory and algorithms,

    Z. Fang, J. Lu, F. Liu, and G. Zhang, “Semi-supervised heterogeneous domain adaptation: Theory and algorithms,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 1, pp. 1087– 1105, 2022

  19. [27]

    A bayesian-inspired, deep learning-based, semi-supervised domain adap- tation technique for land cover mapping,

    B. Lucas, C. Pelletier, D. Schmidt, G. I. Webb, and F. Petitjean, “A bayesian-inspired, deep learning-based, semi-supervised domain adap- tation technique for land cover mapping,”Machine Learning, vol. 112, no. 6, pp. 1941–1973, 2023

  20. [28]

    Domain adaptation for medical image analysis: a survey,

    H. Guan and M. Liu, “Domain adaptation for medical image analysis: a survey,”IEEE Transactions on Biomedical Engineering, vol. 69, no. 3, pp. 1173–1185, 2021

  21. [29]

    A theory of learning from different domains,

    S. Ben-David, J. Blitzer, K. Crammer, A. Kulesza, F. Pereira, and J. W. Vaughan, “A theory of learning from different domains,”Machine learning, vol. 79, pp. 151–175, 2010

  22. [30]

    Mean teachers are better role mod- els: Weight-averaged consistency targets improve semi-supervised deep learning results,

    A. Tarvainen and H. Valpola, “Mean teachers are better role mod- els: Weight-averaged consistency targets improve semi-supervised deep learning results,”Advances in neural information processing systems, vol. 30, 2017

  23. [31]

    The spectral image processing system (sips)—interactive visualization and analysis of imaging spectrometer data,

    F. A. Kruse, A. Lefkoff, y. J. Boardman, K. Heidebrecht, A. Shapiro, P. Barloon, and A. Goetz, “The spectral image processing system (sips)—interactive visualization and analysis of imaging spectrometer data,”Remote sensing of environment, vol. 44, no. 2-3, pp. 145–163, 1993

  24. [32]

    A survey on computational spectral reconstruction methods from rgb to hyperspectral imaging,

    J. Zhang, R. Su, Q. Fu, W. Ren, F. Heide, and Y . Nie, “A survey on computational spectral reconstruction methods from rgb to hyperspectral imaging,”Scientific reports, vol. 12, no. 1, p. 11905, 2022

  25. [33]

    Hypermixnet: hyperspectral image recon- struction with deep mixed network from a snapshot measurement,

    K. Yorimoto and X.-H. Han, “Hypermixnet: hyperspectral image recon- struction with deep mixed network from a snapshot measurement,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 1184–1193

  26. [34]

    Pixel adaptive deep unfolding transformer for hyperspectral image reconstruction,

    M. Li, Y . Fu, J. Liu, and Y . Zhang, “Pixel adaptive deep unfolding transformer for hyperspectral image reconstruction,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 12 959–12 968

  27. [35]

    Reconstruction of hyperspectral data from rgb images with prior category information,

    L. Yan, X. Wang, M. Zhao, M. Kaloorazi, J. Chen, and S. Rahardja, “Reconstruction of hyperspectral data from rgb images with prior category information,”IEEE Transactions on Computational Imaging, vol. 6, pp. 1070–1081, 2020

  28. [36]

    Hsgan: Hy- perspectral reconstruction from rgb images with generative adversarial network,

    Y . Zhao, L.-M. Po, T. Lin, Q. Yan, W. Liu, and P. Xian, “Hsgan: Hy- perspectral reconstruction from rgb images with generative adversarial network,”IEEE Transactions on Neural Networks and Learning Systems, 2023

  29. [37]

    Hyperspectral compressive snapshot reconstruction via coupled low-rank subspace representation and self-supervised deep network,

    Y . Chen, W. Lai, W. He, X.-L. Zhao, and J. Zeng, “Hyperspectral compressive snapshot reconstruction via coupled low-rank subspace representation and self-supervised deep network,”IEEE Transactions on Image Processing, 2024

  30. [38]

    Low-rank tensor meets deep prior: Coupling model-driven and data-driven methods for hyperspectral image reconstruction,

    Y . Chen, F. Yuan, W. Lai, J. Zeng, W. He, and Q. Huang, “Low-rank tensor meets deep prior: Coupling model-driven and data-driven methods for hyperspectral image reconstruction,”IEEE Transactions on Circuits and Systems for Video Technology, 2025

  31. [39]

    Spectral-wise implicit neural representation for hyperspectral image reconstruction,

    H. Chen, W. Zhao, T. Xu, G. Shi, S. Zhou, P. Liu, and J. Li, “Spectral-wise implicit neural representation for hyperspectral image reconstruction,”IEEE Transactions on Circuits and Systems for Video Technology, 2023

  32. [40]

    Spatial-spectral structured sparse low-rank representation for hyper- spectral image super-resolution,

    J. Xue, Y .-Q. Zhao, Y . Bu, W. Liao, J. C.-W. Chan, and W. Philips, “Spatial-spectral structured sparse low-rank representation for hyper- spectral image super-resolution,”IEEE Transactions on Image Process- ing, vol. 30, pp. 3084–3097, 2021

  33. [41]

    Learning tensor low- rank prior for hyperspectral image reconstruction,

    S. Zhang, L. Wang, L. Zhang, and H. Huang, “Learning tensor low- rank prior for hyperspectral image reconstruction,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 12 006–12 015. 13

  34. [42]

    A survey on deep semi- supervised learning,

    X. Yang, Z. Song, I. King, and Z. Xu, “A survey on deep semi- supervised learning,”IEEE transactions on knowledge and data engi- neering, vol. 35, no. 9, pp. 8934–8954, 2022

  35. [43]

    A survey on semi-supervised learning,

    J. E. Van Engelen and H. H. Hoos, “A survey on semi-supervised learning,”Machine learning, vol. 109, no. 2, pp. 373–440, 2020

  36. [44]

    Pseudo-label-assisted subdomain adaptation for hyperspectral image classification,

    Z. Feng, S. Tong, S. Yang, X. Zhang, and L. Jiao, “Pseudo-label-assisted subdomain adaptation for hyperspectral image classification,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 34, no. 6, pp. 4729–4744, 2023

  37. [45]

    Adamatch: A unified approach to semi-supervised learning and domain adaptation,

    D. Berthelot, R. Roelofs, K. Sohn, N. Carlini, and A. Kurakin, “Adamatch: A unified approach to semi-supervised learning and domain adaptation,” inProceedings of the 10th International Conference on Learning Representations (ICLR), 2022

  38. [46]

    Multi-level consistency learning for semi-supervised domain adaptation,

    Z. Yan, Y . Wu, G. Li, Y . Qin, X. Han, and S. Cui, “Multi-level consistency learning for semi-supervised domain adaptation,” inPro- ceedings of the Thirty-First International Joint Conference on Artificial Intelligence (IJCAI). IJCAI, 2022, pp. 1530–1536

  39. [47]

    Context-aware mixup for domain adaptive semantic segmentation,

    Q. Zhou, Z. Feng, Q. Gu, J. Pang, G. Cheng, X. Lu, J. Shi, and L. Ma, “Context-aware mixup for domain adaptive semantic segmentation,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 33, no. 2, pp. 804–817, 2022

  40. [48]

    Inter-domain mixup for semi-supervised domain adaptation,

    J. Li, G. Li, and Y . Yu, “Inter-domain mixup for semi-supervised domain adaptation,”Pattern Recognition, vol. 146, p. 110023, 2024

  41. [49]

    Source-free semi-supervised domain adaptation via progressive mixup,

    N. Ma, H. Wang, Z. Zhang, S. Zhou, H. Chen, and J. Bu, “Source-free semi-supervised domain adaptation via progressive mixup,”Knowledge- Based Systems, vol. 262, p. 110208, 2023

  42. [50]

    Semi-supervised domain adaptation via prototype-based multi-level learning,

    X. Huang, C. Zhu, and W. Chen, “Semi-supervised domain adaptation via prototype-based multi-level learning,” inProceedings of the 32nd International Joint Conference on Artificial Intelligence (IJCAI). IJCAI, 2023, pp. 884–892

  43. [51]

    Unsupervised hyperspectral image domain adaptation through unmixing-based domain alignment,

    R. K. Baghbaderani, Y . Qu, and H. Qi, “Unsupervised hyperspectral image domain adaptation through unmixing-based domain alignment,” in IGARSS 2023-2023 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2023, pp. 5906–5909

  44. [52]

    Maskfu- sionnet: A dual-stream fusion model with masked pre-training mecha- nism for rppg measurement,

    Y . Zhang, J. Shi, J. Wang, Y . Zong, W. Zheng, and G. Zhao, “Maskfu- sionnet: A dual-stream fusion model with masked pre-training mecha- nism for rppg measurement,”IEEE Transactions on Circuits and Systems for Video Technology, 2024

  45. [53]

    A dct-based total jnd profile for spatiotemporal and foveated masking effects,

    S.-H. Bae and M. Kim, “A dct-based total jnd profile for spatiotemporal and foveated masking effects,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 27, no. 6, pp. 1196–1207, 2016

  46. [54]

    Multi-view learning with feature level fusion for cervical dysplasia diagnosis,

    T. Chen, X. Ma, X. Liu, W. Wang, R. Feng, J. Chen, C. Yuan, W. Lu, D. Z. Chen, and J. Wu, “Multi-view learning with feature level fusion for cervical dysplasia diagnosis,” inMedical Image Computing and Computer Assisted Intervention. Springer, 2019, pp. 329–338

  47. [55]

    Endmember variability in spectral mixture analysis: A review,

    B. Somers, G. P. Asner, L. Tits, and P. Coppin, “Endmember variability in spectral mixture analysis: A review,”Remote Sensing of Environment, vol. 115, no. 7, pp. 1603–1616, 2011

  48. [56]

    Impact of initialization on design of endmember extraction algorithms,

    A. Plaza and C.-I. Chang, “Impact of initialization on design of endmember extraction algorithms,”IEEE Transactions on Geoscience and Remote Sensing, vol. 44, no. 11, pp. 3397–3407, 2006

  49. [57]

    Image quality assessment: from error visibility to structural similarity,

    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, vol. 13, no. 4, pp. 600–612, 2004

  50. [58]

    Ntire 2020 challenge on spectral reconstruction from an rgb image,

    B. Arad, R. Timofte, O. Ben-Shahar, Y .-T. Lin, and G. D. Finlayson, “Ntire 2020 challenge on spectral reconstruction from an rgb image,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, 2020, pp. 446–447

  51. [59]

    A multidi- mensional choledoch database and benchmarks for cholangiocarcinoma diagnosis,

    Q. Zhang, Q. Li, G. Yu, L. Sun, M. Zhou, and J. Chu, “A multidi- mensional choledoch database and benchmarks for cholangiocarcinoma diagnosis,”IEEE access, vol. 7, pp. 149 414–149 421, 2019

  52. [60]

    Heiporspectral-the heidelberg porcine hyperspectral imaging dataset of 20 physiological organs,

    A. Studier-Fischer, S. Seidlitz, J. Sellner, M. Bressan, B. ¨Ozdemir, L. Ayala, J. Odenthal, S. Knoedler, K.-F. Kowalewski, C. M. Haney et al., “Heiporspectral-the heidelberg porcine hyperspectral imaging dataset of 20 physiological organs,”Scientific Data, vol. 10, no. 1, p. ...

  53. [61]

    Mst++: Multi-stage Spectral-wise Trans- former for Efficient Spectral Reconstruction,

    Y . Cai, J. Lin, Z. Linet al., “Mst++: Multi-stage Spectral-wise Trans- former for Efficient Spectral Reconstruction,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 745–755

  54. [62]

    Image quality metrics: Psnr vs. ssim,

    A. Hore and D. Ziou, “Image quality metrics: Psnr vs. ssim,” in2010 20th international conference on pattern recognition. IEEE, 2010, pp. 2366–2369

  55. [63]

    Fully convolutional networks for semantic segmentation,

    J. Long, E. Shelhamer, and T. Darrell, “Fully convolutional networks for semantic segmentation,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2015, pp. 3431–3440

  56. [64]

    The pascal visual object classes (voc) challenge,

    M. Everingham, L. Van Gool, C. K. Williams, J. Winn, and A. Zisser- man, “The pascal visual object classes (voc) challenge,”International journal of computer vision, vol. 88, pp. 303–338, 2010

  57. [65]

    U-net: Convolutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” inInternational Conference on Medical image computing and computer-assisted intervention. Springer, 2015, pp. 234–241

  58. [66]

    Not-so-supervised: a survey of semi-supervised, multi-instance, and transfer learning in medical image analysis,

    V . Cheplygina, M. De Bruijne, and J. P. Pluim, “Not-so-supervised: a survey of semi-supervised, multi-instance, and transfer learning in medical image analysis,”Medical image analysis, vol. 54, pp. 280–296, 2019

  59. [67]

    Learning with limited annotations: a survey on deep semi-supervised learning for medical image segmentation,

    R. Jiao, Y . Zhang, L. Ding, B. Xue, J. Zhang, R. Cai, and C. Jin, “Learning with limited annotations: a survey on deep semi-supervised learning for medical image segmentation,”Computers in Biology and Medicine, vol. 169, p. 107840, 2024

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.