pith. sign in

arxiv: 2605.21371 · v1 · pith:NZJ5WNMTnew · submitted 2026-05-20 · 💻 cs.CV

A Non-Reference Diffusion-Based Restoration Framework for Landsat 7 ETM+ SLC-off Imagery in Antarctica

Leyue Tang , Jonathan Louis Bamber , Gang Qiao , Yuanhang Kong This is my paper

Pith reviewed 2026-05-21 04:46 UTC · model grok-4.3

classification 💻 cs.CV
keywords Landsat 7SLC-offdiffusion modelimage restorationAntarcticagap fillingremote sensingnon-reference
0
0 comments X

The pith

A diffusion model trained only on Antarctic data can fill missing pixels in Landsat 7 SLC-off images without any reference imagery.

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

The paper presents DiffGF, a two-stage diffusion framework that restores gaps in Landsat 7 ETM+ SLC-off imagery over Antarctica without external reference data. It first runs a latent-space diffusion process to generate plausible content, then refines the output in pixel space. Training occurs on the authors' new SLCANT dataset of Antarctic scenes. Results show high-fidelity restorations that improve a downstream crevasse segmentation task, addressing the fact that 22 percent of pixels are missing after the 2003 SLC failure and that rapid polar surface changes make reference-based methods unreliable.

Core claim

DiffGF restores Antarctic SLC-off imagery with high fidelity through a non-reference diffusion-based framework that uses a latent-space diffusion process followed by pixel-space refinement, trained on the dedicated SLCANT dataset, and the restored images support improved crevasse segmentation in downstream applications.

What carries the argument

DiffGF, a two-stage non-reference diffusion model that performs gap-filling first in latent space then refines in pixel space by learning priors from Antarctic training data.

If this is right

  • Restored images become usable for multi-year Antarctic surface-change studies that previously excluded SLC-off periods.
  • Downstream tasks such as crevasse mapping achieve higher accuracy on the filled imagery than on the original gappy data.
  • The Landsat 7 archive in polar regions can be exploited more fully for historical analysis without waiting for matching reference scenes.
  • The approach removes the need to search for or align external reference imagery in environments with rapid surface evolution.

Where Pith is reading between the lines

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

  • The same non-reference strategy may reduce dependence on multi-temporal pairs in other remote-sensing domains that face data gaps and changing surfaces.
  • Training on region-specific data like SLCANT suggests that domain-adapted diffusion priors could be tested on SLC-off imagery from non-polar latitudes.
  • If the latent-space stage proves robust, the framework could be adapted to other optical sensors that experience similar line-dropout failures.

Load-bearing premise

The diffusion model trained on the SLCANT dataset learns priors that generalize to real-world SLC-off scenes with variable Antarctic surface conditions and can produce accurate reconstructions without any external reference data.

What would settle it

Apply the method to a test set of SLC-off images that also have complete reference coverage from another sensor or date, then measure whether segmentation or surface-feature accuracy drops below the level achieved by reference-based gap-filling methods.

Figures

Figures reproduced from arXiv: 2605.21371 by Gang Qiao, Jonathan Louis Bamber, Leyue Tang, Yuanhang Kong.

Figure 1
Figure 1. Figure 1: Workflow of the proposed DiffGF framework. (a) Overview of DiffGF framework, consisting of two stages: a diffusion process followed by a refinement stage using a mask-guided harmonization network for final output. (b) Illustration of the forward and reverse processes in the diffusion model. Intermediate results are decoded into pixel space for intuitive visual interpretation. (c) Structure of the proposed … view at source ↗
Figure 2
Figure 2. Figure 2: Architectural details of the PixelShuffle Block (a) and ResBlock (b) used in MGHNet. D. Training Strategy The training of the DiffGF framework is carried out in two separate stages: the denoising network in the diffusion process and the refinement MGHNet are trained independently. The denoising network is trained using the simulated SLC-off images, their corresponding ground truth images, and associated ma… view at source ↗
Figure 4
Figure 4. Figure 4: Restoration results of DiffGF on test images from the SLCANT dataset. From left to right: SLC-off input images, reconstructed results by DiffGF, and ground truth images (GT), and zoomed-in views of the red-boxed regions. (a) Acquired on February 16, 2019, over Shackleton Ice Shelf. (b)-(c) Acquired on January 20, 2020, over Amery Ice Shelf. (d) Acquired on December 29, 2020, over Larsen C Ice Shelf. (e) Ac… view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of DiffGF with and without MGHNet on test images from the SLCANT dataset. From left to right: SLC-off input images, reconstructed results using the diffusion model only, DiffGF results, and ground truth images. (a) Acquired on February 16, 2019, over Shackleton Ice Shelf. (b) Zoomed-in views of the red-boxed region in (a). (c) Acquired on January 7, 2021, over Brunt Ice Shelf. (d) Zoomed-in view… view at source ↗
Figure 6
Figure 6. Figure 6: Architectures of the residual block used in the ablation study for the MGHNet decoder. (a) The designed ResBlock. (b) Variant 1: ResBlock with Batch Normalization (BN). (c) Variant 2: Classical ResNet- style block following [56] [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Reconstruction and downstream crevasse segmentation results on the simulated SLC-off Larsen B Ice Shelf test image. (a) Reconstructed images generated by different methods. (b) Corresponding crevasse segmentation results [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Overall, crevasse structures, textures, and boundaries are more faithfully recovered by DiffGF, showing closer agreement with the ground-truth image than the comparative methods. To further visualize local reconstruction fidelity, pixel-wise absolute differences between the reconstructions and the ground-truth image are computed and normalized to [0,1], as shown in Figs. 8(c) and 9(c), illustrating that Di… view at source ↗
read the original abstract

Acquiring usable optical imagery in Antarctica is inherently challenging due to prolonged polar nights and frequent cloud cover. Landsat provides the longest and most continuous optical observations and constitutes one of the most important remote sensing data sources for Antarctic studies. However, the scan-line corrector (SLC) failure in 2003 resulted in approximately 22% missing pixels in Landsat 7 ETM+ SLC-off imagery, severely limiting its usability. Unlike many non-polar environments, Antarctic surfaces undergo rapid and substantial changes, which makes it difficult to obtain reliable reference imagery and reduces the applicability of conventional reference-based gap-filling methods. To address this challenge, we propose DiffGF, a non-reference diffusion-based framework for restoring Landsat 7 SLC-off imagery without requiring any external reference data. DiffGF adopts a two-stage design consisting of a latent-space diffusion process and a pixel-space refinement. A dedicated Antarctic dataset, SLCANT, is constructed for training and evaluation. Quantitative and qualitative results demonstrate that DiffGF restores Antarctic SLC-off imagery with high fidelity. Its practical value is further examined through a downstream crevasse segmentation application. The results suggest that DiffGF provides a useful approach for exploiting Landsat 7 SLC-off archives in Antarctica, enabling the extraction of valuable information from historical records and supporting related Antarctic studies.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The manuscript proposes DiffGF, a non-reference two-stage diffusion framework (latent-space diffusion followed by pixel-space refinement) for restoring Landsat 7 ETM+ SLC-off imagery in Antarctica. It introduces the SLCANT dataset for training and evaluation, claims high-fidelity restoration via quantitative and qualitative results, and demonstrates downstream utility through a crevasse segmentation application. The approach is positioned as addressing the limitations of reference-based methods in regions with rapid surface changes.

Significance. If the generalization claims hold, the work could enable broader use of the Landsat 7 SLC-off archive for Antarctic studies by providing a reference-free restoration method suited to polar conditions where conventional gap-filling is impractical. The construction of SLCANT and the focus on a downstream task add practical value for the remote-sensing community.

major comments (1)
  1. [Evaluation section] Evaluation section (likely §4 or §5): The quantitative fidelity metrics (PSNR, SSIM, etc.) appear to be computed using simulated SLC-off gaps applied to otherwise complete SLCANT images rather than genuine SLC-off acquisitions. This setup risks under-testing the method's ability to handle the irregular scan-line geometry, radiometric statistics, and rapid surface-change statistics of real Antarctic SLC-off scenes under variable illumination and snow conditions, directly weakening support for the central claim of high-fidelity non-reference restoration on real-world data.
minor comments (2)
  1. Clarify in the methods whether the two-stage pipeline (latent diffusion + pixel refinement) includes any mechanisms to mitigate generative artifacts such as hallucinated features in large gap regions.
  2. Ensure all quantitative results include explicit baseline comparisons (e.g., against existing non-reference or reference-based gap-filling methods) with statistical significance testing.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on the evaluation methodology. We address the major comment below.

read point-by-point responses

  1. Referee: [Evaluation section] Evaluation section (likely §4 or §5): The quantitative fidelity metrics (PSNR, SSIM, etc.) appear to be computed using simulated SLC-off gaps applied to otherwise complete SLCANT images rather than genuine SLC-off acquisitions. This setup risks under-testing the method's ability to handle the irregular scan-line geometry, radiometric statistics, and rapid surface-change statistics of real Antarctic SLC-off scenes under variable illumination and snow conditions, directly weakening support for the central claim of high-fidelity non-reference restoration on real-world data.

    Authors: We agree that the quantitative metrics rely on simulated SLC-off gaps applied to complete SLCANT images, as genuine SLC-off acquisitions lack pixel-level ground truth for direct fidelity computation. The simulation replicates the known SLC failure pattern, including irregular scan-line geometry, using masks derived from actual Landsat 7 SLC-off data. SLCANT was constructed specifically to reflect Antarctic radiometric and surface properties. To address the concern about real-world variability, we will revise the evaluation section to explicitly distinguish simulated quantitative results from qualitative assessments on genuine SLC-off imagery, add further visual examples across diverse illumination and snow conditions, and strengthen discussion of the downstream crevasse segmentation results obtained on real restored scenes. This protocol is standard when ground truth is unavailable for real degraded data. revision: partial

Circularity Check

0 steps flagged

No circularity; derivation is self-contained as independent generative model

full rationale

The paper constructs the SLCANT dataset separately and trains a two-stage latent diffusion model (DiffGF) for non-reference restoration of SLC-off imagery. No equations, parameters, or claims in the abstract or described pipeline reduce the reported restorations or downstream segmentation results to the inputs by construction. The quantitative fidelity metrics are presented as outcomes of the trained generative process rather than fitted values of the target itself, and no load-bearing self-citations or uniqueness theorems are invoked. This matches the reader's assessment that the central claim retains independent content from the training data and model architecture.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard generative-model assumptions plus a new training dataset; no new physical entities are postulated.

free parameters (1)
  • Diffusion model weights and training hyperparameters
    Neural network parameters are learned from the SLCANT dataset and control the restoration behavior.
axioms (1)
  • domain assumption Diffusion models trained on domain-specific imagery can learn priors sufficient to inpaint missing scan lines without reference data
    Invoked by the two-stage latent diffusion plus pixel refinement design.

pith-pipeline@v0.9.0 · 5770 in / 1288 out tokens · 47653 ms · 2026-05-21T04:46:54.978400+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

61 extracted references · 61 canonical work pages

  1. [1]

    Contribution of Antarctica to past and future sea-level rise,

    R. DeConto, and D. Pollard, “Contribution of Antarctica to past and future sea-level rise,” Nature, vol. 531, no. 7596, pp. 591–597, Mar 31, 2016

    work page 2016

  2. [2]

    Accelerated global glacier mass loss in the early twenty -first century,

    R. Hugonnet, R. McNabb, E. Berthier et al., “Accelerated global glacier mass loss in the early twenty -first century,” Nature, vol. 592, no. 7856, pp. 726–+, Apr 29, 2021

    work page 2021

  3. [3]

    Fifty years of Landsat science and impacts,

    M. A. Wulder, D. P. Roy, V. C. Radeloff et al., “Fifty years of Landsat science and impacts,” Remote Sensing of Environment, vol. 280, Oct, 2022

    work page 2022

  4. [4]

    An Accurate and Automated Method for Identifying and Mapping Exposed Rock Outcrop in Antarctica Using Landsat 8 Images,

    J. Kang, X. Cheng, F. Hui et al., “An Accurate and Automated Method for Identifying and Mapping Exposed Rock Outcrop in Antarctica Using Landsat 8 Images,” Ieee Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 11, no. 1, pp. 57–67, Jan, 2018

    work page 2018

  5. [5]

    The Landsat Image Mosaic of Antarctica,

    R. Bindschadler, P. Vornberger, A. Fleming et al., “The Landsat Image Mosaic of Antarctica,” Remote Sensing of Environment, vol. 112, no. 12, pp. 4214–4226, Dec 15, 2008

    work page 2008

  6. [6]

    Mapping blue-ice areas in Antarctica using ETM plus and MODIS data,

    F. Hui, T. Ci, X. Cheng et al., “Mapping blue-ice areas in Antarctica using ETM plus and MODIS data,” Annals of Glaciology, vol. 55, no. 66, pp. 129 – 137, 2014, 2014

    work page 2014

  7. [7]

    Increased West Antarctic and unchanged East Antarctic ice discharge over the last 7 years,

    A. Gardner, G. Moholdt, T. Scambos et al., “Increased West Antarctic and unchanged East Antarctic ice discharge over the last 7 years,” Cryosphere, vol. 12, no. 2, pp. 521–547, Feb 13, 2018

    work page 2018

  8. [8]

    Phase 2 gap -fill algorithm: SLC -off gap -filled products gap -fill algorithm methodology,

    USGS. "Phase 2 gap -fill algorithm: SLC -off gap -filled products gap -fill algorithm methodology," June 24, 2025; https://www.usgs.gov/media/files/landsat-7-slc-gap-filled-products-phase- two-methodology

    work page 2025

  9. [9]

    Image inpainting: Overview and recent advances,

    C. Guillemot, and O. Le Meur, “Image inpainting: Overview and recent advances,” IEEE signal processing magazine, vol. 31, no. 1, pp. 127 –144, 2013

    work page 2013

  10. [10]

    Gaps-fill of SLC-off Landsat ETM plus satellite image using a geostatistical approach,

    C. Zhang, W. Li, and D. Travis, “Gaps-fill of SLC-off Landsat ETM plus satellite image using a geostatistical approach,” International Journal of Remote Sensing, vol. 28, no. 22, pp. 5103–5122, 2007

    work page 2007

  11. [11]

    Gap -filling of landsat 7 imagery using the direct sampling method,

    G. Yin, G. Mariethoz, and M. F. McCabe, “Gap -filling of landsat 7 imagery using the direct sampling method,” Remote Sensing, vol. 9, no. 1, pp. 12, 2016

    work page 2016

  12. [12]

    A comprehensive review of spatial-temporal-spectral information reconstruction techniques,

    Q. M. Wang, Y. J. Tang, Y. Ge et al., “A comprehensive review of spatial-temporal-spectral information reconstruction techniques,” Science of Remote Sensing, vol. 8, pp. 100102, Dec, 2023

    work page 2023

  13. [13]

    A simple and effective method for filling gaps in Landsat ETM plus SLC -off images,

    J. Chen, X. L. Zhu, J. E. Vogelmann et al., “A simple and effective method for filling gaps in Landsat ETM plus SLC -off images,” Remote Sensing of Environment, vol. 115, no. 4, pp. 1053–1064, Apr 15, 2011

    work page 2011

  14. [14]

    A new geostatistical approach for filling gaps in Landsat ETM plus SLC -off images,

    X. L. Zhu, D. S. Liu, and J. Chen, “A new geostatistical approach for filling gaps in Landsat ETM plus SLC -off images,” Remote Sensing of Environment, vol. 124, pp. 49–60, Sep, 2012

    work page 2012

  15. [15]

    Recovering missing pixels for Landsat ETM plus SLC -off imagery using multi -temporal regression analysis and a regularization method,

    C. Zeng, H. F. Shen, and L. P. Zhang, “Recovering missing pixels for Landsat ETM plus SLC -off imagery using multi -temporal regression analysis and a regularization method,” Remote Sensing of Environment, vol. 131, pp. 182–194, Apr 15, 2013

    work page 2013

  16. [16]

    Spatial –spectral radial basis function - based interpolation for Landsat ETM+ SLC -off image gap filling,

    Q. Wang, L. Wang, Z. Li et al., “Spatial –spectral radial basis function - based interpolation for Landsat ETM+ SLC -off image gap filling,” IEEE Transactions on Geoscience and Remote Sensing, vol. 59, no. 9, pp. 7901 – 7917, 2020

    work page 2020

  17. [17]

    Progressive gap -filling in optical remote sensing imagery through a cascade of temporal and spatial reconstruction models,

    J. G. Wu, T. W. Li, L. P. Lin et al., “Progressive gap -filling in optical remote sensing imagery through a cascade of temporal and spatial reconstruction models,” Remote Sensing of Environment, vol. 311, pp. 114245, Sep 1, 2024

    work page 2024

  18. [18]

    Context encoders: Feature learning by inpainting,

    D. Pathak, P. Krahenbuhl, J. Donahue et al., “Context encoders: Feature learning by inpainting,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 2536–2544

    work page 2016

  19. [19]

    Deep Image Prior,

    D. Ulyanov, A. Vedaldi, V. Lempitsky et al., “Deep Image Prior,” in 31st IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Salt Lake City, UT, 2018, pp. 9446–9454

    work page 2018

  20. [20]

    Image inpainting for irregular holes using partial convolutions,

    G. Liu, F. A. Reda, K. J. Shih et al., “Image inpainting for irregular holes using partial convolutions,” in Proceedings of the European conference on computer vision (ECCV), 2018, pp. 85–100

    work page 2018

  21. [21]

    Edgeconnect: Structure guided image inpainting using edge prediction,

    K. Nazeri, E. Ng, T. Joseph et al., “Edgeconnect: Structure guided image inpainting using edge prediction,” in Proceedings of the IEEE/CVF international conference on computer vision workshops, 2019, pp. 0–0

    work page 2019

  22. [22]

    Resolution -robust large mask inpainting with fourier convolutions,

    R. Suvorov, E. Logacheva, A. Mashikhin et al., “Resolution -robust large mask inpainting with fourier convolutions,” in Proceedings of the IEEE/CVF winter conference on applications of computer vision, 2022, pp. 2149–2159

    work page 2022

  23. [23]

    Missing Data Reconstruction in Remote Sensing Image With a Unified Spatial -Temporal-Spectral Deep Convolutional Neural Network,

    Q. Zhang, Q. Q. Yuan, C. Zeng et al., “Missing Data Reconstruction in Remote Sensing Image With a Unified Spatial -Temporal-Spectral Deep Convolutional Neural Network,” IEEE Transactions on Geoscience and Remote Sensing, vol. 56, no. 8, pp. 4274–4288, Aug, 2018

    work page 2018

  24. [24]

    DecRecNet: A Decoupling - Reconstruction Network for Restoring the Missing Information of Optical Remote Sensing Images,

    W. Liu, H. Cui, Y. Jiang et al., “DecRecNet: A Decoupling - Reconstruction Network for Restoring the Missing Information of Optical Remote Sensing Images,” Ieee Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 16, pp. 10204–10228, 2023, 2023

    work page 2023

  25. [25]

    Collaborative dual -harmonization reconstruction network for large -ratio cloud occlusion missing information in high-resolution remote sensing images,

    W. Liu, Y. Jiang, F. Li et al., “Collaborative dual -harmonization reconstruction network for large -ratio cloud occlusion missing information in high-resolution remote sensing images,” Engineering Applications of Artificial Intelligence, vol. 136, Oct, 2024

    work page 2024

  26. [26]

    Satellite Image Inpainting With Edge-Conditional Expectation Attention,

    D. Zhou, Y. Chen, Y. Zhang et al., “Satellite Image Inpainting With Edge-Conditional Expectation Attention,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 18, pp. 10830 –10845, 2025

    work page 2025

  27. [27]

    A Single Image Deep Learning Approach to Restoration of Corrupted Landsat -7 Satellite Images,

    A. Petrovskaia, R. Jana, and I. Oseledets, “A Single Image Deep Learning Approach to Restoration of Corrupted Landsat -7 Satellite Images,” Sensors, vol. 22, no. 23, pp. 9273, Nov 28, 2022

    work page 2022

  28. [28]

    Stripe Error Correction for Landsat -7 Using Deep Learning,

    H. Adıyaman, Y. Emre Varul, T. Bakırman et al., “Stripe Error Correction for Landsat -7 Using Deep Learning,” PFG –Journal of Photogrammetry, Remote Sensing and Geoinformation Science, vol. 93, no. 1, pp. 51–63, 2025

    work page 2025

  29. [29]

    Removal of line striping and shot noise from remote sensing imagery using a deep neural network with post - processing for improved restoration quality,

    A. Deshpande, S. Patale, and S. Roy, “Removal of line striping and shot noise from remote sensing imagery using a deep neural network with post - processing for improved restoration quality,” International Journal of Remote Sensing, vol. 42, no. 19, pp. 7357–7380, Oct 2, 2021

    work page 2021

  30. [30]

    Diffusion Models Beat GANs on Image Synthesis,

    P. Dhariwal, and A. Nichol, “Diffusion Models Beat GANs on Image Synthesis,” Advances in Neural Information Processing Systems 34 (Neurips 2021), vol. 34, pp. 8780–8794, 2021

    work page 2021

  31. [31]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” Advances in neural information processing systems, vol. 33, pp. 6840 –6851, 2020

    work page 2020

  32. [32]

    RePaint: Inpainting using Denoising Diffusion Probabilistic Models,

    A. Lugmayr, M. Danelljan, A. Romero et al., “RePaint: Inpainting using Denoising Diffusion Probabilistic Models,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), New Orleans, LA, 2022, pp. 11451–11461

    work page 2022

  33. [33]

    High -resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz et al., “High -resolution image synthesis with latent diffusion models,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 10684 – 10695

    work page 2022

  34. [34]

    BrushNet: A Plug -and-Play Image Inpainting Model with Decomposed Dual -Branch Diffusion,

    X. Ju, X. Liu, X. Wang et al., "BrushNet: A Plug -and-Play Image Inpainting Model with Decomposed Dual -Branch Diffusion," Lecture Notes in Computer Science. pp. 150–168, 2025

    work page 2025

  35. [35]

    LatentPaint: Image Inpainting in Latent Space with Diffusion Models,

    C. Corneanu, R. Gadde, A. Martinez et al., "LatentPaint: Image Inpainting in Latent Space with Diffusion Models," IEEE Winter Conference on Applications of Computer Vision. pp. 4322–4331, 2024. [36]L. Pang, X. Rui, L. Cui et al., "HIR -Diff: Unsupervised Hyperspectral Image Restoration Via Improved Diffusion Models," IEEE Conference on Computer Vision and...

    work page 2024

  36. [36]

    Latent Diffusion Enhanced Rectangle Transformer for Hyperspectral Image Restoration,

    M. Li, Y. Fu, T. Zhang et al., “Latent Diffusion Enhanced Rectangle Transformer for Hyperspectral Image Restoration,” Ieee Transactions on Pattern Analysis and Machine Intelligence, vol. 47, no. 1, pp. 549 –564, Jan, 2025

    work page 2025

  37. [37]

    IDF -CR: Iterative Diffusion Process for Divide-and-Conquer Cloud Removal in Remote-Sensing Images,

    M. L. Wang, Y. X. Song, P. X. Wei et al., “IDF -CR: Iterative Diffusion Process for Divide-and-Conquer Cloud Removal in Remote-Sensing Images,” IEEE Transactions on Geoscience and Remote Sensing, vol. 62, 2024

    work page 2024

  38. [38]

    Diffusion Enhancement for Cloud Removal in Ultra-Resolution Remote Sensing Imagery,

    J. Sui, Y. Ma, W. Yang et al., “Diffusion Enhancement for Cloud Removal in Ultra-Resolution Remote Sensing Imagery,” Ieee Transactions on Geoscience and Remote Sensing, vol. 62, 2024, 2024

    work page 2024

  39. [39]

    OS -MambaDiff: An Optical -SAR Fusion Diffusion Framework With Mamba for Robust Cloud Removal in Remote Sensing Imagery,

    P. Yang, B. Huang, Y. Ye et al., “OS -MambaDiff: An Optical -SAR Fusion Diffusion Framework With Mamba for Robust Cloud Removal in Remote Sensing Imagery,” Ieee Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 19, pp. 10716–10729, 2026, 2026

    work page 2026

  40. [40]

    A global -local interaction and conditional consistency constrained diffusion model for SAR -guided optical image cloud removal,

    L. Cao, J. Pan, J. Xu et al., “A global -local interaction and conditional consistency constrained diffusion model for SAR -guided optical image cloud removal,” International Journal of Applied Earth Observation and Geoinformation, vol. 146, Feb, 2026

    work page 2026

  41. [41]

    A Conditional Diffusion Model With Fast Sampling Strategy for Remote Sensing Image Super - Resolution,

    F. N. Meng, Y. J. Chen, H. Y. Jing et al., “A Conditional Diffusion Model With Fast Sampling Strategy for Remote Sensing Image Super - Resolution,” IEEE Transactions on Geoscience and Remote Sensing, vol. 62, 2024

    work page 2024

  42. [42]

    ORDiffSR: An effective one -step diffusion network for single remote sensing image super -resolution,

    T. Tang, J. Liu, X. Luo et al., “ORDiffSR: An effective one -step diffusion network for single remote sensing image super -resolution,” Expert Systems with Applications, vol. 320, Jul 15, 2026

    work page 2026

  43. [43]

    Cross-platform super-resolution: A diffusion model approach for enhancing satellite imagery with aerial data,

    Z. Wang, C. García, and B. Halpern, “Cross-platform super-resolution: A diffusion model approach for enhancing satellite imagery with aerial data,” International Journal of Applied Earth Observation and Geoinformation, vol. 146, Feb, 2026

    work page 2026

  44. [44]

    SFSR: Spectral Fusion Super - Resolution for Multisensor Remote Sensing With Degraded References,

    S. Gwak, S. Yang, and M. Kang, “SFSR: Spectral Fusion Super - Resolution for Multisensor Remote Sensing With Degraded References,” Ieee Transactions on Geoscience and Remote Sensing, vol. 64, 2026, 2026

    work page 2026

  45. [45]

    Ship -Go: SAR ship images inpainting via instance-to-image generative diffusion models,

    X. Zhang, Y. Li, F. Li et al., “Ship -Go: SAR ship images inpainting via instance-to-image generative diffusion models,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. 207, pp. 203–217, 2024

    work page 2024

  46. [46]

    Open-vocabulary generative vision - language models for creating a large -scale remote sensing change detection dataset☆,

    Y. Zan, S. Ji, S. Chao et al., “ Open-vocabulary generative vision - language models for creating a large -scale remote sensing change detection dataset☆,” Isprs Journal of Photogrammetry and Remote Sensing, vol. 225, pp. 275–290, Jul, 2025

    work page 2025

  47. [47]

    MetaEarth: A Generative Foundation Model for Global -Scale Remote Sensing Image Generation,

    Z. Yu, C. Liu, L. Liu et al., “MetaEarth: A Generative Foundation Model for Global -Scale Remote Sensing Image Generation,” Ieee Transactions on Pattern Analysis and Machine Intelligence, vol. 47, no. 3, pp. 1764 –1781, Mar, 2025

    work page 2025

  48. [48]

    Diffusion Models Meet Remote Sensing: Principles, Methods, and Perspectives,

    Y. D. Liu, J. Yue, S. B. Xia et al., “Diffusion Models Meet Remote Sensing: Principles, Methods, and Perspectives,” IEEE Transactions on Geoscience and Remote Sensing, vol. 62, 2024

    work page 2024

  49. [49]

    Efficient Diffusion Model for Image Restoration by Residual Shifting,

    Z. Yue, J. Wang, and C. C. Loy, “Efficient Diffusion Model for Image Restoration by Residual Shifting,” IEEE Trans Pattern Anal Mach Intell, vol. 47, no. 1, pp. 116–130, Jan, 2025

    work page 2025

  50. [50]

    The Unreasonable Effectiveness of Deep Features as a Perceptual Metric,

    R. Zhang, P. Isola, A. Efros et al., “The Unreasonable Effectiveness of Deep Features as a Perceptual Metric,” in 31st IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Salt Lake City, UT, 2018, pp. 586–595

    work page 2018

  51. [51]

    Swin Transformer: Hierarchical Vision Transformer using Shifted Windows,

    Z. Liu, Y. Lin, Y. Cao et al., “Swin Transformer: Hierarchical Vision Transformer using Shifted Windows,” in 18th IEEE/CVF International Conference on Computer Vision (ICCV), Electr Network, 2021, pp. 9992 – 10002

    work page 2021

  52. [52]

    Taming Transformers for High- Resolution Image Synthesis,

    P. Esser, R. Rombach, B. Ommer et al., “Taming Transformers for High- Resolution Image Synthesis,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), null, ELECTR NETWORK, 2021, pp. 12868–12878

    work page 2021

  53. [53]

    Gradient -free Decoder Inversion in Latent Diffusion Models,

    S. Hong, S. Y. Jeon, K. Lee et al., “Gradient -free Decoder Inversion in Latent Diffusion Models,” in Advances in Neural Information Processing Systems, 2024, pp. 82982––83007

    work page 2024

  54. [54]

    Real -Time Single Image and Video Super -Resolution Using an Efficient Sub -Pixel Convolutional Neural Network,

    W. Shi, J. Caballero, F. Huszár et al., “Real -Time Single Image and Video Super -Resolution Using an Efficient Sub -Pixel Convolutional Neural Network,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, WA, 2016, pp. 1874–1883

    work page 2016

  55. [55]

    Deep Residual Learning for Image Recognition,

    K. He, X. Zhang, S. Ren et al., “Deep Residual Learning for Image Recognition,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, WA, 2016, pp. 770–778

    work page 2016

  56. [56]

    Foreground -aware Semantic Representations for Image Harmonization,

    K. Sofiiuk, P. Popenova, A. Konushin et al., “Foreground -aware Semantic Representations for Image Harmonization,” in IEEE Winter Conference on Applications of Computer Vision (WACV), Electr Network, 2021, pp. 1619–1628

    work page 2021

  57. [57]

    Image quality metrics: PSNR vs. SSIM,

    A. Hore, and D. Ziou, “Image quality metrics: PSNR vs. SSIM,” in 2010 20th international conference on pattern recognition, 2010, pp. 2366–2369

    work page 2010

  58. [58]

    A universal image quality index,

    Z. Wang, and A. C. Bovik, “A universal image quality index,” IEEE Signal Processing Letters, vol. 9, no. 3, pp. 81–84, Mar, 2002

    work page 2002

  59. [59]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh et al., “Image quality assessment: from error visibility to structural similarity,” IEEE transactions on image processing, vol. 13, no. 4, pp. 600–612, 2004

    work page 2004

  60. [60]

    SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers,

    E. Z. Xie, W. H. Wang, Z. D. Yu et al., “SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers,” Advances in Neural Information Processing Systems 34 (Neurips 2021), vol. 34, pp. 12077–12090, 2021

    work page 2021

  61. [61]

    Bedmap2: improved ice bed, surface and thickness datasets for Antarctica,

    P. Fretwell, H. Pritchard, D. Vaughan et al., “Bedmap2: improved ice bed, surface and thickness datasets for Antarctica,” Cryosphere, vol. 7, no. 1, pp. 375–393, 2013, 2013

    work page 2013