Pith. sign in

REVIEW 2 major objections 4 minor 54 references

Efficient Seismic Data Interpolation via Sparse Attention Transformer and Diffusion Model

T0 review · 2 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A transformer and diffusion model, joined by a small prior-extraction network, is claimed to interpolate missing seismic traces more accurately than seven baselines while using only four reverse diffusion steps.

desk verdict The core sparse-attention mechanism is a mathematical no-op as written: Eq. 25's negative squared Euclidean distance fed through ReLU is identically zero, so the paper's headline contribution cannot explain its reported results. read the letter →

arxiv 2506.07923 v1 pith:42464AKH submitted 2025-06-09 physics.geo-ph

Xiaoli Wei , Chunxia Zhang , Baisong Jiang , Anxiang Di , Deng Xiong , Jiangshe Zhang , Mingming Gong This is my paper
classification physics.geo-ph
keywords seismicdatainterpolationdiffusionmodelsparseattentiontransformerPriorExtractionNetworknegativesquaredEuclideandistancemissingtracereconstructionsingle-stagetraining
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 sets out to show that a transformer and a diffusion model can be joined, through a small bridge network, to interpolate missing seismic traces both faster and more accurately than current deep learning methods. The proposed Diff-spaformer feeds a sparse self-attention U-Net with observed data and a missing-trace mask, while a compact latent prior generated by a four-step diffusion process is fused into every encoder-decoder layer. The authors identify two design choices as decisive: computing self-attention along channels rather than spatial positions, and scoring affinity with negative squared Euclidean distance instead of cosine similarity. On three test sets, for random and continuous missing traces, the model is reported to beat seven baselines on all metrics, including gains of roughly 1 dB in SNR over the strongest diffusion baseline. If the claims hold, plug-and-play diffusion interpolation's expensive iterative resampling can be replaced by a cheap deterministic sampler without sacrificing fidelity.

What carries the argument

The load-bearing object is the SPEN-SPA coupling. SPEN (Seismic Prior Extraction Network) is a convolutional-residual encoder that uses PixelUnshuffle, average pooling, and a multilayer perceptron to compress a seismic patch into a compact latent vector representing its energy distribution. A four-step diffusion process uses this latent as the clean state and generates a prior feature from the observed data. The SPA (sparse attention) module then computes multi-head self-attention along the channel dimension, so the affinity matrix is $c_n\times c_n$ rather than $N\times N$, scores similarity with the negative squared Euclidean distance $s(q,k)=-\|q-k\|_2^2$, and applies ReLU with a learnable scaling parameter to zero out low-correlation entries. Dynamic feature calibration gates the prior into the main feature flow at every layer, which the paper credits for the model's accuracy-efficiency balance.

What would settle it

Run a blind evaluation in which Diff-spaformer and the seven baselines are retrained from the paper's training splits with all hyperparameters frozen, and scored only on held-out shots never used during development; if the reported SNR advantage over SeisDDIMCR shrinks below about 1 dB or reverses, the central superiority claim would not survive.

Watch

Extended reading notes

Core claim

The central claim is that Diff-spaformer reconstructs missing seismic data more accurately than existing GAN-, U-Net-, attention-, and diffusion-based methods while requiring far fewer diffusion steps. The paper reports the best SNR on every tested configuration: for random missing traces, 38.29 dB on SEG C3, 34.35 dB on MA VO, and 36.95 dB on Model94, and for consecutive missing traces, 32.68, 34.81, and 26.98 dB, respectively. These numbers correspond to roughly 1 dB improvement over SeisDDIMCR on SEG C3 and Model94 random-missing tests and smaller margins elsewhere. The mechanism is the Seismic Prior Extraction Network (SPEN), which compresses complete and observed seismic data into a shared latent space; the diffusion process sharpens this prior in four reverse steps, and the Spaformer U-Net injects it into each layer through gated residual connections. The paper argues that channel-wise sparse attention with ReLU filtering and negative squared Euclidean distance gives the global modeling needed for large gaps at lower FLOPs and memory than spatial attention.

Load-bearing premise

The paper's superiority claim rests on the assumption that its reported test-set metrics were not inflated by using the test data to choose components and hyperparameters; the ablation section selects SPEN, FFN, gate, L2 similarity, and channel width with reference to the MA VO test set.

Editorial extensions

If this is right

  • Four reverse diffusion steps are enough for high-fidelity interpolation, so inference drops from hundreds of resampling iterations to about 0.9 seconds per MA VO slice.
  • Channel-wise sparse attention with L2 similarity offers a cheaper way to keep global context in seismic reconstruction, with reported FLOPs and memory only mildly higher than cosine similarity at the same channel width.
  • The method's advantage widens at high missing rates: on SEG C3 continuous gaps, the paper reports a 16.1 dB SNR drop for the plug-and-play baseline from 10% to 60% missing, while Diff-spaformer stays ahead at every rate.
  • Removing SPEN, or unsharing its parameters between complete and observed data, lowers test SNR, so the prior branch is doing real work rather than acting as a bypass.
  • The full model at 19M parameters reaches the top SNR with a moderate model size, whereas the stronger diffusion baseline is roughly twice as large.

Reading between the lines

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

  • Editorial inference: if the MA VO test set influenced the choice of architecture components and channel width, the test-set numbers in Tables I-II may be optimistic; an independent evaluation on a frozen development set would be needed to confirm the margins.
  • Editorial inference: the L2-affinity idea transfers naturally to other seismic inverse problems such as denoising and regularization, where amplitude fidelity matters; one could test whether the same margin over cosine similarity appears in those tasks.
  • Editorial inference: because diffusion operates on a compressed latent rather than on the full wavefield, the framework suggests a general recipe of small latent diffusion plus a strong reconstruction network that could reduce cost in other geophysical imaging applications.
  • Editorial inference: a direct test of the four-step claim would be to measure inversion quality versus diffusion steps on a fresh dataset; the reported gain from 2 to 100 steps is only 0.24 dB, implying much of the benefit is already captured at very few steps.
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

2 major / 4 minor

Summary. The paper proposes Diff-spaformer, a seismic data interpolation framework that combines a U-Net-style Spaformer with sparse channel-wise multi-head attention and a diffusion-based prior obtained from a Seismic Prior Extraction Network (SPEN). The authors report improved SNR/PSNR/SSIM over seven baselines on three public datasets for both random and continuous missing traces, with inference reduced to four reverse diffusion steps. The central claims are empirical: higher fidelity than SeisDDIMCR and comparable inference time to end-to-end models.

Significance. If the method worked as described, the contribution would be practically valuable: it offers a concrete way to cut the iterative sampling cost of diffusion-based seismic interpolation while preserving or improving fidelity. The experimental design is broad, covering three datasets, two missing-data scenarios, seven baselines, and efficiency comparisons. However, the significance currently rests on an internally inconsistent definition of the sparse attention mechanism and on model selection performed directly on the test set, so the evidence base is not yet reliable.

major comments (2)
  1. [Section III-D2/D3, Eqs. (23) and (25)] The sparse-attention branch is mathematically inert as written. Eq. (25) defines every affinity score as non-positive, namely s(Q_i,j, K^T_i,k) = -||Q_i,j - K^T_i,k||^2_2, and Eq. (23) then computes W_i = (1/omega) ReLU(S_i). Since ReLU of a non-positive number is zero, W_i is identically zero, head_i is zero in Eq. (24), and the SPA branch contributes nothing to Eq. (14). A faithful reimplementation would collapse the model to a gated U-Net with FFN only, so the reported improvements over SeisDDIMCR cannot be produced by the proposed sparse attention as specified. Please correct the similarity definition (e.g., an additive constant, an exponential form, or a similarity that can yield positive values), clarify the actual implementation, and ideally release code; without this, the core mechanism is not reproducible.
  2. [Section V, Tables III-IV and Figs. 9-10] The architecture and hyperparameter choices are selected using the MA VO test set, not a held-out validation set. Section V-A states that the ablation study is evaluated on the MA VO test set with continuous missing traces, and Figs. 9(b) and 10(a) report test SNR for choosing the diffusion steps T and the channel dimension C_k0. Table IV similarly compares L2 versus cosine similarity on the test set. Since the test set informed model selection, the test metrics in Tables I-II are optimistically biased, and the comparison with baselines (whose hyperparameters were not tuned on this test set) is not a fair assessment of generalization. Please re-run model selection on a validation split and report final metrics on an untouched test set.
minor comments (4)
  1. [Section III-D3] The text says that L2 similarity 'is not dominated by the vector magnitudes', but Eq. (25) explicitly depends on magnitudes through the norm of the difference; cosine similarity is the magnitude-invariant measure. This statement contradicts the definition.
  2. [Section IV-E1, Tables I-II] The claim that the method 'consistently demonstrates superior performance across all evaluation metrics' is overstated because ties occur with SeisDDIMCR on SSIM for continuous missing MA VO (0.979) and SEG C3 (0.973).
  3. [Section III-E, Eq. (26)] The text describes W_f1 and W_f2 as 3x3 convolution operations but then says 1x1 convolutions are applied at both ends; please clarify which weight corresponds to which convolution and how the channel expansion/restoration is realized.
  4. [Section V-B, Table IV] The FLOPs values in Table IV are computed only for the similarity operation, while Fig. 10(b) reports overall model FLOPs; the main text should state this distinction explicitly to avoid inconsistent efficiency comparisons.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: all claims are empirical comparisons on public data; the only notable flaw is an internal consistency issue, not a circular derivation.

full rationale

The paper is an empirical deep-learning interpolation study. Its central claim, that Diff-spaformer outperforms existing methods on SEG C3, MA VO, and Model94, is supported by direct quantitative comparisons on public test splits. No step in the method section derives a prediction from a fitted constant, and no parameter is relabeled as a prediction. The sparse-attention mechanism and the SPEN design are adopted from cited prior work, but those citations are not used as a load-bearing justification that forces the reported results; they supply architectural building blocks, and the reported gains rest on the paper's own experiments. The ablation study does select components (L2 versus cosine similarity, C_k0 dimension, diffusion steps) with reference to the MA VO test set, so the test numbers may be optimistically selected, but this is a model-selection/data-leakage concern, not a circularity: the reported metrics are not equal to the selection criterion by construction. The paper's self-citations to prior work such as SeisDDIMCR and Coarse-to-Fine are baseline comparisons or data-processing choices, not uniqueness theorems or unverified premises that the present claims reduce to. The internally inconsistent equations for the SPA branch, where Eq. 25 defines a non-positive similarity and Eq. 23 applies ReLU to it, would make the attention output identically zero as written; this is a serious correctness and reproducibility problem, but it is not a circular derivation because it does not make any claimed prediction equivalent to its input. No circular step can be exhibited from the manuscript text.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

No new physical entities, particles, forces, or conserved quantities are proposed. The new components are network modules (SPEN, SPA) whose architectures are adapted from existing image-restoration literature, particularly DiffIR [49] and sparse-attention work [51]. The free parameters listed above are the main hand-selected or data-selected quantities that the empirical claims depend on.

free parameters (6)
  • Loss weights lambda1, lambda2, lambda3 = 6.0, 1.0, 6.0
    Chosen by hand to weight missing-region error and diffusion loss; no search procedure or validation criterion is reported (Section IV-D).
  • Diffusion steps T = 4
    Selected from the trade-off in Fig. 9(b): increasing to 100 steps improves SNR by 0.24 dB but extends training time by over an hour.
  • First-layer channel dimension Ck0 = 64
    Chosen via ablation on the MA VO test set (Fig. 10a), balancing SNR against parameters and FLOPs; 128 gives higher SNR but about 3.3x FLOPs.
  • Compressed prior dimension C' = 256
    Set to 64x4, matching the fourth-layer channel count of the U-Net; no independent validation is reported.
  • Diffusion noise schedule beta = linear 0.1 to 0.99
    A standard linear DDPM schedule fixed by the authors; it is a modeling choice that affects the diffusion dynamics.
  • Learning rate schedule = 1e-4, decay by 10 after 50 epochs
    Standard Adam optimizer configuration used uniformly across datasets; not tuned per dataset.
assumptions (5)
  • standard math The DDPM forward and reverse process equations (Eqs. 1-5) describe a valid generative model.
    The paper builds directly on Ho et al. [48]; the ELBO derivation is standard and unmodified.
  • domain assumption Seismic data can be represented as 2D patches normalized to [0,1] without losing information needed for interpolation.
    All experiments use min-max normalized patches (Sections IV-B and IV-D); this discards absolute amplitudes, which may matter for field data with varying gain.
  • ad hoc to paper The latent prior Z0 = SPEN(X) captures sufficient seismic information to guide reconstruction, and Zc = SPEN(X_obs) provides a useful condition.
    No theoretical guarantee is provided; the paper supports this only through the 0.32 dB ablation drop in Table III on the MA VO test set.
  • ad hoc to paper Negative squared Euclidean distance is a better similarity function for seismic amplitude modeling than cosine similarity.
    Asserted from PCA visualizations and a single ablation (Section III-D3, Table IV), not derived from seismic signal properties.
  • domain assumption Patch-based training with random crops generalizes to full shot gathers at inference.
    Training uses randomly cropped patches while inference appears to operate on larger data; no stitching or boundary handling details are given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Efficient Seismic Data Interpolation via Sparse Attention Transformer and Diffusion Model." pith.science (2026). https://pith.science/paper/42464AKH

@misc{pith2026250607923,
  author       = {Pith},
  title        = {Pith review of: Efficient Seismic Data Interpolation via Sparse Attention Transformer and Diffusion Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/42464AKH}},
  note         = {Machine review of arXiv:2506.07923}
}
read the original abstract

Seismic data interpolation is a critical pre-processing step for improving seismic imaging quality and remains a focus of academic innovation. To address the computational inefficiencies caused by extensive iterative resampling in current plug-and-play diffusion interpolation methods, we propose the diffusion-enhanced sparse attention transformer (Diff-spaformer), a novel deep learning framework. Our model integrates transformer architectures and diffusion models via a Seismic Prior Extraction Network (SPEN), which serves as a bridge module. Full-layer sparse multi-head attention and feed-forward propagation capture global information distributions, while the diffusion model provides robust prior guidance. To mitigate the computational burden of high-dimensional representations, self-attention is computed along the channel rather than the spatial dimension. We show that using negative squared Euclidean distance to compute sparse affinity matrices better suits seismic data modeling, enabling broader contribution from amplitude feature nodes. An adaptive ReLU function further discards low or irrelevant self-attention values. We conduct training within a single-stage optimization framework, requiring only a few reverse diffusion sampling steps during inference. Extensive experiments demonstrate improved interpolation fidelity and computational efficiency for both random and continuous missing data, offering a new paradigm for high-efficiency seismic data reconstruction under complex geological conditions.

Figures

Figures reproduced from arXiv: 2506.07923 by the authors.

Figure 1
Figure 1. The overall framework of Diff-spaformer. It comprises three components, i.e., SPEN, diffusion process, and Spaformer. SPEN encodes seismic prior [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. SPA module between feature vectors. Cosine similarity is a widely used metric that quantifies the similarity between vectors based on their directional alignment in the vector space. Its core principle involves evaluating the degree of similarity between two non-zero vectors by calculating the cosine of the angle between them. However, studies have shown that for tasks requiring strong global pixel attention, the op… view at source ↗
Figure 3
Figure 3. (a) PCA projection of the affinity matrix [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: FFN module This Diff-spaformer framework optimizes the interplay be￾tween conditional feature encoding, diffusion-based prior gen￾eration, and multi-scale feature integration, ensuring efficient [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Interpolation results of the Model94 complete test slice with a 79.2% random missing ratio on different methods. We restore the seismic data to its [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: The f-k spectra of Model94 test data interpolation results with 79.2% random missing traces on different methods. quantitative interpolation evaluation results on the three test sets. Compared with comparative methods, our method consis￾tently demonstrates superior per…
Figure 7
Figure 7. Figure 7: Interpolation results of the SEG C3 patch with a 35% continuous missing ratio on different methods. We restore seismic data to its native amplitude [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: The wiggle plot of continuous missing gap interpolation results for a [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: (a) Curves of SNR versus the continuous missing ratio for different [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]
Figure 10
Figure 10. Figure 10: (a) The comparison of the test SNR, number of parameters (Params), and training time cost for our method under different selections of [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

54 extracted references · 50 canonical work pages

  1. [1]

    Five-dimensional interpolation: Recovering from ac- quisition constraints,

    D. Trad, “Five-dimensional interpolation: Recovering from ac- quisition constraints,”Geophysics, vol. 74, no. 6, pp. V123– V132, 2009

    work page 2009

  2. [2]

    Seismic trace interpolation in the fx domain,

    S. Spitz, “Seismic trace interpolation in the fx domain,”Geo- physics, vol. 56, no. 6, pp. 785–794, 1991

    work page 1991

  3. [3]

    Seismic trace interpolation in the f-x-y domain,

    Y . Wang, “Seismic trace interpolation in the f-x-y domain,” Geophysics, vol. 67, no. 4, pp. 1232–1239, 2002

    work page 2002

  4. [4]

    5d dealiased seismic data interpolation using nonstationary prediction-error filter,

    Y . Chen, S. Fomel, H. Wang, and S. Zu, “5d dealiased seismic data interpolation using nonstationary prediction-error filter,” Geophysics, vol. 86, no. 5, pp. V419–V429, 2021

    work page 2021

  5. [5]

    Wave-equation trace interpolation,

    J. Ronen, “Wave-equation trace interpolation,”Geophysics, vol. 52, no. 7, pp. 973–984, 1987

    work page 1987

  6. [6]

    Applications of plane-wave destruction filters,

    S. Fomel, “Applications of plane-wave destruction filters,”Geo- physics, vol. 67, no. 6, pp. 1946–1960, 2002

    work page 1946

  7. [7]

    Interpolation and extrapolation using a high-resolution discrete fourier transform,

    M. D. Sacchi, T. J. Ulrych, and C. J. Walker, “Interpolation and extrapolation using a high-resolution discrete fourier transform,” IEEE Transactions on Signal Processing, vol. 46, no. 1, pp. 31– 38, 1998

    work page 1998

  8. [8]

    Fourier reconstruction of nonuni- formly sampled, aliased seismic data,

    P. Zwartjes and M. Sacchi, “Fourier reconstruction of nonuni- formly sampled, aliased seismic data,”Geophysics, vol. 72, no. 1, pp. V21–V32, 2007

    work page 2007

Show all 54 references
  1. [9]

    Fourier reconstruction with sparse inversion,

    P. Zwartjes and A. Gisolf, “Fourier reconstruction with sparse inversion,”Geophysical Prospecting, vol. 55, no. 2, pp. 199– 221, 2007

    2007

  2. [10]

    Beyond alias hierarchical scale curvelet interpolation of regularly and irregularly sampled seismic data,

    M. Naghizadeh and M. D. Sacchi, “Beyond alias hierarchical scale curvelet interpolation of regularly and irregularly sampled seismic data,”Geophysics, vol. 75, no. 6, pp. WB189–WB202, 2010

    2010

  3. [11]

    Application of randomized sampling schemes to curvelet-based sparsity- promoting seismic data recovery,

    R. Shahidi, G. Tang, J. Ma, and F. J. Herrmann, “Application of randomized sampling schemes to curvelet-based sparsity- promoting seismic data recovery,”Geophysical Prospecting, vol. 61, no. 5, pp. 973–997, 2013

    2013

  4. [12]

    Seismic data interpolation by greedy local radon transform,

    J. Wang, M. Ng, and M. Perz, “Seismic data interpolation by greedy local radon transform,”Geophysics, vol. 75, no. 6, pp. WB225–WB234, 2010

    2010

  5. [13]

    Rank-reduction-based trace interpolation,

    S. Trickett, L. Burroughs, A. Milton, L. Walton, and R. Dack, “Rank-reduction-based trace interpolation,” inSEG Interna- tional Exposition and Annual Meeting, no. SEG-2010-3829, October 2010

    2010

  6. [14]

    Three-dimensional irregular seismic data reconstruction via low-rank matrix completion,

    J. Ma, “Three-dimensional irregular seismic data reconstruction via low-rank matrix completion,”Geophysics, vol. 78, no. 5, pp. V181–V192, 2013

    2013

  7. [15]

    Faultseg3d: Using synthetic data sets to train an end-to-end convolutional neural network for 3d seismic fault segmentation,

    X. Wu, L. Liang, Y . Shi, and S. Fomel, “Faultseg3d: Using synthetic data sets to train an end-to-end convolutional neural network for 3d seismic fault segmentation,”GEOPHYSICS, vol. 84, no. 3, pp. IM35–IM45, 2019

    2019

  8. [16]

    Quadratic unet for seismic random noise attenuation,

    X. Wang, Y . Sui, and J. Ma, “Quadratic unet for seismic random noise attenuation,”Geophysics, vol. 90, no. 2, pp. V43–V55, 2025

    2025

  9. [17]

    Upnet: Uncertainty-based picking deep learning network for robust first break picking,

    H. Wang, J. Zhang, X. Wei, L. Long, C. Zhang, and Z. Guo, “Upnet: Uncertainty-based picking deep learning network for robust first break picking,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, no. 5924214, 2024

    2024

  10. [18]

    Seismic data interpolation through convolutional autoencoder,

    S. Mandelli, F. Borra, V . Lipari, P. Bestagini, A. Sarti, and S. Tubaro, “Seismic data interpolation through convolutional autoencoder,” inSEG International Exposition and Annual Meeting, October 2018, no. SEG-2018-2995428

    2018

  11. [19]

    Self-supervised learning for seismic data reconstruction and denoising,

    F. Meng, Q. Fan, and Y . Li, “Self-supervised learning for seismic data reconstruction and denoising,”IEEE Geoscience and Remote Sensing Letters, vol. 19, no. 7502805, 2021

    2021

  12. [20]

    Reconstruction of sparsely sampled seismic data via residual u-net,

    S. Tang, Y . Ding, H.-W. Zhou, and H. Zhou, “Reconstruction of sparsely sampled seismic data via residual u-net,”IEEE Geoscience and Remote Sensing Letters, vol. 19, 2022

    2022

  13. [21]

    Seismic data interpo- lation based on u-net with texture loss,

    W. Fang, L. Fu, M. Zhang, and Z. Li, “Seismic data interpo- lation based on u-net with texture loss,”Geophysics, vol. 86, no. 1, pp. V41–V54, 2021

    2021

  14. [22]

    A partial convolution-based deep-learning network for seismic data regularization,

    S. Pan, K. Chen, J. Chen, Z. Qin, Q. Cui, and J. Li, “A partial convolution-based deep-learning network for seismic data regularization,”Computers & Geosciences, vol. 145, no. 104609, 2020

    2020

  15. [23]

    Seismic data reconstruction via wavelet-based residual deep learning,

    N. Liu, L. Wu, J. Wang, H. Wu, J. Gao, and D. Wang, “Seismic data reconstruction via wavelet-based residual deep learning,” IEEE Transactions on Geoscience and Remote Sensing, vol. 60, no. 4508213, 2022

    2022

  16. [24]

    Seismic data interpolation using CycleGAN,

    H. Kaur, N. Pham, and S. Fomel, “Seismic data interpolation using CycleGAN,” inSEG Technical Program Expanded Ab- stracts, 2019, pp. 2202–2206

    2019

  17. [25]

    Big gaps seismic data interpolation using conditional Wasserstein generative adversarial networks with gradient penalty,

    Q. Wei and X. Li, “Big gaps seismic data interpolation using conditional Wasserstein generative adversarial networks with gradient penalty,”Exploration Geophysics, vol. 53, no. 5, pp. 477–486, 2022

    2022

  18. [26]

    Ensemble deep learning for enhanced seismic data reconstruction,

    M. M. Abedi, D. Pardo, and T. Alkhalifah, “Ensemble deep learning for enhanced seismic data reconstruction,”IEEE Trans- actions on Geoscience and Remote Sensing, vol. 62, no. 5916311, 2024

    2024

  19. [27]

    Seismic data consecutively missing trace interpolation based on multistage neural network training process,

    T. He, B. Wu, and X. Zhu, “Seismic data consecutively missing trace interpolation based on multistage neural network training process,”IEEE Geoscience and Remote Sensing Letters, vol. 19, no. 7504105, 2022

    2022

  20. [28]

    Self-supervised simul- taneous deblending and interpolation of incomplete blended data using a multistep blind-trace u-net,

    B.-F. Wang, S.-C. Lin, and X.-Y . Chen, “Self-supervised simul- taneous deblending and interpolation of incomplete blended data using a multistep blind-trace u-net,”Petroleum Science, vol. 22, no. 3, pp. 1098–1109, 2025

    2025

  21. [29]

    Seismic data reconstruction via recurrent residual multiscale inference,

    A. Song, C. Wang, C. Zhang, J. Zhang, and D. Xiong, “Seismic data reconstruction via recurrent residual multiscale inference,” IEEE Geoscience and Remote Sensing Letters, vol. 19, no. 8029105, 2022

    2022

  22. [30]

    Seismic data interpolation using dual-domain conditional generative adversarial networks,

    D. Chang, W. Yang, X. Yong, G. Zhang, W. Wang, H. Li, and Y . Wang, “Seismic data interpolation using dual-domain conditional generative adversarial networks,”IEEE Geoscience and Remote Sensing Letters, vol. 18, no. 10, pp. 1856–1860, 2020

    2020

  23. [31]

    Hybrid loss-guided coarse-to-fine model for seismic data consecutively missing trace reconstruction,

    X. Wei, C. Zhang, H. Wang, Z. Zhao, D. Xiong, S. Xu, J. Zhang, and S.-W. Kim, “Hybrid loss-guided coarse-to-fine model for seismic data consecutively missing trace reconstruction,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, no. 5923315, 2022

    2022

  24. [32]

    Attention and hybrid loss guided deep learning for consecutively missing seismic data reconstruction,

    J. Yu and B. Wu, “Attention and hybrid loss guided deep learning for consecutively missing seismic data reconstruction,” IEEE Transactions on Geoscience and Remote Sensing, vol. 60, no. 5902108, 2022

    2022

  25. [33]

    Self-attention generative ad- versarial network interpolating and denoising seismic signals simultaneously,

    M. Ding, Y . Zhou, and Y . Chi, “Self-attention generative ad- versarial network interpolating and denoising seismic signals simultaneously,”Remote Sensing, vol. 16, no. 2, p. 305, 2024

    2024

  26. [34]

    Seismic impedance inversion based on residual attention network,

    B. Wu, Q. Xie, and B. Wu, “Seismic impedance inversion based on residual attention network,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, no. 4511117, 2022

    2022

  27. [35]

    Seismic data interpolation based on multi-scale transformer,

    Y . Guo, L. Fu, and H. Li, “Seismic data interpolation based on multi-scale transformer,”IEEE Geoscience and Remote Sensing Letters, vol. 20, no. 7504205, 2023

    2023

  28. [36]

    Can learning from natural image denoising be used for seismic data interpolation?

    H. Zhang, X. Yang, and J. Ma, “Can learning from natural image denoising be used for seismic data interpolation?”Geophysics, vol. 85, no. 4, pp. W A115–W A136, 2020

    2020

  29. [37]

    A projection-onto-convex-sets network for 3d seismic data interpolation,

    Y . Chen, S. Yu, and J. Ma, “A projection-onto-convex-sets network for 3d seismic data interpolation,”Geophysics, vol. 88, no. 3, pp. V249–V265, 2023

    2023

  30. [38]

    Physics- informed neural networks: A deep learning framework for solv- ing forward and inverse problems involving nonlinear partial differential equations,

    M. Raissi, P. Perdikaris, and G. E. Karniadakis, “Physics- informed neural networks: A deep learning framework for solv- ing forward and inverse problems involving nonlinear partial differential equations,”Journal of Computational Physics, vol. 378, pp. 686–707, 2019

    2019

  31. [39]

    PINNslope: seis- JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 14 mic data interpolation and local slope estimation with physics informed neural networks,

    F. Brandolin, M. Ravasi, and T. Alkhalifah, “PINNslope: seis- JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 14 mic data interpolation and local slope estimation with physics informed neural networks,”Geophysics, vol. 89, no. 4, pp. V331–V345, 2024

    2015

  32. [40]

    Deep diffusion models for seismic processing,

    R. Durall, A. Ghanim, M. R. Fernandez, N. Ettrich, and J. Keu- per, “Deep diffusion models for seismic processing,”Computers & Geosciences, vol. 177, no. 105377, 2023

    2023

  33. [41]

    Generative interpolation via a diffusion probabilistic model,

    Q. Liu and J. Ma, “Generative interpolation via a diffusion probabilistic model,”Geophysics, vol. 89, no. 1, pp. V65–V85, 2024

    2024

  34. [42]

    Repaint: Inpainting using denoising diffusion probabilistic models,

    A. Lugmayr, M. Danelljan, A. Romero, F. Yu, R. Timofte, and L. Van Gool, “Repaint: Inpainting using denoising diffusion probabilistic models,” inProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR), June 2022, pp. 11 461–11 471

    2022

  35. [43]

    Seismic data interpolation via denoising diffusion implicit models with coherence-corrected resampling,

    X. Wei, C. Zhang, H. Wang, C. Tan, D. Xiong, B. Jiang, J. Zhang, and S.-W. Kim, “Seismic data interpolation via denoising diffusion implicit models with coherence-corrected resampling,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, no. 5933217, 2024

    2024

  36. [44]

    Stochastic solutions for simultaneous seismic data denoising and reconstruction via score-based generative models,

    C. Meng, J. Gao, Y . Tian, H. Chen, W. Zhang, and R. Luo, “Stochastic solutions for simultaneous seismic data denoising and reconstruction via score-based generative models,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, no. 5919415, 2024

    2024

  37. [45]

    Seismic data reconstruction based on conditional constraint diffusion model,

    F. Deng, S. Wang, X. Wang, and P. Fang, “Seismic data reconstruction based on conditional constraint diffusion model,” IEEE Geoscience and Remote Sensing Letters, vol. 21, no. 7502305, 2024

    2024

  38. [46]

    Seisfusion: Constrained diffusion model with input guidance for 3-d seismic data interpolation and reconstruction,

    S. Wang, F. Deng, P. Jiang, Z. Gong, X. Wei, and Y . Wang, “Seisfusion: Constrained diffusion model with input guidance for 3-d seismic data interpolation and reconstruction,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, no. 5929815, 2024

    2024

  39. [47]

    Re- constructing regularly missing seismic traces with a classifier- guided diffusion model,

    X. Wang, Z. Wang, Z. Xiong, Y . Yang, C. Zhu, and J. Gao, “Re- constructing regularly missing seismic traces with a classifier- guided diffusion model,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, no. 5906914, 2024

    2024

  40. [48]

    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

    2020

  41. [49]

    Diffir: Efficient diffusion model for image restoration,

    B. Xia, Y . Zhang, S. Wang, Y . Wang, X. Wu, Y . Tian, W. Yang, and L. Van Gool, “Diffir: Efficient diffusion model for image restoration,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2023, pp. 13 095–13 105

    2023

  42. [50]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. u. Kaiser, and I. Polosukhin, “Attention is all you need,” inAdvances in Neural Information Processing Systems, vol. 30, 2017

    2017

  43. [51]

    Sparse self-attention transformer for image inpainting,

    W. Huang, Y . Deng, S. Hui, Y . Wu, S. Zhou, and J. Wang, “Sparse self-attention transformer for image inpainting,”Pattern Recognition, vol. 145, no. 109897, 2024

    2024

  44. [52]

    Rethinking space- time networks with improved memory coverage for efficient video object segmentation,

    H. K. Cheng, Y .-W. Tai, and C.-K. Tang, “Rethinking space- time networks with improved memory coverage for efficient video object segmentation,”Advances in neural information processing systems, vol. 34, pp. 11 781–11 794, 2021

    2021

  45. [53]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. Bovik, H. Sheikh, and E. Simoncelli, “Image quality assessment: from error visibility to structural similarity,” IEEE Transactions on Image Processing, vol. 13, no. 4, pp. 600–612, 2004

    2004

  46. [54]

    Regeneration-constrained self-supervised seismic data interpo- lation,

    A. Song, C. Wang, C. Zhang, J. Zhang, D. Xiong, and X. Wei, “Regeneration-constrained self-supervised seismic data interpo- lation,”IEEE Transactions on Geoscience and Remote Sensing, vol. 61, no. 5901610, 2023

    2023

Pith tools

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