Pith. sign in

REVIEW 4 major objections 6 minor 54 references

Burst Image Super-Resolution via Multi-Cross Attention Encoding and Multi-Scan State-Space Decoding

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read A burst super-resolution architecture that pairs overlapping cross-window and cross-frame attention with a multi-scan state-space decoder claims state-of-the-art results on the NTIRE2022 benchmarks.

desk verdict A plausible burst-SR architecture with a genuine synthetic SOTA claim, but the real-world SOTA claims in the abstract and Section 4.3 are contradicted by Table 1. read the letter →

arxiv 2505.19668 v1 pith:C56XASFT submitted 2025-05-26 cs.CV

classification cs.CV
keywords burstsuper-resolutionmulti-imagemulti-crossattentioncross-windowcross-framestate-spacemodelMambaRAWimages
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

Burst super-resolution aims to recover a high-resolution image from a short burst of noisy, slightly misaligned RAW frames, using the sub-pixel shifts between frames as extra information. This paper proposes an encoder-decoder network in which a Multi-Cross Attention feature extractor combines overlapping cross-window attention with cross-frame attention, and a Multi-Scan State-Space decoder fuses the aligned features while correcting residual misalignment left by deformable-convolution alignment. On the NTIRE2022 synthetic benchmark the paper reports 43.20 dB PSNR, beating the prior best SeBIR by 0.34 dB, and it reports the best SSIM and LPIPS on the real-world benchmark. The authors also evaluate on an ISO 12233 resolution chart with a SWIR camera and report higher resolvable line pairs per millimeter than Burstormer and BSRT. The significance, if the comparisons hold, is that combining windowed attention with linear-complexity state-space scanning is a practical route to extracting and fusing sub-pixel detail across frames.

What carries the argument

The paper's central objects are Multi-Cross Attention (MCA) and the Multi-Scan State-Space Module (MS-SSM). MCA combines overlapping Cross-Window Attention, in which the key/value windows are larger than the query window so attention reaches beyond the fixed window, with Cross-Frame Attention, a lightweight channel-compressed convolutional gating block that pools global frame information. MS-SSM is a four-direction scan of the flattened feature sequence inside a Residual Mamba Block, so long-range dependencies are modeled with linear complexity while a local convolution restores neighborhood regularity after the 1D scan. Together they carry the argument: CWA and CFA widen the receptive field during feature extraction, and MS-SSM plus CFA compensates for the limited receptive field of the DCN alignment in the fusion stage.

What would settle it

Retraining every compared method under identical settings (patch size 64, same epochs, same aligned-L1 loss, same data) and recomputing the real-world PSNR column would settle the claim; as printed, Table 1 already shows BurstM at 48.65 and AFCNet at 48.63, both above the paper's 48.60, so the asserted 'second in PSNR' ranking does not follow from the table.

Watch

Extended reading notes

Core claim

The central claim is that the standard burst-SR pipeline of feature extraction, flow-guided deformable alignment, fusion, and reconstruction can be improved by replacing the Swin Transformer stages with two purpose-built modules. In the encoder, Multi-Cross Attention (MCA) runs overlapping Cross-Window Attention and Cross-Frame Attention in parallel: the query uses a $P\times P$ window while keys and values use overlapping $P'\times P'$ windows, so each query token can attend to a wider neighborhood and to information from other frames. In the decoder, a Multi-Scan State-Space Module (MS-SSM) scans the fused features in four directions inside a Residual Mamba Block, and a Cross-Frame Attention block then re-weights the frame-wise information. The paper reports that this architecture achieves 43.20 PSNR, 0.972 SSIM, and 0.028 LPIPS on the synthetic NTIRE2022 dataset, ahead of all compared methods, and 48.60 PSNR, 0.986 SSIM, 0.021 LPIPS on the real-world dataset, where it says it is first in SSIM and LPIPS and second in PSNR. The ablation attributes 0.78 dB of the synthetic gain to MCA and a further 0.11 dB to the RMB fusion module.

Load-bearing premise

The reported state-of-the-art ranking assumes that all numbers in the comparison table are directly comparable, even though the baselines were trained with different patch sizes (80 versus 64) and some were retrained while others were taken from their original papers.

Editorial extensions

If this is right

  • On the synthetic NTIRE2022 burst benchmark, the method claims 43.20 dB PSNR, a 0.34 dB gain over the previous best SeBIR, together with the best SSIM and LPIPS among the compared methods.
  • On the real-world benchmark, it reports the best SSIM and LPIPS scores, reaching 48.60 dB PSNR while training at a smaller patch size than several baselines.
  • On the ISO 12233 chart experiments, it reports 66 LP/mm horizontal, 59 LP/mm vertical, and 65 LP/mm diagonal resolution, exceeding the BSRT and Burstormer reconstructions.
  • The ablation study reports that replacing the Swin extraction with MCA adds 0.78 dB PSNR on synthetic bursts, and adding the MS-SSM-based RMB adds a further 0.11 dB.
  • The design implies that retaining DCN-based alignment is not a liability when a multi-scan state-space fusion module is available to correct the residual misalignment.

Reading between the lines

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

  • The attention recipe may transfer to other multi-frame low-level vision tasks: any alignment-plus-fusion pipeline that suffers from fixed-window attention could use overlapping cross-window attention and cross-frame gating to pull sub-pixel information from neighboring frames.
  • Since the multi-scan SSM explicitly scans the dithering range of each pixel, it could be tested as a generic post-alignment error corrector, potentially reducing dependence on accurate optical flow estimates in video restoration.
  • The real-world PSNR column of Table 1 lists BurstM (48.65) and AFCNet (48.63) above the paper's 48.60, so the 'second in PSNR' statement is an editorial reading of the table rather than a direct numerical consequence.
  • The ISO 12233 result uses a short-wave infrared camera with 14-frame bursts and controlled 2-pixel dithering; those LP/mm numbers indicate fine-detail recovery in that specific setup and should not be treated as a direct RGB burst-SR measurement.
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 / 6 minor

Summary. The manuscript proposes a burst super-resolution network that combines a Multi-Cross Attention (MCA) encoder, built from overlapping cross-window attention and cross-frame attention, with a decoder based on a Multi-Scan State-Space Module (MS-SSM) and cross-frame attention, while retaining DCN-based alignment. The method is evaluated on the NTIRE2022 synthetic and real-world burst-SR benchmarks and on an ISO 12233 resolution chart. The central claims are that the method achieves state-of-the-art results on the synthetic dataset across PSNR, SSIM, and LPIPS, and that on the real-world dataset it achieves the highest SSIM and LPIPS with second-best PSNR.

Significance. If the reported synthetic result is reproducible, the proposed architecture is a credible integration of window attention, cross-frame attention, and state-space models for burst super-resolution, offering competitive performance at a moderate parameter count (12.60M). The synthetic comparison in Table 1 supports the claim of improved PSNR/SSIM/LPIPS over the listed baselines, including SeBIR. The real-world claims, however, are not supported by the table as presented, because the comparison mixes published patch-80 numbers with retrained patch-64 numbers and the text asserts rankings that the table does not show. The ISO 12233 experiment is a useful practical extension, but its protocol is not described in enough detail to verify. Overall, the work is of moderate significance for the burst-SR community, provided the comparison claims are corrected and the retraining protocol is documented.

major comments (4)
  1. [Table 1 caption, §4.3] The real-world ranking claims are not supported by Table 1. The text states that on the real-world dataset the method 'achieves the highest scores for SSIM and LPIPS, while ranking second in PSNR,' and the abstract repeats this. Sorting the real-world columns shows that Ours (PSNR 48.60, SSIM 0.986, LPIPS 0.021) is below BurstM at patch 64 (48.65 PSNR) and below several patch-80 entries (BurstM 49.12, SBFBurst 48.87, Burstormer 48.82, AFCNet 48.63). For SSIM, Ours 0.986 is below BurstM (0.987) and SBFBurst (0.987), and for LPIPS Ours 0.021 is tied with BSRT (0.021) and SeBIR (0.021), not strictly best. The claims are therefore only defensible if restricted to the 64-patch retrained models, but the text does not state that restriction. This is a load-bearing issue because the abstract, §4.3, and the conclusion all repeat the real-world state-of-the-art claim. The authors should either revise the claims to match the table or provide a consistent comparison protocol that justifies the stated ranking.
  2. [Table 1 caption / §4.2] The retraining protocol for the entries labeled '(64)' is not adequately specified. The caption says these values are obtained by retraining on the real-world dataset with the same 64-patch size as the proposed method, 'based on the open-sourced code from the paper,' but the manuscript does not report the retraining epochs, learning-rate schedule, loss function (aligned or not), data splits, or any other details needed to reproduce those numbers. Because the real-world comparison mixes published patch-80 numbers with retrained patch-64 numbers in a single table, the ranking of methods across these columns is not interpretable without knowing whether the retraining was faithful and comparable. The authors should either report full retraining details or present patch-80 and patch-64 results in separate, clearly labeled comparisons.
  3. [§4.5] The ablation study reports quantitative gains of 0.78 dB (MCA versus baseline) and 0.11 dB (full model versus MCA only) but does not provide a table with final PSNR, SSIM, or LPIPS values for the baseline, MCA-only, RMB-only, and full model. Only training curves are shown, and the sole numeric value, '≈43.07' for the RMB-only model, appears in the caption of Fig. 13. Without a numeric ablation table, the claimed contributions of the individual modules cannot be verified. Please add a table with final metrics and parameter counts for each ablated configuration.
  4. [§4.4] The ISO 12233 resolution-chart experiment does not describe a controlled comparison protocol. It states that low-resolution images were upsampled using 'several state-of-the-art BurstSR methods' but does not say whether those methods were retrained or fine-tuned on the same short-wave infrared data as the proposed method. The conclusion later asserts that 'the results, obtained using the same training data,' but this is not described in §4.4. Without a clear statement of how the competing models were adapted to the SWIR domain, the reported LP/mm differences cannot be attributed to the proposed architecture rather than to training conditions.
minor comments (6)
  1. [Header] The manuscript header reads 'Preprint submitted to Nuclear Physics B,' which is not an appropriate journal for a computer-vision paper; this appears to be a template error and should be corrected.
  2. [§3, §3.1] Equation numbering is inconsistent: Eq. (1) is used twice, first for Fd = E(F0) in §3 and again for the four equations inside the encoder description in §3.1. The equations in §3.1 should be renumbered sequentially.
  3. [Eq. (10)] The loss in Eq. (10) is written as L = ||ISR − IHR|| without specifying the norm; given the text mentions L1 loss, the equation should state the L1 norm explicitly and clarify whether the aligned L1 loss is used for real-world training.
  4. [§4.2] The patch-size description is confusing: 'For synthetic training, we used a patch size of 384 × 384 for high-resolution images. However, due to memory constraints, the patch size for low-resolution images was reduced to 64 × 64 during training on real-world data.' Please clarify which patch size applies to synthetic versus real-world training and the relationship between the HR and LR patch sizes.
  5. [§4.5] The ablation text says the orange line uses 'the proposed MCA modules in the decoder,' but MCA is described as an encoder module; this is likely a typo and should be corrected. Also, the baseline 'original BSRT' should be described precisely so the reader knows which configuration is used as the blue line.
  6. [§4.4] The conversion factor of 9.26 used to convert LW/PH to LP/mm is stated without derivation. Please provide the formula and the sensor parameters used, so the reported resolution values can be checked.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the method is an architecture proposal evaluated against external benchmarks, and the claimed rankings do not reduce by construction to the paper's own equations or citations.

full rationale

The paper's derivation chain is an architecture proposal plus empirical evaluation. The central claims (synthetic 43.20 PSNR, real-world LPIPS 0.021, etc.) are obtained by training the proposed MCA/MS-SSM network on NTIRE2022 data and comparing it with published baselines; there is no fitted parameter inserted into an equation that is then renamed a prediction, and no result is defined in terms of the target metric it claims to achieve. The equations in the paper (Eqs. 1-10) are forward-model definitions and the L1 loss; none of them encodes PSNR, SSIM, or LPIPS. Citations to MambaIR, VMamba, SwinIR, and similar works are used as component-design inspiration, not as load-bearing proof of this paper's SOTA claim. There is no self-citation chain and no imported uniqueness theorem. The skeptical observation that Table 1 does not support the statements 'highest scores for SSIM and LPIPS' and 'ranking second in PSNR' on the real-world dataset (e.g., BurstM and SBFBurst report higher PSNR/SSIM at the 80-patch setting) is a factual and benchmarking-consistency concern about mixed patch sizes, not a circularity: the claim, even if overstated, does not reduce by construction to a fitted input or to a self-cited result. Under the hard rules, that correctness risk does not raise the circularity score. The paper is self-contained against external benchmarks, so the honest finding is no significant circularity.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

The central empirical claim relies on a standard deep-learning scaffold: pretrained optical flow, deformable convolution, attention, and Mamba-style state-space blocks. The ledger lists the hand-chosen hyperparameters and domain assumptions that are load-bearing for the reported performance. No physically invented entities are introduced.

free parameters (8)
  • Overlap rate r in cross-window attention = not reported
    Controls the expanded key/value window size P' = P(1+r); chosen by hand and not ablated.
  • Query window size P = not reported
    Attention window size in CWA; chosen by hand.
  • Channel reduction factor beta in CFA = not reported
    Compresses channels in the cross-frame attention gating layers; chosen by hand.
  • Channel expansion factor lambda in RMB = not reported
    Expands feature channels in the residual Mamba block; chosen by hand.
  • Number of MS-SSM scan directions = 4
    Four fixed scan directions are used to cover pixel displacement; the count is not ablated over alternatives.
  • Learnable residual scales gamma and s' = learned during training
    Equations 7 and 9 introduce learnable scale factors for residual connections.
  • Feature embedding dimensions C and C' = not reported
    Network width is chosen by hand and not specified numerically in the text.
  • Training patch sizes = 384x384 HR synthetic, 64x64 LR real
    Patch size choice affects the comparison because some baselines were trained at patch size 80.
assumptions (5)
  • standard math Softmax attention with relative position bias (Eq. 3) is a suitable operation for fusing window and frame features.
    The paper builds on standard attention machinery without proving that it captures sub-pixel relationships.
  • domain assumption Pre-trained SpyNet optical flow and DCN offsets produce alignments accurate enough for the decoder to correct residuals.
    The whole decoder design assumes DCN-based alignment leaves only small residual misalignments that four scan directions can fix.
  • domain assumption Training on synthetic NTIRE bursts and fine-tuning on real-world bursts transfers to real captures.
    Standard practice in the field, but the gap between synthetic degradation and real sensor noise is not quantified.
  • domain assumption PWC-Net alignment of real-world ground truth does not bias the reported metrics.
    The real-world training and evaluation use PWC-Net to align LR and HR images; errors in this alignment affect both training and metric values.
  • ad hoc to paper Four fixed scan directions are sufficient to cover the sub-pixel displacement range in bursts.
    The paper asserts four scans without a derivation or ablation over the number or orientation of scans.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Burst Image Super-Resolution via Multi-Cross Attention Encoding and Multi-Scan State-Space Decoding." pith.science (2026). https://pith.science/paper/C56XASFT

@misc{pith2026250519668,
  author       = {Pith},
  title        = {Pith review of: Burst Image Super-Resolution via Multi-Cross Attention Encoding and Multi-Scan State-Space Decoding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/C56XASFT}},
  note         = {Machine review of arXiv:2505.19668}
}
read the original abstract

Multi-image super-resolution (MISR) can achieve higher image quality than single-image super-resolution (SISR) by aggregating sub-pixel information from multiple spatially shifted frames. Among MISR tasks, burst super-resolution (BurstSR) has gained significant attention due to its wide range of applications. Recent methods have increasingly adopted Transformers over convolutional neural networks (CNNs) in super-resolution tasks, due to their superior ability to capture both local and global context. However, most existing approaches still rely on fixed and narrow attention windows that restrict the perception of features beyond the local field. This limitation hampers alignment and feature aggregation, both of which are crucial for high-quality super-resolution. To address these limitations, we propose a novel feature extractor that incorporates two newly designed attention mechanisms: overlapping cross-window attention and cross-frame attention, enabling more precise and efficient extraction of sub-pixel information across multiple frames. Furthermore, we introduce a Multi-scan State-Space Module with the cross-frame attention mechanism to enhance feature aggregation. Extensive experiments on both synthetic and real-world benchmarks demonstrate the superiority of our approach. Additional evaluations on ISO 12233 resolution test charts further confirm its enhanced super-resolution performance.

Figures

Figures reproduced from arXiv: 2505.19668 by the authors.

Figure 1
Figure 1. Compared with the current SOTA method without DCN, our method shows [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. An overview of the proposed method. The network takes as input a burst of [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Details of the proposed Encoder. Parallel Cross-Window Attention (CWA) and [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Details of the proposed Decoder (a). The core pipeline is shown in (b). The [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: Visual comparison results on Synthetic Datasets [26]. The odd rows the [PITH_FULL_IMAGE:figures/full_fig_p029_5.png]
Figure 6
Figure 6. Figure 6: Visual comparison results on BurstSR Real World Datasets [26].The odd rows [PITH_FULL_IMAGE:figures/full_fig_p030_6.png]
Figure 7
Figure 7. Figure 7: Burst super-resolution results on the ISO 12233 resolution test chart (center). [PITH_FULL_IMAGE:figures/full_fig_p031_7.png]
Figure 8
Figure 8. Figure 8: Burst super-resolution results on the ISO 12233 resolution test chart (horizon [PITH_FULL_IMAGE:figures/full_fig_p031_8.png]
Figure 9
Figure 9. Figure 9: Burst super-resolution results on the ISO 12233 resolution test chart (vertical). [PITH_FULL_IMAGE:figures/full_fig_p032_9.png]
Figure 10
Figure 10. Figure 10: Burst super-resolution results on the ISO 12233 resolution test chart (45 [PITH_FULL_IMAGE:figures/full_fig_p032_10.png]
Figure 11
Figure 11. Figure 11: Experimental setup for evaluating super-resolution methods using the ISO [PITH_FULL_IMAGE:figures/full_fig_p032_11.png]
Figure 12
Figure 12. Figure 12: Comparison of training processes among the baseline method (Standard Swin [PITH_FULL_IMAGE:figures/full_fig_p033_12.png]
Figure 13
Figure 13. Figure 13: Comparison between the training processes of the model with only RMB and [PITH_FULL_IMAGE:figures/full_fig_p033_13.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

54 extracted references · 30 canonical work pages

  1. [1]

    J. Li, Z. Pei, W. Li, G. Gao, L. Wang, Y. Wang, T. Zeng, A system- atic survey of deep learning-based single-image super-resolution, ACM Computing Surveys 56 (10) (2024) 1–40

  2. [2]

    Dudhane, S

    A. Dudhane, S. W. Zamir, S. Khan, F. S. Khan, M.-H. Yang, Burstormer: Burst Image Restoration and Enhancement Transformer (2023). 21

  3. [3]

    W. Ye, B. Lin, J. Lao, Y. Liu, Z. Lin, Mra-idn: a lightweight super- resolution framework of remote sensing images based on multi-scale residual attention fusion mechanism, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (2024)

  4. [4]

    Kappeler, S

    A. Kappeler, S. Yoo, Q. Dai, A. K. Katsaggelos, Video Super-Resolution With Convolutional Neural Networks, IEEE Transactions on Computa- tional Imaging 2 (2) (2016) 109–122. doi:10.1109/TCI.2016.2532323

  5. [5]

    Z. Luo, L. Yu, X. Mo, Y. Li, L. Jia, H. Fan, J. Sun, S. Liu, EBSR: Feature Enhanced Burst Super-Resolution with Deformable Alignment, in: 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPR W), IEEE, 2021, pp. 471–478. doi: 10.1109/CVPRW53098.2021.00058

  6. [6]

    Z. Luo, Y. Li, S. Cheng, L. Yu, Q. Wu, Z. Wen, H. Fan, J. Sun, S. Liu, BSRT: Improving Burst Super-Resolution with Swin Transformer and Flow-Guided Deformable Alignment (2022)

  7. [7]

    A. N. Cotrim, G. Barbosa, C. A. N. Santos, H. Pedrini, Enhanced residual network for burst image super-resolution using simple base frame guidance, Image and Vision Computing 155 (2025) 105444. doi: 10.1016/j.imavis.2025.105444

  8. [8]

    Wronski, I

    B. Wronski, I. Garcia-Dorado, M. Ernst, D. Kelly, M. Krainin, C.-K. Liang, M. Levoy, P. Milanfar, Handheld multi-frame super-resolution, ACM Transactions on Graphics 38 (4) (2019) 1–18. doi:10.1145/ 3306346.3323024

Show all 54 references
  1. [9]

    Kawulok, P

    M. Kawulok, P. Benecki, S. Piechaczek, K. Hrynczenko, D. Kostrzewa, J. Nalepa, Deep Learning for Multiple-Image Super-Resolution, IEEE Geoscience and Remote Sensing Letters 17 (6) (2020) 1062–1066

  2. [10]

    Isobe, X

    T. Isobe, X. Jia, S. Gu, S. Li, S. Wang, Q. Tian, Video Super-Resolution with Recurrent Structure-Detail Network, arXiv:2008.00455 [cs] (Aug. 2020). doi:10.48550/arXiv.2008.00455

  3. [11]

    Rifat Arefin, V

    M. Rifat Arefin, V. Michalski, P.-L. St-Charles, A. Kalaitzis, S. Kim, S. E. Kahou, Y. Bengio, Multi-Image Super-Resolution for Remote Sens- ing using Deep Recurrent Networks, in: 2020 IEEE/CVF Conference on 22 Computer Vision and Pattern Recognition Workshops (CVPR W), IEEE, ...

  4. [12]

    Ballas, L

    N. Ballas, L. Yao, C. Pal, A. Courville, Delving Deeper into Convolu- tional Networks for Learning Video Representations, arXiv:1511.06432 [cs] (Mar. 2016). doi:10.48550/arXiv.1511.06432

  5. [13]

    A. B. Molini, D. Valsesia, G. Fracastoro, E. Magli, DeepSUM: Deep neu- ral network for Super-resolution of Unregistered Multitemporal images, IEEE Transactions on Geoscience and Remote Sensing 58 (5) (2020) 3644–3656, arXiv:1907.06490 [eess]. doi:10.1109/TGRS.2019.2959248

  6. [14]

    Dudhane, S

    A. Dudhane, S. W. Zamir, S. Khan, F. S. Khan, M.-H. Yang, Burst im- age restoration and enhancement, IEEE transactions on pattern analysis and machine intelligence (2024)

  7. [15]

    E. Kang, B. Lee, S. Im, K. H. Jin, Burstm: Deep burst multi-scale sr using fourier space with optical flow, in: European Conference on Computer Vision, Springer, 2024, pp. 459–477

  8. [16]

    D. Y. Fu, T. Dao, K. K. Saab, A. W. Thomas, A. Rudra, C. R´ e, Hungry Hungry Hippos: Towards Language Modeling with State Space Models, arXiv:2212.14052 [cs] (Apr. 2023). doi:10.48550/arXiv.2212.14052

  9. [17]

    A. Gu, T. Dao, Mamba: Linear-Time Sequence Modeling with Selective State Spaces, arXiv:2312.00752 [cs] (May 2024). doi:10.48550/arXiv. 2312.00752

  10. [18]

    A. Gu, K. Goel, C. R´ e, Efficiently Modeling Long Sequences with Struc- tured State Spaces, arXiv:2111.00396 [cs] (Aug. 2022). doi:10.48550/ arXiv.2111.00396

  11. [19]

    J. T. H. Smith, A. Warrington, S. W. Linderman, Simplified State Space Layers for Sequence Modeling, arXiv:2208.04933 [cs] (Mar. 2023). doi: 10.48550/arXiv.2208.04933

  12. [20]

    H. Guo, J. Li, T. Dai, Z. Ouyang, X. Ren, S.-T. Xia, MambaIR: A Simple Baseline for Image Restoration with State-Space Model, arXiv:2402.15648 [cs] (Oct. 2024). doi:10.48550/arXiv.2402.15648. 23

  13. [21]

    Simonyan, A

    K. Simonyan, A. Zisserman, Very Deep Convolutional Networks for Large-Scale Image Recognition, arXiv:1409.1556 [cs] (Apr. 2015). doi: 10.48550/arXiv.1409.1556

  14. [22]

    X. Wang, L. Xie, C. Dong, Y. Shan, Real-ESRGAN: Training Real- World Blind Super-Resolution with Pure Synthetic Data, in: 2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), IEEE, 2021, pp. 1905–1914. doi:10.1109/ICCVW54120. 2021.00217

  15. [23]

    N. Bose, N. Ahuja, Superresolution and noise filtering using moving least squares, IEEE Transactions on Image Processing 15 (8) (2006) 2239–2248. doi:10.1109/TIP.2006.877406

  16. [24]

    Fermuller, Robust Wavelet-Based Super-Resolution Recon- struction: Theory and Algorithm, IEEE Transactions on Pattern Anal- ysis and Machine Intelligence 31 (4) (2009) 649–660

    Hui Ji, C. Fermuller, Robust Wavelet-Based Super-Resolution Recon- struction: Theory and Algorithm, IEEE Transactions on Pattern Anal- ysis and Machine Intelligence 31 (4) (2009) 649–660. doi:10.1109/ TPAMI.2008.103

  17. [25]

    S. P. Belekos, N. P. Galatsanos, A. K. Katsaggelos, Maximum a Pos- teriori Video Super-Resolution Using a New Multichannel Image Prior, IEEE Transactions on Image Processing 19 (6) (2010) 1451–1464. doi: 10.1109/TIP.2010.2042115

  18. [26]

    G. Bhat, M. Danelljan, L. Van Gool, R. Timofte, Deep Burst Super- Resolution, in: 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2021, pp. 9205–9214

  19. [27]

    G. Bhat, M. Danelljan, F. Yu, L. Van Gool, R. Timofte, Deep Reparametrization of Multi-Frame Super-Resolution and Denoising, in: 2021 IEEE/CVF International Conference on Computer Vision (ICCV), IEEE, 2021, pp. 2440–2450

  20. [28]

    J. Dai, H. Qi, Y. Xiong, Y. Li, G. Zhang, H. Hu, Y. Wei, Deformable convolutional networks, in: Proceedings of the IEEE international con- ference on computer vision, 2017, pp. 764–773

  21. [29]

    Z. Liu, Y. Lin, Y. Cao, H. Hu, Y. Wei, Z. Zhang, S. Lin, B. Guo, Swin Transformer: Hierarchical Vision Transformer using Shifted Win- dows, in: 2021 IEEE/CVF International Conference on Computer Vi- sion (ICCV), IEEE, 2021, pp. 9992–10002. 24

  22. [30]

    Dosovitskiy, L

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, N. Houlsby, An Image is Worth 16x16 Words: Transform- ers for Image Recognition at Scale (2021)

  23. [31]

    Parmar, P

    N. Parmar, P. Ramachandran, A. Vaswani, I. Bello, A. Levskaya, J. Shlens, Stand-Alone Self-Attention in Vision Models, Advances in neural information processing systems 32 (2019)

  24. [32]

    Carion, F

    N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, S. Zagoruyko, End-to-end object detection with transformers, in: European conference on computer vision, Springer, 2020, pp. 213–229

  25. [33]

    Touvron, M

    H. Touvron, M. Cord, M. Douze, F. Massa, A. Sablayrolles, H. J´ egou, Training data-efficient image transformers & distillation through atten- tion, in: International conference on machine learning, PMLR, 2021, pp. 10347–10357

  26. [34]

    X. Chu, Z. Tian, Y. Wang, B. Zhang, H. Ren, X. Wei, H. Xia, C. Shen, Twins: Revisiting the Design of Spatial Attention in Vision Transform- ers, Advances in neural information processing systems 34 (2021) 9355– 9366

  27. [36]

    Huang, Y

    G. Huang, Y. Wang, K. Lv, H. Jiang, W. Huang, P. Qi, S. Song, Glance and Focus Networks for Dynamic Visual Recognition, arXiv:2201.03014 [cs] (Aug. 2022). doi:10.48550/arXiv.2201.03014

  28. [37]

    H. Cao, Y. Wang, J. Chen, D. Jiang, X. Zhang, Q. Tian, M. Wang, Swin-Unet: Unet-like Pure Transformer for Medical Image Segmenta- tion, arXiv:2105.05537 [eess] (May 2021). doi:10.48550/arXiv.2105. 05537

  29. [38]

    R. Liu, H. Deng, Y. Huang, X. Shi, L. Lu, W. Sun, X. Wang, J. Dai, H. Li, FuseFormer: Fusing Fine-Grained Information in Transformers 25 for Video Inpainting, in: 2021 IEEE/CVF International Conference on Computer Vision (ICCV), IEEE, 2021, pp. 14020–14029. doi:10.1109/ ICCV48...

  30. [39]

    Zhang, J

    K. Zhang, J. Peng, J. Fu, D. Liu, Exploiting Optical Flow Guid- ance for Transformer-Based Video Inpainting, IEEE Transactions on Pattern Analysis and Machine Intelligence 46 (7) (2024) 4977–4992, arXiv:2301.10048 [cs]. doi:10.1109/TPAMI.2024.3361010

  31. [40]

    Liang, J

    J. Liang, J. Cao, G. Sun, K. Zhang, L. Van Gool, R. Timofte, SwinIR: Image Restoration Using Swin Transformer, in: 2021 IEEE/CVF Inter- national Conference on Computer Vision Workshops (ICCVW), IEEE, 2021, pp. 1833–1844

  32. [41]

    H. Chen, Y. Wang, T. Guo, C. Xu, Y. Deng, Z. Liu, S. Ma, C. Xu, C. Xu, W. Gao, Pre-Trained Image Processing Transformer, in: 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2021, pp. 12294–12305. doi:10.1109/CVPR46437. 2021.01212

  33. [42]

    A. Gu, I. Johnson, K. Goel, K. Saab, T. Dao, A. Rudra, C. R´ e, Combin- ing Recurrent, Convolutional, and Continuous-time Models with Linear State-Space Layers, Advances in neural information processing systems 34 (2021) 572–585

  34. [43]

    L. Zhu, B. Liao, Q. Zhang, X. Wang, W. Liu, X. Wang, Vision Mamba: Efficient Visual Representation Learning with Bidirectional State Space Model, arXiv:2401.09417 [cs] (Nov. 2024). doi:10.48550/arXiv.2401. 09417

  35. [44]

    B. N. Patro, V. S. Agneeswaran, SiMBA: Simplified Mamba-Based Ar- chitecture for Vision and Multivariate Time series, arXiv:2403.15360 [cs] (Apr. 2024). doi:10.48550/arXiv.2403.15360

  36. [45]

    Ranjan, M

    A. Ranjan, M. J. Black, Optical Flow Estimation Using a Spatial Pyramid Network, in: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2017, pp. 2720–2729. doi: 10.1109/CVPR.2017.291. 26

  37. [46]

    Huang, H

    C. Huang, H. Nien, Multi chaotic systems based pixel shuffle for image encryption, Optics Communications 282 (11) (2009) 2123–2127. doi: 10.1016/j.optcom.2009.02.044

  38. [47]

    K. Li, Y. Wang, J. Zhang, P. Gao, G. Song, Y. Liu, H. Li, Y. Qiao, UniFormer: Unifying Convolution and Self-Attention for Visual Recog- nition, IEEE Transactions on Pattern Analysis and Machine Intelligence 45 (10) (2023) 12581–12600. doi:10.1109/TPAMI.2023.3282631. URL https:...

  39. [48]

    T. Xiao, M. Singh, E. Mintun, T. Darrell, P. Doll´ ar, R. Girshick, Early convolutions help transformers see better, Advances in neural informa- tion processing systems 34 (2021) 30392–30400

  40. [49]

    Y. Liu, Y. Tian, Y. Zhao, H. Yu, L. Xie, Y. Wang, Q. Ye, Y. Liu, VMamba: Visual State Space Model, arXiv:2401.10166 [cs] (May 2024). doi:10.48550/arXiv.2401.10166

  41. [50]

    Shazeer, GLU Variants Improve Transformer, arXiv:2002.05202 [cs] (Feb

    N. Shazeer, GLU Variants Improve Transformer, arXiv:2002.05202 [cs] (Feb. 2020). doi:10.48550/arXiv.2002.05202

  42. [51]

    G. Bhat, M. Danelljan, R. Timofte, Y. Cao, Y. Cao, M. Chen, X. Chen, S. Cheng, A. Dudhane, H. Fan, et al., Ntire 2022 burst super-resolution challenge, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 1041–1061

  43. [52]

    Loshchilov, F

    I. Loshchilov, F. Hutter, SGDR: STOCHASTIC GRADIENT DE- SCENT WITH W ARM RESTARTS, arXiv preprint arXiv:1608.03983 (2017)

  44. [53]

    D. Sun, X. Yang, M.-Y. Liu, J. Kautz, PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume, in: 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, IEEE, 2018, pp. 8934–8943. doi:10.1109/CVPR.2018.00931

  45. [54]

    Mehta, A

    N. Mehta, A. Dudhane, S. Murala, S. W. Zamir, S. Khan, F. S. Khan, Adaptive Feature Consolidation Network for Burst Super-Resolution, in: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPR W), IEEE, 2022, pp. 1278–1285. doi: 10.1109/CVPRW56347....

  46. [55]

    H. Liu, M. Shao, Y. Wan, Y. Liu, K. Shang, Sebir: Semantic-guided burst image restoration, Neural Networks 181 (2025) 106834. 28 BIPNet BSRT Burstormer OursBurstM SBFBurst Ground Truth Lr Figure 5: Visual comparison results on Synthetic Datasets [26]. The odd rows the Ground t...

Pith tools

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