pith. sign in

arxiv: 2606.30030 · v1 · pith:FGREYZ7Hnew · submitted 2026-06-29 · 💻 cs.CV

CogSENet: Blind Image Deblurring with Blur-Conditioned Semantic Routing and Explicit Frequency Fusion

Pan Wang , Yihao Hu , Xiujin Liu This is my paper

Pith reviewed 2026-06-30 06:16 UTC · model grok-4.3

classification 💻 cs.CV
keywords blind image deblurringsemantic routingfrequency fusionstate space moduleblur field estimationimage restorationwavelet transform
0
0 comments X

The pith

CogSENet restores blurred images by routing semantic tokens and fusing frequencies, outperforming prior deblurring methods with fewer parameters.

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

The paper proposes a network called CogSENet that mimics an eagle's visual system to tackle blind image deblurring. It introduces modules for semantic-aware token regrouping, high-low frequency decomposition, and blur field estimation fused with semantic priors. This is meant to address the lack of semantic awareness in handling spatially varying real-world blur. A sympathetic reader would care because better deblurring could improve photo quality from cameras and phones without knowing the exact blur. The work also claims benefits for related tasks like dehazing and denoising.

Core claim

CogSENet is a dynamic semantic-aligned reconstruction framework that uses a Semantic-Driven State Space Module for prompt-conditioned long-range dependency modeling via semantic-aware token regrouping, a BiFreqFusionBlock that decomposes features into high and low frequencies with wavelet transforms for physically interpretable recovery, and estimation of a continuous Blur Field from the blur image fused with CLIP semantic priors to modulate latent features for adaptive restoration under non-uniform blur.

What carries the argument

The Semantic-Driven State Space Module (SDSSM) with differentiable semantic-aware token regrouping, combined with BiFreqFusionBlock (BFFB) using wavelet transforms and continuous Blur Field (CBF) estimation fused with CLIP priors.

If this is right

  • CogSENet achieves higher visual quality and structural fidelity than state-of-the-art deblurring methods.
  • It uses fewer parameters while doing so.
  • It performs well on dehazing, deraining, and denoising tasks as well.

Where Pith is reading between the lines

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

  • The approach may suggest that biologically inspired dynamic mechanisms can enhance other image restoration problems beyond deblurring.
  • Explicit frequency separation could be tested for improving interpretability in related computer vision tasks.
  • Combining state space models with semantic priors might reduce the need for large parameter counts in restoration networks.

Load-bearing premise

The eagle-inspired Semantic-Driven State Space Module, BiFreqFusionBlock, and continuous Blur Field estimation supply the semantic awareness and physically interpretable recovery needed to handle spatially varying real-world degradations better than existing methods.

What would settle it

A direct comparison on standard real-world deblurring benchmarks showing that CogSENet does not exceed current methods in PSNR, SSIM, or visual quality, or requires more parameters.

Figures

Figures reproduced from arXiv: 2606.30030 by Pan Wang, Xiujin Liu, Yihao Hu.

Figure 1
Figure 1. Figure 1: Biological inspiration of CogSENet. Emulating the eagle’s visual system: (1) Active saccadic scanning for semantic targets inspires our SDSSM to perform semantic￾grouped scanning; (2) Retinal differentiation motivates our BFFB to decouple high/low frequencies; (3) Continuous focal adaptation inspires our Continuous Blur Field to model spatially varying blur; (4) Memory recall inspires a frozen CLIP prior t… view at source ↗
Figure 2
Figure 2. Figure 2: Our proposed CogSENet. A U-shaped backbone is built with stacked CogSF blocks across multi-scale encoder-decoder stages. Each CogSF block integrates an SDSSM module for prompt-guided dynamic token routing and a BFFB core composed of HPB units for explicit high/low-frequency decoupling and fusion. At the deepest bottleneck, we inject a continuous blur field conditioned on frozen CLIP embeddings to modulate … view at source ↗
Figure 3
Figure 3. Figure 3: Visual comparison on the GoPro dataset. Compared with the methods in (b)–(f), the proposed method produces a deblurred image with clearer structures and finer details in (g), yielding results that are more faithful to the ground truth in (h). 1 × 10−7 . We use cosine annealing for learning-rate scheduling throughout. For cross-task training, we disable the blur-field branch. In our implementation, this bra… view at source ↗
Figure 4
Figure 4. Figure 4: Deblurred results on the RealBlur dataset. The characters and structural details in (b)–(f) are still blurry, while our method restores sharper characters and clearer structural details [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Region-aware token routing in SDSSM. Effects of SDSSM [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualization of Continuous Blur Field and Semantic Priors. (a) Input blurry image. (b) CLIP-guided attention map, highlighting semantically informative struc￾tures. (c) Estimated continuous blur field. (d) Corresponding blur magnitude map. Effectiveness of Continuous Blur Field and Semantic Priors. As quan￾titatively verified in Tab. 6, applying either the physical blur field or cognitive CLIP semantics i… view at source ↗
Figure 7
Figure 7. Figure 7: Failure analysis. 5 Conclusion In this paper, we propose an efficient region- and frequency-aware framework for image deblurring. We introduce the Semantic-Driven State-Space Module (SDSSM) for linear-complexity, content-adaptive long-range modeling, and the Bi-Frequency Fusion Block (BFFB) to explicitly decouple and refine high/low￾frequency components. To tackle spatially non-uniform degradations, we pre… view at source ↗
read the original abstract

Blind image deblurring demands the recovery of high-fidelity details and coherent structures from complex, unknown degradations. Current blind image deblurring methods struggle with real-world, spatially varying degradations, and lack the semantic awareness necessary to reliably differentiate valid textures from artifacts. To bridge this gap, we propose CogSENet, a dynamic, semantic-aligned reconstruction framework inspired by the eagle's visual system. By mimicking the eagle's active saccadic scanning, we devise a Semantic-Driven State Space Module (SDSSM) with semantic-aware token regrouping via differentiable routing, enabling prompt-conditioned long-range dependency modeling. To ensure physically interpretable recovery of textures and structures, a BiFreqFusionBlock (BFFB) mirrors functional differentiation of the eagle's retina by decomposing features into high and low frequencies using wavelet transforms. Finally, we estimate a continuous Blur Field (CBF) from blur image and fuse it with CLIP semantic priors to modulate the deepest latent features, emulating focal adaptation and enabling adaptive restoration under spatially non-uniform blur. Extensive experiments demonstrate that CogSENetoutperforms state-of-the-art deblurring methods in both visual quality and structural fidelity with fewer parameters, while also performing favorably on dehazing, deraining, and denoising tasks.

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

2 major / 1 minor

Summary. The paper proposes CogSENet, a bio-inspired blind image deblurring network that introduces a Semantic-Driven State Space Module (SDSSM) for differentiable semantic-aware token routing and long-range modeling, a BiFreqFusionBlock (BFFB) that uses wavelet transforms to separate and fuse high/low-frequency features, and a continuous Blur Field (CBF) estimator that conditions the deepest features with CLIP semantic priors. The central claim is that this architecture yields superior visual quality and structural fidelity on real-world spatially varying blur while using fewer parameters than prior art, and generalizes favorably to dehazing, deraining, and denoising.

Significance. If the reported gains are reproducible, the combination of semantic routing with explicit frequency decomposition and blur-field conditioning could provide a useful inductive bias for handling non-uniform degradations; the eagle-inspired framing is mainly motivational and does not itself constitute a technical contribution.

major comments (2)
  1. [Abstract] Abstract (and any experimental section): the claim that CogSENet 'outperforms state-of-the-art deblurring methods in both visual quality and structural fidelity with fewer parameters' is presented without any quantitative tables, PSNR/SSIM values, parameter counts, dataset names, or statistical significance tests, rendering the central empirical claim impossible to evaluate.
  2. [Abstract] Abstract: the descriptions of SDSSM (semantic-aware token regrouping), BFFB (wavelet-based frequency fusion), and CBF (continuous blur-field estimation fused with CLIP priors) contain no equations, pseudocode, or architectural diagrams; without these, it is impossible to verify whether the modules implement the claimed semantic routing or physically interpretable recovery, which are load-bearing for the novelty argument.
minor comments (1)
  1. [Abstract] Abstract: 'CogSENetoutperforms' is missing a space.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the abstract. We address the two major comments point-by-point below, clarifying the role of the abstract versus the full manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract (and any experimental section): the claim that CogSENet 'outperforms state-of-the-art deblurring methods in both visual quality and structural fidelity with fewer parameters' is presented without any quantitative tables, PSNR/SSIM values, parameter counts, dataset names, or statistical significance tests, rendering the central empirical claim impossible to evaluate.

    Authors: The abstract serves as a concise summary and does not contain tables or full statistics, which is standard. The full manuscript includes quantitative results with PSNR/SSIM values, parameter counts, dataset names (GoPro, RealBlur, etc.), and comparisons in the Experiments section (Tables 1–4). We will revise the abstract to include a brief mention of key metrics and datasets for improved clarity. revision: yes

  2. Referee: [Abstract] Abstract: the descriptions of SDSSM (semantic-aware token regrouping), BFFB (wavelet-based frequency fusion), and CBF (continuous blur-field estimation fused with CLIP priors) contain no equations, pseudocode, or architectural diagrams; without these, it is impossible to verify whether the modules implement the claimed semantic routing or physically interpretable recovery, which are load-bearing for the novelty argument.

    Authors: Abstract length constraints preclude equations, pseudocode, or diagrams. Full technical details, including equations for semantic routing in SDSSM, wavelet decomposition in BFFB, and CLIP-conditioned blur field estimation in CBF, along with architectural diagrams and pseudocode, are provided in Sections 3.2–3.4 of the manuscript. These sections substantiate the claimed mechanisms and novelty. revision: no

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The provided abstract and description contain no equations, derivations, predictions, or first-principles results. The paper introduces an architecture (SDSSM, BFFB, CBF) with bio-inspired modules and reports empirical outperformance on deblurring and related tasks. No load-bearing step reduces by construction to fitted inputs, self-citations, or renamed known results; the central claim rests on experimental validation of the proposed modules rather than any self-referential reduction. This is the expected outcome for an architecture paper without visible mathematical derivations.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no free parameters, axioms, or invented entities can be identified from the provided text.

pith-pipeline@v0.9.1-grok · 5758 in / 1148 out tokens · 41156 ms · 2026-06-30T06:16:59.628300+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

37 extracted references · 5 canonical work pages · 2 internal anchors

  1. [1]

    In: ECCV

    Chen, L., Chu, X., Zhang, X., Sun, J.: Simple baselines for image restoration. In: ECCV. pp. 17–33. Springer (2022)

    2022

  2. [2]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Cho, S.J., Ji, S.W., Hong, J.P., Jung, S.W., Ko, S.J.: Rethinking coarse-to-fine ap- proach in single image deblurring. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 4641–4650 (2021)

    2021

  3. [3]

    In: European Conference on Computer Vision

    Conde, M.V., Choi, U.J., Burchi, M., Timofte, R.: Swin2sr: Swinv2 transformer for compressed image super-resolution and restoration. In: European Conference on Computer Vision. pp. 669–687. Springer (2022)

    2022

  4. [4]

    In: Proceedings of the IEEE international conference on computer vision

    Dong, C., Deng, Y., Loy, C.C., Tang, X.: Compression artifacts reduction by a deep convolutional network. In: Proceedings of the IEEE international conference on computer vision. pp. 576–584 (2015)

    2015

  5. [5]

    In: European conference on computer vision

    Dong, C., Loy, C.C., He, K., Tang, X.: Learning a deep convolutional network for image super-resolution. In: European conference on computer vision. pp. 184–199. Springer (2014)

    2014

  6. [6]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Dong, J., Pan, J., Yang, Z., Tang, J.: Multi-scale residual low-pass filter network for image deblurring. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 12345–12354 (2023) 16 P. Wang et al

    2023

  7. [7]

    In: Proceedings of the AAAI conference on artificial intelligence

    Dong, Y., Liu, Y., Zhang, H., Chen, S., Qiao, Y.: Fd-gan: Generative adversarial networks with fusion-discriminator for single image dehazing. In: Proceedings of the AAAI conference on artificial intelligence. vol. 34, pp. 10729–10736 (2020)

    2020

  8. [8]

    IEEE transactions on pattern analysis and machine intelligence43(1), 33–47 (2019)

    Fan, Q., Chen, D., Yuan, L., Hua, G., Yu, N., Chen, B.: A general decoupled learn- ing framework for parameterized image operators. IEEE transactions on pattern analysis and machine intelligence43(1), 33–47 (2019)

    2019

  9. [9]

    arXiv preprint arXiv:2212.14052 , year=

    Fu, D.Y., Dao, T., Saab, K.K., Thomas, A.W., Rudra, A., Ré, C.: Hungry hun- gry hippos: Towards language modeling with state space models. arXiv preprint arXiv:2212.14052 (2022)

    work page arXiv 2022

  10. [10]

    IEEE Transactions on Image Processing 26(6), 2944–2956 (2017)

    Fu, X., Huang, J., Ding, X., Liao, Y., Paisley, J.: Clearing the skies: A deep network architecture for single-image rain removal. IEEE Transactions on Image Processing 26(6), 2944–2956 (2017)

    2017

  11. [11]

    IEEE transactions on neural networks and learning systems 31(6), 1794–1807 (2019)

    Fu, X., Liang, B., Huang, Y., Ding, X., Paisley, J.: Lightweight pyramid networks for image deraining. IEEE transactions on neural networks and learning systems 31(6), 1794–1807 (2019)

    2019

  12. [12]

    In: Proceedings of the 32nd ACM International Conference on Multimedia

    Gao,H.,Ma,B.,Zhang,Y.,Yang,J.,Yang,J.,Dang,D.:Learningenrichedfeatures via selective state spaces model for efficient image deblurring. In: Proceedings of the 32nd ACM International Conference on Multimedia. pp. 710–718 (2024)

    2024

  13. [13]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Gao, X., Qiu, T., Zhang, X., Bai, H., Liu, K., Huang, X., Wei, H., Zhang, G., Liu, H.: Efficient multi-scale network with learnable discrete wavelet transform for blind motion deblurring. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 2733–2742 (2024)

    2024

  14. [14]

    Efficiently Modeling Long Sequences with Structured State Spaces

    Gu, A., Goel, K., Ré, C.: Efficiently modeling long sequences with structured state spaces. arXiv preprint arXiv:2111.00396 (2021)

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  15. [15]

    In: European conference on computer vision

    Guo, H., Li, J., Dai, T., Ouyang, Z., Ren, X., Xia, S.T.: Mambair: A simple baseline for image restoration with state-space model. In: European conference on computer vision. pp. 222–241. Springer (2024)

    2024

  16. [16]

    In: European Conference on Computer Vision

    Huang, T., Pei, X., You, S., Wang, F., Qian, C., Xu, C.: Localmamba: Visual state space model with windowed selective scan. In: European Conference on Computer Vision. pp. 12–22. Springer (2024)

    2024

  17. [17]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Jiang, K., Wang, Z., Yi, P., Chen, C., Huang, B., Luo, Y., Ma, J., Jiang, J.: Multi- scale progressive fusion network for single image deraining. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 8346–8355 (2020)

    2020

  18. [18]

    In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition

    Kong, L., Dong, J., Ge, J., Li, M., Pan, J.: Efficient frequency domain-based trans- formers for high-quality image deblurring. In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition. pp. 5886–5895 (2023)

    2023

  19. [19]

    In: Proceedings of the computer vision and pattern recognition conference

    Kong, L., Dong, J., Tang, J., Yang, M.H., Pan, J.: Efficient visual state space model for image deblurring. In: Proceedings of the computer vision and pattern recognition conference. pp. 12710–12719 (2025)

    2025

  20. [20]

    IEEE Transactions on Image Processing28(1), 492–505 (2019).https://doi.org/10.1109/TIP.2018.2867951

    Li, B., Ren, W., Fu, D., Tao, D., Feng, D., Zeng, W., Wang, Z.: Benchmarking single-image dehazing and beyond. IEEE Transactions on Image Processing28(1), 492–505 (2019).https://doi.org/10.1109/TIP.2018.2867951

    work page doi:10.1109/tip.2018.2867951 2019

  21. [21]

    In: CVPR

    Li, B., Liu, X., Hu, P., Wu, Z., Lv, J., Peng, X.: All-in-one image restoration for unknown corruption. In: CVPR. pp. 17452–17462 (2022)

    2022

  22. [22]

    Liang, J., Cao, J., Sun, G., Zhang, K., Van Gool, L., Timofte, R.: Swinir: Image restorationusingswintransformer.In:ProceedingsoftheIEEE/CVFInternational ConferenceonComputerVision(ICCV)Workshops.pp.1833–1844(October2021)

  23. [23]

    Advances in neural information processing systems37, 103031–103063 (2024) CogSENet17

    Liu, Y., Tian, Y., Zhao, Y., Yu, H., Xie, L., Wang, Y., Ye, Q., Jiao, J., Liu, Y.: Vmamba: Visual state space model. Advances in neural information processing systems37, 103031–103063 (2024) CogSENet17

    2024

  24. [24]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Mao, X., Li, Q., Wang, Y.: Adarevd: Adaptive patch exiting reversible decoder pushes the limit of image deblurring. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 25681–25690 (June 2024)

    2024

  25. [25]

    In: Proceedings eighth IEEE international conference on com- puter vision

    Martin, D., Fowlkes, C., Tal, D., Malik, J.: A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics. In: Proceedings eighth IEEE international conference on com- puter vision. ICCV 2001. vol. 2, pp. 416–423. Ieee (2001)

    2001

  26. [26]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Nah, S., Hyun Kim, T., Mu Lee, K.: Deep multi-scale convolutional neural network for dynamic scene deblurring. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 3883–3891 (2017)

    2017

  27. [27]

    In: European conference on computer vision

    Rim, J., Lee, H., Won, J., Cho, S.: Real-world blur dataset for learning and bench- marking deblurring algorithms. In: European conference on computer vision. pp. 184–201. Springer (2020)

    2020

  28. [28]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Shen, Z., Wang, W., Lu, X., Shen, J., Ling, H., Xu, T., Shao, L.: Human-aware motion deblurring. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 5572–5581 (2019)

    2019

  29. [29]

    Simplified State Space Layers for Sequence Modeling

    Smith, J.T., Warrington, A., Linderman, S.W.: Simplified state space layers for sequence modeling. arXiv preprint arXiv:2208.04933 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  30. [30]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Tao, X., Gao, H., Shen, X., Wang, J., Jia, J.: Scale-recurrent network for deep image deblurring. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 8174–8182 (2018)

    2018

  31. [31]

    Neural Networks121, 461–473 (2020)

    Tian, C., Xu, Y., Zuo, W.: Image denoising using deep cnn with batch renormal- ization. Neural Networks121, 461–473 (2020)

    2020

  32. [32]

    In: European conference on computer vision

    Tsai, F.J., Peng, Y.T., Lin, Y.Y., Tsai, C.C., Lin, C.W.: Stripformer: Strip trans- former for fast image deblurring. In: European conference on computer vision. pp. 146–162. Springer (2022)

    2022

  33. [33]

    Tsai, F.J., Peng, Y.T., Tsai, C.C., Lin, Y.Y., Lin, C.W.: Banet: A blur-aware atten- tionnetworkfordynamicscenedeblurring.IEEETransactionsonImageProcessing 31, 6789–6799 (2022)

    2022

  34. [34]

    IEEE Transactions on Pattern Analysis and Machine Intelligence45(11), 12978–12995 (2023).https://doi.org/10.1109/TPAMI.2022.3183612

    Xiao, J., Fu, X., Liu, A., Wu, F., Zha, Z.J.: Image de-raining transformer. IEEE Transactions on Pattern Analysis and Machine Intelligence45(11), 12978–12995 (2023).https://doi.org/10.1109/TPAMI.2022.3183612

    work page doi:10.1109/tpami.2022.3183612 2023

  35. [35]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Zamir, S.W., Arora, A., Khan, S., Hayat, M., Khan, F.S., Yang, M.H.: Restormer: Efficient transformer for high-resolution image restoration. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 5728–5739 (2022)

    2022

  36. [36]

    In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition

    Zamir, S.W., Arora, A., Khan, S., Hayat, M., Khan, F.S., Yang, M.H., Shao, L.: Multi-stage progressive image restoration. In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition. pp. 14821–14831 (2021)

    2021

  37. [37]

    IEEE transactions on image processing26(7), 3142–3155 (2017)

    Zhang, K., Zuo, W., Chen, Y., Meng, D., Zhang, L.: Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising. IEEE transactions on image processing26(7), 3142–3155 (2017)

    2017