arxiv: 2605.09566 · v1 · submitted 2026-05-10 · 💻 cs.CV

Recognition: no theorem link

Dual-Path Hyperprior Informed Deep Unfolding Network for Image Compressive Sensing

Tianyi Lu , Wenxue Cui , Shaohui Liu

Authors on Pith no claims yet

Pith reviewed 2026-05-12 03:51 UTC · model grok-4.3

classification 💻 cs.CV

keywords compressive sensingdeep unfolding networkshyperpriordual-path architectureimage reconstructionattention mechanismsadaptive step size

0 comments

The pith

DPH-DUN splits compressive measurements into two subsets and routes them through a hyperprior-guided dual-path unfolding network to adapt step sizes and attention to local reconstruction difficulty.

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

Existing deep unfolding networks for compressive sensing process all measurements in one stream and apply the same operations to every image region, which limits how well they can use complementary information or handle textures of different complexity. The proposed DPH-DUN first partitions the measurements into two subsets, then runs one lightweight branch to extract hyperprior knowledge across domains and a second unfolding branch that uses that knowledge to steer each iteration. Inside the unfolding loop, a step-size network produces spatially varying update maps and two attention modules (hard and soft, both gradient-driven) emphasize regions that are harder to reconstruct. Experiments show the resulting reconstructions exceed those of prior CS methods on standard benchmarks.

Core claim

Partitioning measurements into double subsets enables a Deep Hyperprior Learning branch that produces collaborative domain-specific priors; these priors then drive a Hyperprior Informed Reconstruction branch whose gradient-descent step uses a Hyperprior Informed Step Size Generation network for adaptive per-pixel updates and whose proximal-mapping step uses gradient-based hard and soft attention to focus computation on difficult regions, yielding higher-fidelity image recovery than single-stream uniform DUNs.

What carries the argument

Dual-path architecture consisting of a hyperprior learning branch that generates multi-domain priors and a hyperprior-informed unfolding reconstruction branch whose step-size generator and gradient-driven hard/soft attention modules adapt the iteration to local image content.

Load-bearing premise

That splitting measurements into two subsets and feeding hyperprior signals from one path into the iterative updates of the other path will consistently overcome the information-sharing and texture-uniformity limits of single-stream networks across diverse sensing rates and image statistics.

What would settle it

A controlled experiment in which the same network is run once with the dual-subset partitioning and hyperprior modules enabled and once with them disabled (single stream, uniform step size and attention) on the same set of test images at multiple sampling ratios; if the disabled version matches or exceeds the enabled version on PSNR/SSIM, the core benefit claim is falsified.

Figures

Figures reproduced from arXiv: 2605.09566 by Shaohui Liu, Tianyi Lu, Wenxue Cui.

**Figure 2.** Figure 2: Overview of the proposed DPH-DUN. The upper part shows the dual-path unfolding framework, which consists: the Deep [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: Details of the Deep Hyperprior Learning (DHL), which [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: Details of the Hyperprior Informed Step-size Generation [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

**Figure 5.** Figure 5: Details of the Hyperprior Informed Hard Attention [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗

**Figure 6.** Figure 6: Details of the Hyperprior Informed Soft Attention [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗

**Figure 7.** Figure 7: Visual comparisons on Set11 images, with CS ratio [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗

**Figure 8.** Figure 8: Visualization of the reconstructed image, hard structural [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗

**Figure 10.** Figure 10: Comparisons of PSNR/SSIM performance on Set11 at [PITH_FULL_IMAGE:figures/full_fig_p008_10.png] view at source ↗

**Figure 11.** Figure 11: Visual comparisons on Brain dataset using Pseudo Ra [PITH_FULL_IMAGE:figures/full_fig_p008_11.png] view at source ↗

**Figure 9.** Figure 9: Comparisons of PSNR/SSIM performance on Set11 at [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗

read the original abstract

Recent Deep Unfolding Networks (DUNs) have significantly advanced Compressive Sensing (CS) by integrating iterative optimization with deep networks. However, existing DUNs still suffer from two challenges: 1) Reliance on a single measurement stream, which limits effective information interaction across distinct measurement subsets. 2) Uniform processing of all image regions, which overlooks varying reconstruction difficulties induced by diverse textures. To address these limitations, a novel Dual-Path Hyperprior Informed Deep Unfolding Network (DPH-DUN) is proposed, which partitions measurements into double subsets to enable hyperprior-guided reconstruction via a dual-path architecture. In the Deep Hyperprior Learning branch, a series of lightweight neural modules are designed to efficiently generate hyperprior knowledge of different domains, enabling collaborative guidance for the CS reconstruction. In the Hyperprior Informed Reconstruction branch, a deep unfolding framework with hyperprior guidance is constructed to iteratively refine reconstruction. Specifically, i) in the gradient descent step, a Hyperprior Informed Step Size Generation network is designed to dynamically generate spatially varying step maps, enabling adaptive fine-grained gradient updates. ii) In the proximal mapping step, two well-designed hyperprior informed attention mechanisms are introduced to dynamically focus on challenging regions via gradient-based hard and soft attentions, facilitating CS reconstruction accuracy. Extensive experiments demonstrate that the proposed DPH-DUN outperforms existing CS methods.

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.

Desk Editor's Note private letter to a colleague

DPH-DUN splits measurements into dual paths with hyperprior guidance for adaptive steps and attention in unfolding, but the outperformance claim needs checks on capacity-matched baselines.

read the letter

The main thing here is that the paper splits measurements into two subsets, runs a lightweight hyperprior branch to extract domain knowledge, and then feeds that into the unfolding loop to create spatially varying step sizes plus hard and soft attentions. This directly targets the single-stream and uniform-processing problems they identify in prior DUNs. The architecture is a clean combination of ideas that have not appeared together in the cited work. The design choices look deliberate: the hyperprior modules stay lightweight, the step-size network produces per-pixel maps, and the attentions focus on hard regions during proximal mapping. That part is well motivated and internally consistent. The experiments are presented as showing clear gains over existing CS methods, which is the central claim. The soft spot is that the abstract gives no sign the baselines were matched on parameter count, FLOPs, or training regime, nor that ablations isolate the dual-subset partitioning from the simple effect of adding a second path. If those controls are missing, the attribution to hyperprior guidance becomes less secure, exactly as the stress-test note flags. This paper is for people already working on deep unfolding for compressive sensing who want concrete architectural tweaks for adaptive reconstruction. A reader in that niche can extract useful module ideas even if they plan to rerun the comparisons themselves. It deserves peer review so the experimental fairness can be examined and any needed ablations requested.

Referee Report

2 major / 1 minor

Summary. The paper proposes a Dual-Path Hyperprior Informed Deep Unfolding Network (DPH-DUN) for image compressive sensing. It partitions measurements into double subsets to enable a dual-path architecture: a Deep Hyperprior Learning branch generates hyperprior knowledge across domains for collaborative guidance, while the Hyperprior Informed Reconstruction branch performs deep unfolding with two specific mechanisms—(i) a Hyperprior Informed Step Size Generation network that produces spatially varying step maps for adaptive gradient descent updates, and (ii) hyperprior-informed hard and soft attention mechanisms in the proximal mapping step to emphasize challenging regions. The central claim is that extensive experiments demonstrate DPH-DUN outperforms existing CS methods.

Significance. If the outperformance holds under controlled comparisons, the dual-path hyperprior design could meaningfully advance deep unfolding networks for CS by enabling subset-specific information interaction and texture-adaptive processing, which standard single-stream DUNs lack. The adaptive step-size and attention modules represent a concrete way to inject hyperprior guidance into the iterative optimization, potentially improving reconstruction accuracy on diverse image content.

major comments (2)

[Abstract] Abstract and experimental claims: The central claim that DPH-DUN outperforms existing CS methods is load-bearing, yet the manuscript provides no indication that baselines were matched for parameter count, FLOPs, or training protocol. The dual-path architecture inherently increases capacity relative to single-stream DUNs; without explicit controls or ablations isolating the dual-subset partitioning, it is impossible to attribute gains to the hyperprior mechanisms rather than model size.
[Method (Reconstruction branch)] Method section describing the Hyperprior Informed Reconstruction branch: The integration of the Hyperprior Informed Step Size Generation network into the gradient descent step and the hard/soft attentions into the proximal mapping step is described at a high level but lacks the explicit update equations or iteration pseudocode that would allow verification of how hyperprior knowledge is injected at each unfolding stage. This detail is necessary to assess whether the approach differs substantively from prior hyperprior or attention-augmented DUNs.

minor comments (1)

[Abstract] Abstract: Adding the specific datasets, CS sampling ratios, and quantitative metrics (e.g., PSNR/SSIM deltas) used in the 'extensive experiments' would strengthen the summary without lengthening it excessively.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. Below we provide point-by-point responses to the major comments and indicate the revisions we will make to strengthen the paper.

read point-by-point responses

Referee: [Abstract] Abstract and experimental claims: The central claim that DPH-DUN outperforms existing CS methods is load-bearing, yet the manuscript provides no indication that baselines were matched for parameter count, FLOPs, or training protocol. The dual-path architecture inherently increases capacity relative to single-stream DUNs; without explicit controls or ablations isolating the dual-subset partitioning, it is impossible to attribute gains to the hyperprior mechanisms rather than model size.

Authors: We acknowledge the importance of controlled comparisons to isolate the contributions of the proposed components. In the revised manuscript we will add a table reporting parameter counts, FLOPs, and training protocols for DPH-DUN and all baselines. We will also include new ablation experiments that separately evaluate the dual-subset partitioning and the hyperprior-informed modules, allowing clearer attribution of performance gains. revision: yes
Referee: [Method (Reconstruction branch)] Method section describing the Hyperprior Informed Reconstruction branch: The integration of the Hyperprior Informed Step Size Generation network into the gradient descent step and the hard/soft attentions into the proximal mapping step is described at a high level but lacks the explicit update equations or iteration pseudocode that would allow verification of how hyperprior knowledge is injected at each unfolding stage. This detail is necessary to assess whether the approach differs substantively from prior hyperprior or attention-augmented DUNs.

Authors: We agree that explicit equations and pseudocode will improve verifiability. In the revision we will insert the full update equations for the gradient-descent step (incorporating the spatially varying step-size map) and the proximal-mapping step (with gradient-based hard and soft attention), together with an iteration-level pseudocode that shows the precise injection of hyperprior information at each unfolding stage. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical architecture validated by experiments

full rationale

The paper proposes DPH-DUN, a dual-path deep unfolding network with hyperprior modules for image compressive sensing, and supports its claims solely through experimental comparisons. No derivation chain exists that reduces a claimed result to a self-defined quantity, a fitted parameter renamed as prediction, or a load-bearing self-citation. The architecture descriptions (step-size generation, attention mechanisms) are explicit design choices whose effectiveness is tested externally on benchmarks rather than assumed by construction. Self-citations, if any, are not invoked to prove uniqueness or forbid alternatives. The outperformance statement is an empirical observation, not a tautology.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 3 invented entities

The central claim rests on standard deep-unfolding assumptions plus several newly introduced neural modules whose effectiveness is asserted via experiments; no independent evidence outside the reported results is supplied in the abstract.

free parameters (1)

network weights and hyperparameters
All parameters of the hyperprior generation modules, step-size network, and attention mechanisms are learned from training data.

axioms (1)

domain assumption Iterative optimization algorithms for compressive sensing can be unfolded into a finite number of trainable layers that preserve convergence properties.
This is the foundational premise of all deep unfolding networks referenced in the abstract.

invented entities (3)

Dual-Path Hyperprior Informed Deep Unfolding Network (DPH-DUN) no independent evidence
purpose: To enable collaborative hyperprior-guided reconstruction by partitioning measurements into two paths.
New overall architecture introduced in the paper.
Hyperprior Informed Step Size Generation network no independent evidence
purpose: To dynamically produce spatially varying step maps for the gradient descent step.
New module described for the gradient descent update.
Hyperprior informed attention mechanisms (hard and soft) no independent evidence
purpose: To focus reconstruction effort on challenging image regions in the proximal mapping step.
New attention components introduced for the proximal step.

pith-pipeline@v0.9.0 · 5541 in / 1457 out tokens · 55949 ms · 2026-05-12T03:51:02.680033+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

52 extracted references · 52 canonical work pages

[1]

Compressed sensing with wavelet domain dependencies for coronary mri: a retrospective study.IEEE Transactions on Medical Imaging, 30(5):1090–1099, 2010

Mehmet Akc ¸akaya, Seunghoon Nam, Peng Hu, Mehdi H Moghari, Long H Ngo, Vahid Tarokh, Warren J Manning, and Reza Nezafat. Compressed sensing with wavelet domain dependencies for coronary mri: a retrospective study.IEEE Transactions on Medical Imaging, 30(5):1090–1099, 2010. 1

work page 2010
[2]

An introduction to compressive sampling.IEEE Signal Processing Maga- zine, 25(2):21–30, 2008

Emmanuel J Cand `es and Michael B Wakin. An introduction to compressive sampling.IEEE Signal Processing Maga- zine, 25(2):21–30, 2008. 1

work page 2008
[3]

Content-aware scalable deep com- pressed sensing.IEEE Transactions on Image Processing, 31:5412–5426, 2022

Bin Chen and Jian Zhang. Content-aware scalable deep com- pressed sensing.IEEE Transactions on Image Processing, 31:5412–5426, 2022. 3

work page 2022
[4]

Wenjun Chen, Chunling Yang, and Xin Yang. Fsoinet: Feature-space optimization-inspired network for image com- pressive sensing.Proceedings of the IEEE Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 2460–2464, 2022. 3, 6

work page 2022
[5]

Image compressed sensing using non-local neural network

Wenxue Cui, Shaohui Liu, Feng Jiang, and Debin Zhao. Image compressed sensing using non-local neural network. IEEE Transactions on Multimedia, 25:816–830, 2021. 2

work page 2021
[6]

Image compressed sensing using non-local neural network

Wenxue Cui, Shaohui Liu, Feng Jiang, and Debin Zhao. Image compressed sensing using non-local neural network. IEEE Transactions on Multimedia, 25:816–830, 2023. 2

work page 2023
[7]

Wenxue Cui, Xiaopeng Fan, Jian Zhang, and Debin Zhao. Deep unfolding network for image compressed sensing by content-adaptive gradient updating and deformation- invariant non-local modeling.IEEE Transactions on Mul- timedia, 26:4012–4027, 2024. 3

work page 2024
[8]

Image denoising by sparse 3-d transform- domain collaborative filtering.IEEE Transactions on Image Processing, 16(8):2080–2095, 2007

Kostadin Dabov, Alessandro Foi, Vladimir Katkovnik, and Karen Egiazarian. Image denoising by sparse 3-d transform- domain collaborative filtering.IEEE Transactions on Image Processing, 16(8):2080–2095, 2007. 2

work page 2080
[9]

Accelerat- ing the super-resolution convolutional neural network

Chao Dong, Chen Change Loy, and Xiaoou Tang. Accelerat- ing the super-resolution convolutional neural network. InEu- ropean conference on computer vision (ECCV), pages 391–

work page
[10]

Compressive sensing via nonlocal low-rank regu- larization.IEEE Transactions on Image Processing, 23(8): 3618–3632, 2014

Weisheng Dong, Guangming Shi, Xin Li, Yi Ma, and Feng Huang. Compressive sensing via nonlocal low-rank regu- larization.IEEE Transactions on Image Processing, 23(8): 3618–3632, 2014. 1

work page 2014
[11]

Compressed sensing.IEEE Transactions on Information Theory, 52(4):1289–1306, 2006

David L Donoho. Compressed sensing.IEEE Transactions on Information Theory, 52(4):1289–1306, 2006. 1

work page 2006
[12]

Single-pixel imaging via compressive sampling.IEEE Signal Processing Magazine, 25(2):83–91, 2008

Marco F Duarte, Mark A Davenport, Dharmpal Takhar, Ja- son N Laska, Ting Sun, Kevin F Kelly, and Richard G Bara- niuk. Single-pixel imaging via compressive sampling.IEEE Signal Processing Magazine, 25(2):83–91, 2008. 1

work page 2008
[13]

Image recovery using total variation minimization on compressive sensing

Assia El Mahdaoui, Abdeldjalil Ouahabi, and Mo- hamed Said Moulay. Image recovery using total variation minimization on compressive sensing. In2019 6th Interna- tional Conference on Image and Signal Processing and their Applications (ISPA), pages 1–5. IEEE, 2019. 1

work page 2019
[14]

From patch to pixel: A transformer-based hierarchi- cal framework for compressive image sensing.IEEE Trans- actions on Computational Imaging, 9:133–146, 2023

Hongping Gan, Minghe Shen, Yi Hua, Chunyan Ma, and Tao Zhang. From patch to pixel: A transformer-based hierarchi- cal framework for compressive image sensing.IEEE Trans- actions on Computational Imaging, 9:133–146, 2023. 2, 6

work page 2023
[15]

Nestd-net: Deep nesta-inspired unfolding network with dual-path deblocking structure for image compressive sensing.IEEE Transactions on Image Processing, 33:1923–1937, 2024

Hongping Gan, Zhen Guo, and Feng Liu. Nestd-net: Deep nesta-inspired unfolding network with dual-path deblocking structure for image compressive sensing.IEEE Transactions on Image Processing, 33:1923–1937, 2024. 2, 6

work page 1923
[16]

Cpp-net: Embracing multi- scale feature fusion into deep unfolding cp-ppa network for compressive sensing

Zhen Guo and Hongping Gan. Cpp-net: Embracing multi- scale feature fusion into deep unfolding cp-ppa network for compressive sensing. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 25086–25095, 2024. 3, 6

work page 2024
[17]

Usb-net: Unfolding split bregman method with multi-phase feature integration for compressive imaging.IEEE Transactions on Image Process- ing, 34:925 – 938, 2025

Zhen Guo and Hongping Gan. Usb-net: Unfolding split bregman method with multi-phase feature integration for compressive imaging.IEEE Transactions on Image Process- ing, 34:925 – 938, 2025. 6, 8

work page 2025
[18]

Sin- gle image super-resolution from transformed self-exemplars

Jia-Bin Huang, Abhishek Singh, and Narendra Ahuja. Sin- gle image super-resolution from transformed self-exemplars. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5197–5206,

work page
[19]

Csdet: A compressed sensing object detection ar- chitecture with lightweight networks.IEEE Transactions on Circuits and Systems for Video Technology, 35(3):2355 – 2368, 2024

Xiaoqian Huang, Yong Gong, Wenhao Wu, Saike Zhu, and Yi Zhao. Csdet: A compressed sensing object detection ar- chitecture with lightweight networks.IEEE Transactions on Circuits and Systems for Video Technology, 35(3):2355 – 2368, 2024. 2

work page 2024
[20]

Bayesian compres- sive sensing.IEEE Transactions on Signal Processing, 56 (6):2346–2356, 2008

Shihao Ji, Ya Xue, and Lawrence Carin. Bayesian compres- sive sensing.IEEE Transactions on Signal Processing, 56 (6):2346–2356, 2008. 1

work page 2008
[21]

Dynamic path- controllable deep unfolding network for compressive sens- ing.IEEE Transactions on Image Processing, 32:2202– 2214, 2023

Jian Zhang Jiechong Song, Bin Chen. Dynamic path- controllable deep unfolding network for compressive sens- ing.IEEE Transactions on Image Processing, 32:2202– 2214, 2023. 2, 3, 6

work page 2023
[22]

Reconnet: Non-iterative reconstruc- tion of images from compressively sensed measurements

Kuldeep Kulkarni, Suhas Lohit, Pavan Turaga, Ronan Ker- viche, and Amit Ashok. Reconnet: Non-iterative reconstruc- tion of images from compressively sensed measurements. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition (CVPR), pages 449–458, 2016. 2, 6

work page 2016
[23]

Sparse signal recovery via tree search matching pursuit.Journal of Communications and Networks, 18(5):699–712, 2016

Jaeseok Lee, Jun Won Choi, and Byonghyo Shim. Sparse signal recovery via tree search matching pursuit.Journal of Communications and Networks, 18(5):699–712, 2016. 1

work page 2016
[24]

D3c2-net: Dual-domain deep convolutional coding network for compressive sensing

Weiqi Li, Bin Chen, Shuai Liu, Shijie Zhao, Bowen Du, Yongbing Zhang, and Jian Zhang. D3c2-net: Dual-domain deep convolutional coding network for compressive sensing. IEEE Transactions on Circuits and Systems for Video Tech- nology, 34(10):9341 – 9355, 2024. 3

work page 2024
[25]

Nonconvex l1/2-regularized non- local self-similarity denoiser for compressive sensing based ct reconstruction.Journal of the Franklin Institute, 360(6): 4172–4195, 2023

Yunyi Li, Yiqiu Jiang, Hengmin Zhang, Jianxun Liu, Xian- gling Ding, and Guan Gui. Nonconvex l1/2-regularized non- local self-similarity denoiser for compressive sensing based ct reconstruction.Journal of the Franklin Institute, 360(6): 4172–4195, 2023. 2

work page 2023
[26]

Imaging with nature: Com- pressive imaging using a multiply scattering medium.Scien- tific Reports, 4(1):5552, 2014

Antoine Liutkus, David Martina, S ´ebastien Popoff, Gilles Chardon, Ori Katz, Geoffroy Lerosey, Sylvain Gigan, Lau- rent Daudet, and Igor Carron. Imaging with nature: Com- pressive imaging using a multiply scattering medium.Scien- tific Reports, 4(1):5552, 2014. 1

work page 2014
[27]

Convolutional neural networks for noniterative reconstruction of compressively sensed im- ages.IEEE Transactions on Computational Imaging, 4(3): 326–340, 2018

Suhas Lohit, Kuldeep Kulkarni, Ronan Kerviche, Pavan Turaga, and Amit Ashok. Convolutional neural networks for noniterative reconstruction of compressively sensed im- ages.IEEE Transactions on Computational Imaging, 4(3): 326–340, 2018. 2

work page 2018
[28]

Michael Lustig, David Donoho, and John M Pauly. Sparse mri: The application of compressed sensing for rapid mr imaging.Magnetic Resonance in Medicine: An Official Journal of the International Society for Magnetic Resonance in Medicine, 58(6):1182–1195, 2007. 1

work page 2007
[29]

Waterloo ex- ploration database: New challenges for image quality assess- ment models.IEEE Transactions on Image Processing, 26 (2):1004–1016, 2017

Kede Ma, Zhengfang Duanmu, Qingbo Wu, Zhou Wang, Hongwei Yong, Hongliang Li, and Lei Zhang. Waterloo ex- ploration database: New challenges for image quality assess- ment models.IEEE Transactions on Image Processing, 26 (2):1004–1016, 2017. 6

work page 2017
[30]

Improving synthesis and analysis prior blind compressed sensing with low-rank constraints for dy- namic mri reconstruction.Magnetic Resonance Imaging, 33 (1):174–179, 2015

Angshul Majumdar. Improving synthesis and analysis prior blind compressed sensing with low-rank constraints for dy- namic mri reconstruction.Magnetic Resonance Imaging, 33 (1):174–179, 2015. 1

work page 2015
[31]

Learning to invert: Signal recovery via deep convolutional networks

Ali Mousavi and Richard G Baraniuk. Learning to invert: Signal recovery via deep convolutional networks. InPro- ceedings of the IEEE Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 2272–2276, 2017. 2

work page 2017
[32]

Near-optimal com- pressed sensing guarantees for total variation minimization

Deanna Needell and Rachel Ward. Near-optimal com- pressed sensing guarantees for total variation minimization. IEEE Transactions on Image Processing, 22(10):3941–3949,

work page
[33]

A deep cascade of convolutional neural networks for dynamic mr image reconstruction.IEEE transactions on Medical Imaging, 37(2):491–503, 2017

Jo Schlemper, Jose Caballero, Joseph V Hajnal, Anthony N Price, and Daniel Rueckert. A deep cascade of convolutional neural networks for dynamic mr image reconstruction.IEEE transactions on Medical Imaging, 37(2):491–503, 2017. 8

work page 2017
[34]

Mtc-csnet: Marrying transformer and convolution for image compressed sensing.IEEE Trans- actions on Cybernetics, 54(9):4949–4961, 2024

Minghe Shen, Hongping Gan, Chunyan Ma, Chao Ning, Hongqi Li, and Feng Liu. Mtc-csnet: Marrying transformer and convolution for image compressed sensing.IEEE Trans- actions on Cybernetics, 54(9):4949–4961, 2024. 2

work page 2024
[35]

Image compressed sens- ing using convolutional neural network.IEEE Transactions on Image Processing, 29:375–388, 2019

W Shi, F Jiang, S Liu, and D Zhao. Image compressed sens- ing using convolutional neural network.IEEE Transactions on Image Processing, 29:375–388, 2019. 2, 6

work page 2019
[36]

Wuzhen Shi, Feng Jiang, Shaohui Liu, and Debin Zhao. Scalable convolutional neural network for image compressed sensing.Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 12282– 12291, 2019. 2

work page 2019
[37]

Single-pixel imaging using compressed sensing and wavelength-dependent scattering.Optics letters, 41(5):886– 889, 2016

Jaewook Shin, Bryan T Bosworth, and Mark A Fos- ter. Single-pixel imaging using compressed sensing and wavelength-dependent scattering.Optics letters, 41(5):886– 889, 2016. 1

work page 2016
[38]

Optimization-inspired cross-attention transformer for compressive sensing

Jiechong Song, Chong Mou, Shiqi Wang, Siwei Ma, and Jian Zhang. Optimization-inspired cross-attention transformer for compressive sensing. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 6174–6184, 2023. 6

work page 2023
[39]

Compressed sensing mri using a recursive di- lated network

Liyan Sun, Zhiwen Fan, Yue Huang, Xinghao Ding, and John Paisley. Compressed sensing mri using a recursive di- lated network. InProceedings of the AAAI Conference on Artificial Intelligence (AAAI), 2018. 8

work page 2018
[40]

Compressive sensing reconstruction algorithm using l1-norm minimization via l2-norm minimization.In- ternational Journal on Electrical Engineering & Informat- ics, 10(1), 2018

Koredianto Usman, Hendra Gunawan, and Andriyan Bayu Suksmono. Compressive sensing reconstruction algorithm using l1-norm minimization via l2-norm minimization.In- ternational Journal on Electrical Engineering & Informat- ics, 10(1), 2018. 1

work page 2018
[41]

Ufc-net: Unrolling fixed-point continuous network for deep compressive sens- ing

Xiaoyang Wang and Hongping Gan. Ufc-net: Unrolling fixed-point continuous network for deep compressive sens- ing. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 25149– 25159, 2024. 2, 3, 6

work page 2024
[42]

Adaptive measurement network for cs image reconstruction

Xuemei Xie, Yuxiang Wang, Guangming Shi, Chenye Wang, Jiang Du, and Xiao Han. Adaptive measurement network for cs image reconstruction. InComputer Vision: Second CCF Chinese Conference, CCCV 2017, Tianjin, China, October 11–14, 2017, Proceedings, Part II, pages 407–417. Springer,

work page 2017
[43]

Admm- csnet: A deep learning approach for image compressive sens- ing.IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(3):521–538, 2020

Yan Yang, Jian Sun, Huibin Li, and Zongben Xu. Admm- csnet: A deep learning approach for image compressive sens- ing.IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(3):521–538, 2020. 3

work page 2020
[44]

On sin- gle image scale-up using sparse-representations

Roman Zeyde, Michael Elad, and Matan Protter. On sin- gle image scale-up using sparse-representations. InInterna- tional conference on curves and surfaces, pages 711–730. Springer, 2010. 6

work page 2010
[45]

Ista-net: Interpretable optimization-inspired deep network for image compressive sensing

Jian Zhang and Bernard Ghanem. Ista-net: Interpretable optimization-inspired deep network for image compressive sensing. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 1828–1837, 2018. 3, 6, 8

work page 2018
[46]

High-throughput deep unfolding network for com- pressive sensing mri.IEEE Journal of Selected Topics in Signal Processing, 16(4):750–761, 2022

Jian Zhang, Zhenyu Zhang, Jingfen Xie, and Yongbing Zhang. High-throughput deep unfolding network for com- pressive sensing mri.IEEE Journal of Selected Topics in Signal Processing, 16(4):750–761, 2022. 1, 8

work page 2022
[47]

Locally similar sparsity-based hyperspec- tral compressive sensing using unmixing.IEEE Transactions on Computational Imaging, 2(2):86–100, 2016

Lei Zhang, Wei Wei, Yanning Zhang, Hangqi Yan, Fei Li, and Chunna Tian. Locally similar sparsity-based hyperspec- tral compressive sensing using unmixing.IEEE Transactions on Computational Imaging, 2(2):86–100, 2016. 1

work page 2016
[48]

Progres- sive content-aware coded hyperspectral snapshot compres- sive imaging.IEEE Transactions on Circuits and Systems for Video Technology, 34(11):10817 – 10830, 2024

Xuanyu Zhang, Bin Chen, Wenzhen Zou, Shuai Liu, Yong- bing Zhang, Ruiqin Xiong, and Jian Zhang. Progres- sive content-aware coded hyperspectral snapshot compres- sive imaging.IEEE Transactions on Circuits and Systems for Video Technology, 34(11):10817 – 10830, 2024. 1

work page 2024
[49]

Amp-net: Denoising-based deep unfolding for com- pressive image sensing.IEEE Transactions on Image Pro- cessing, 30:1487–1500, 2021

Zhonghao Zhang, Yipeng Liu, Jiani Liu, Fei Wen, and Ce Zhu. Amp-net: Denoising-based deep unfolding for com- pressive image sensing.IEEE Transactions on Image Pro- cessing, 30:1487–1500, 2021. 3, 6

work page 2021
[50]

Image compressive- sensing recovery using structured laplacian sparsity in dct domain and multi-hypothesis prediction

Chen Zhao, Siwei Ma, and Wen Gao. Image compressive- sensing recovery using structured laplacian sparsity in dct domain and multi-hypothesis prediction. In2014 IEEE Inter- national Conference on Multimedia and Expo (ICME), pages 1–6. IEEE, 2014. 1

work page 2014
[51]

Cascaded di- lated dense network with two-step data consistency for mri reconstruction.Advances in Neural Information Processing Systems, 32, 2019

Hao Zheng, Faming Fang, and Guixu Zhang. Cascaded di- lated dense network with two-step data consistency for mri reconstruction.Advances in Neural Information Processing Systems, 32, 2019. 8

work page 2019
[52]

Multi-channel deep networks for block-based image compressive sensing.IEEE Transactions on Multime- dia, 23:2627–2640, 2021

Siwang Zhou, Yan He, Yonghe Liu, Chengqing Li, and Jian- ming Zhang. Multi-channel deep networks for block-based image compressive sensing.IEEE Transactions on Multime- dia, 23:2627–2640, 2021. 2, 6

work page 2021