pith. sign in

arxiv: 2606.24153 · v1 · pith:VEBXEDYCnew · submitted 2026-06-23 · 💻 cs.CV

Differential Unfolding: Efficient Unfolding Reconstruction for Video Snapshot Compressive Imaging

Muyuan Zhang , Jiancheng Zhang , Haijin Zeng , Yin-ping Zhao This is my paper

Pith reviewed 2026-06-26 01:35 UTC · model grok-4.3

classification 💻 cs.CV
keywords Differential UnfoldingVideo Snapshot Compressive ImagingDeep Unfolding NetworksDifferential Representation PriorEfficient ReconstructionHeterogeneous Architecture
0
0 comments X

The pith

Differential Unfolding replaces uniform repetition in deep unfolding networks with sparse high-parameter anchoring stages and lightweight differential evolution stages to improve the accuracy-efficiency trade-off for video snapshot compress

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

The paper claims that standard deep unfolding networks for video SCI waste computation by repeatedly applying identical high-complexity priors even after optimization trajectories have converged to near-static states. It introduces Differential Unfolding, a heterogeneous architecture that deploys expensive general stages only sparsely to create feature foundations and then uses cheap differential stages to propagate and refine those foundations across iterations. The differential stages rely on a Differential Representation Prior built from evolving attention maps and modulated feed-forward layers that explicitly model cross-stage feature changes. Experiments are said to show that this design reaches new state-of-the-art reconstruction quality while cutting computational cost substantially compared with uniform baselines.

Core claim

The central claim is that partitioning the unfolding process into structural anchoring (high-parameter general stages placed sparsely) and differential evolution (lightweight stages driven by the Differential Representation Prior) allows the network to focus computation on meaningful cross-stage variations instead of redundant static representations, thereby delivering a better accuracy-efficiency balance than uniform deep unfolding networks.

What carries the argument

The Differential Evolutionary Framework (DEF), which assigns complementary roles of structural anchoring to sparse high-parameter stages and differential evolution to lightweight stages equipped with Differential Representation Attention and Differential Modulated FFN inside the Differential Representation Prior.

If this is right

  • Video SCI reconstruction can reach higher fidelity at lower parameter and FLOPs budgets than uniform DUN baselines.
  • Cross-stage feature variations can be modeled explicitly with lightweight modules rather than by repeating full-capacity priors.
  • The same heterogeneous anchoring-plus-evolution pattern may apply to other iterative reconstruction tasks that exhibit convergence to static representations.

Where Pith is reading between the lines

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

  • If the differential mechanism proves robust, similar stage-wise differentiation could be tested in non-SCI domains such as iterative solvers for inverse problems in medical imaging.
  • The approach implicitly questions whether all stages in any unfolding network need identical capacity once early stages have stabilized the representation.
  • An open extension would be to learn the placement and depth of the sparse anchoring stages rather than fixing them manually.

Load-bearing premise

The premise that optimization trajectories in existing deep unfolding networks converge toward static states, producing representation stagnation that uniform repetition cannot avoid.

What would settle it

A controlled comparison in which a uniform deep unfolding network is trained on the same video SCI data and shows no measurable stagnation in feature updates across stages, or where the differential stages fail to improve the accuracy-compute curve.

Figures

Figures reproduced from arXiv: 2606.24153 by Haijin Zeng, Jiancheng Zhang, Muyuan Zhang, Yin-ping Zhao.

Figure 1
Figure 1. Figure 1: Comparison of PSNR-FLOPs￾Params on 6 grayscale datasets of the pro￾posed DU and competing methods. The ra￾dius indicates the number of parameters. and plug-and-play [40, 41] methods to modern end-to-end learning ap￾proaches [4, 5, 7, 8, 18, 22, 25–29]. Among these, deep unfolding meth￾ods [19,20,31–34,37,42] have emerged as a particularly promising direction. By embedding deep unfolding net￾works (DUNs) wi… view at source ↗
Figure 2
Figure 2. Figure 2: The compression process of VideoSCI and the reconstruction process of the unfolding framework. We present feature maps from some stages of the general un￾folding, along with the cosine similarities between the stages of general unfolding and differential unfolding. where ⊙ denotes the element-wise multiplication (Hadamard product), h and w represent spatial coordinates, and N ∈ R H×W represents noise intro… view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the Differential Evolutionary Framework (DEF). Our DEF starts from the general unfolding stages within one period, where the intermediate states are subsequently utilized in the next differential evolutionary stage for adaptive feature update. (a) The general unfolding stages. (b) The differential evolutionary stages. redundancy of repeatedly applying high-capacity operations but also allow… view at source ↗
Figure 4
Figure 4. Figure 4: (a) The backbone structure of DRP. (b) and (c) are details of DRA and DM￾FFN. differential attention by comparing the current feature representation with the intermediate state from the previous stage. The core idea behind DRA is that attention scores corresponding to similar regions are attenuated, while those exhibiting greater discrepancies between stages are amplified. This attention mechanism highligh… view at source ↗
Figure 5
Figure 5. Figure 5: Visual results of competitive methods on grayscale video frames. Experiment Setting. The proposed method is implemented using PyTorch and trained on a single RTX PRO 6000 GPU. Following previous works, our models are trained in the DAVIS2017 dataset [21] with the same data aug￾mentation in [25]. To verify model performance, we conduct experiments on six grayscale/color benchmark videos with a resolution of… view at source ↗
Figure 6
Figure 6. Figure 6: Visual results of competitive methods on color video frames. 4.4 Results on Real Captured Video Due to the presence of more noise in real captured videos, the reconstruction task becomes more challenging. We provide the results of DU compared with GAP-TV [38], BIRNAT [8], PnP-FFDNet [40], STFormer [26], and HiSViT [28] on real datasets [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Reconstructed results of real captured Duonimo and WaterBallon [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
read the original abstract

While Deep Unfolding Networks (DUNs) dominate video Snapshot Compressive Imaging (SCI), they remain constrained by a uniform design philosophy. Existing methods repeatedly stack high-complexity priors with identical structures, ignoring the fact that optimization trajectories converge toward static states. This results in representation stagnation, where high-cost computations are wasted on minimal feature updates. To address this inefficiency, we present Differential Unfolding (DU), a heterogeneous framework that replaces uniform repetition with dynamic evolution. Central to DU is the Differential Evolutionary Framework (DEF), which partitions the unfolding process into two complementary roles: structural anchoring and differential evolution. In this scheme, high-parameter general stages are sparsely deployed to generate high-fidelity feature foundations. Complementing these, lightweight differential stages employ a Differential Representation Prior (DRP) to propagate and refine these foundational features through a differential mechanism. By integrating Differential Representation Attention (DRA) for evolving attention maps and a Differential Modulated FFN (DM-FFN) for feature rectification, DRP effectively models cross-stage variations with minimal overhead. By focusing computational resources on dynamic evolution rather than static redundancy, DU achieves a superior trade-off between accuracy and efficiency. Extensive experiments verify that our method establishes new state-of-the-art results while significantly slashing computational overhead. https://github.com/Muyuan-Zhang/DU

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

3 major / 2 minor

Summary. The paper proposes Differential Unfolding (DU) for video Snapshot Compressive Imaging (SCI), replacing the uniform stacking of identical high-complexity priors in Deep Unfolding Networks (DUNs) with a heterogeneous Differential Evolutionary Framework (DEF). This partitions unfolding into sparse high-parameter anchoring stages and lightweight differential stages that use a Differential Representation Prior (DRP) incorporating Differential Representation Attention (DRA) and Differential Modulated FFN (DM-FFN) to model cross-stage variations. The central claim is that this addresses representation stagnation in optimization trajectories, yielding new state-of-the-art accuracy with substantially lower computational overhead, as verified by experiments; code is released at https://github.com/Muyuan-Zhang/DU.

Significance. If the accuracy-efficiency gains hold under rigorous verification, the work could meaningfully advance efficient DUN designs for SCI by shifting compute from redundant static stages to dynamic evolution. The open-source implementation supports reproducibility, which strengthens the contribution.

major comments (3)
  1. [Abstract / §1] Abstract and §1 (motivation): The claim that 'optimization trajectories converge toward static states' producing 'representation stagnation' is load-bearing for the entire DEF design (sparse anchoring + DRP/DRA/DM-FFN), yet no supporting measurement—such as per-stage feature delta norms, gradient magnitudes, or activation change statistics across uniform DUN baselines—is provided. Without this, the differential mechanism risks being an architectural search artifact rather than a principled response to observed stagnation.
  2. [§3] §3 (DEF and DRP): The differential mechanism is asserted to 'effectively model cross-stage variations with minimal overhead,' but the manuscript does not derive or bound the complexity reduction relative to uniform repetition (e.g., no FLOPs or parameter comparison that isolates the effect of DRP versus simply using fewer total stages). This leaves the efficiency claim unanchored.
  3. [Experiments] Experiments section (SOTA claims): The abstract states 'new state-of-the-art results while significantly slashing computational overhead,' but without reported error bars, multiple random seeds, or explicit baseline re-implementations with identical training protocols, it is impossible to assess whether the reported gains exceed typical variance in SCI reconstruction benchmarks.
minor comments (2)
  1. [§3] Notation for the new modules (DRP, DRA, DM-FFN) should be introduced with explicit equations in §3 before their use in the overall architecture diagram.
  2. The GitHub link is provided; confirming that the released code reproduces the exact tables and figures would strengthen the submission.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the presentation of our contributions. We address each major comment below and indicate planned revisions to the manuscript.

read point-by-point responses

  1. Referee: [Abstract / §1] Abstract and §1 (motivation): The claim that 'optimization trajectories converge toward static states' producing 'representation stagnation' is load-bearing for the entire DEF design (sparse anchoring + DRP/DRA/DM-FFN), yet no supporting measurement—such as per-stage feature delta norms, gradient magnitudes, or activation change statistics across uniform DUN baselines—is provided. Without this, the differential mechanism risks being an architectural search artifact rather than a principled response to observed stagnation.

    Authors: We agree that quantitative support for the stagnation phenomenon would strengthen the motivation. In the revised manuscript we will add a short analysis subsection (or appendix) reporting per-stage feature delta norms (L2) and activation change statistics computed on a representative uniform DUN baseline, confirming the observed convergence toward static states. This will directly tie the DEF design to the measured behavior rather than leaving it as an unverified premise. revision: yes

  2. Referee: [§3] §3 (DEF and DRP): The differential mechanism is asserted to 'effectively model cross-stage variations with minimal overhead,' but the manuscript does not derive or bound the complexity reduction relative to uniform repetition (e.g., no FLOPs or parameter comparison that isolates the effect of DRP versus simply using fewer total stages). This leaves the efficiency claim unanchored.

    Authors: The efficiency advantage arises because high-parameter anchoring stages are used only sparsely while the majority of stages employ the lightweight DRP. Although overall FLOPs and parameter counts are reported in the experiments, we acknowledge that an explicit isolation (e.g., a controlled comparison of uniform vs. differential designs at matched total complexity) is missing. We will add such a comparison table and brief derivation in §3 of the revision to anchor the overhead reduction claim. revision: yes

  3. Referee: [Experiments] Experiments section (SOTA claims): The abstract states 'new state-of-the-art results while significantly slashing computational overhead,' but without reported error bars, multiple random seeds, or explicit baseline re-implementations with identical training protocols, it is impossible to assess whether the reported gains exceed typical variance in SCI reconstruction benchmarks.

    Authors: Our reported numbers follow the evaluation protocol common in the SCI literature (single training run per method, averaged over test sequences). To address the concern we will (i) rerun the proposed method and key baselines with three random seeds and report mean ± std, and (ii) explicitly state in the revision which baselines were re-trained under identical protocols versus taken from original papers. This will allow readers to judge whether gains exceed typical variance. revision: partial

Circularity Check

0 steps flagged

No circularity: structural redesign with independent premise

full rationale

The paper's central claim is a heterogeneous unfolding architecture (DEF with DRP/DRA/DM-FFN) motivated by an unverified premise about stagnation in uniform DUNs. No equations, parameter fits, or self-citations are shown that reduce the claimed accuracy-efficiency gains to the inputs by construction. The derivation chain consists of a design choice justified by the premise rather than any self-referential reduction, making the framework self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 4 invented entities

The central claim rests on the domain assumption that uniform DUNs suffer from representation stagnation due to converging trajectories, plus several newly introduced components whose value is demonstrated only within the proposed system.

axioms (1)
  • domain assumption Optimization trajectories in deep unfolding networks converge toward static states, causing representation stagnation when identical high-complexity priors are repeated.
    Explicitly stated in the abstract as the ignored fact that motivates the new design.
invented entities (4)
  • Differential Evolutionary Framework (DEF) no independent evidence
    purpose: Partitions the unfolding process into structural anchoring and differential evolution roles.
    Newly introduced partitioning scheme with no independent evidence outside the paper.
  • Differential Representation Prior (DRP) no independent evidence
    purpose: Models cross-stage variations using differential mechanisms to propagate and refine features.
    Core new prior introduced by the paper.
  • Differential Representation Attention (DRA) no independent evidence
    purpose: Generates evolving attention maps within the differential stages.
    Component of DRP with no external validation shown.
  • Differential Modulated FFN (DM-FFN) no independent evidence
    purpose: Performs feature rectification in the differential stages.
    Component of DRP with no external validation shown.

pith-pipeline@v0.9.1-grok · 5771 in / 1438 out tokens · 24701 ms · 2026-06-26T01:35:33.857500+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

40 extracted references · 24 canonical work pages · 1 internal anchor

  1. [2]

    (eds.) Proceedings of the 38th In- ternational Conference on Machine Learning

    Bertasius, G., Wang, H., Torresani, L.: Is space-time attention all you need for video understanding? In: Meila, M., Zhang, T. (eds.) Proceedings of the 38th In- ternational Conference on Machine Learning. Proceedings of Machine Learning Research, vol. 139, pp. 813–824. PMLR (18–24 Jul 2021)

    2021

  2. [3]

    In: Koyejo, S., Mohamed, S., Agarwal, A., Belgrave, D., Cho, K., Oh, A

    Cai, Y., Lin, J., Wang, H., Yuan, X., Ding, H., Zhang, Y., Timofte, R., Gool, L.V.:Degradation-awareunfoldinghalf-shuffletransformerforspectralcompressive imaging. In: Koyejo, S., Mohamed, S., Agarwal, A., Belgrave, D., Cho, K., Oh, A. (eds.) Advances in Neural Information Processing Systems. vol. 35, pp. 37749– 37761. Curran Associates, Inc. (2022)

    2022

  3. [4]

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

    Cao, M., Wang, L., Wang, H., Wang, G., Yuan, X.: Towards real-time video com- pressive sensing on mobile devices. In: Proceedings of the 32nd ACM International Conference on Multimedia. p. 11080–11088. MM ’24, Association for Computing Machinery, New York, NY, USA (2024).https://doi.org/10.1145/3664647. 3680561

    work page doi:10.1145/3664647 2024

  4. [5]

    In: Leonardis, A., Ricci, E., Roth, S., Russakovsky, O., Sattler, T., Varol, G

    Cao, M., Wang, L., Wang, H., Yuan, X.: A simple low-bit quantization framework for video snapshot compressive imaging. In: Leonardis, A., Ricci, E., Roth, S., Russakovsky, O., Sattler, T., Varol, G. (eds.) Computer Vision – ECCV 2024. pp. 112–129. Springer Nature Switzerland, Cham (2025)

    2024

  5. [6]

    International Journal of Computer Vision pp

    Cao, M., Wang, L., Zhu, M., Yuan, X.: Hybrid cnn-transformer architecture for efficient large-scale video snapshot compressive imaging. International Journal of Computer Vision pp. 1–20 (2024)

    2024

  6. [7]

    2021 IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR) pp

    Cheng, Z., Chen, B., Liu, G., Zhang, H., Lu, R., Wang, Z., Yuan, X.: Memory- efficient network for large-scale video compressive sensing. 2021 IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR) pp. 16241–16250 (2021)

    2021

  7. [8]

    IEEE Transactions on Pattern Analysis and Machine Intelligence45(2), 2264–2281 (2023).https://doi.org/10

    Cheng, Z., Chen, B., Lu, R., Wang, Z., Zhang, H., Meng, Z., Yuan, X.: Recurrent neural networks for snapshot compressive imaging. IEEE Transactions on Pattern Analysis and Machine Intelligence45(2), 2264–2281 (2023).https://doi.org/10. 1109/TPAMI.2022.3161934

    arXiv 2023

  8. [9]

    In: International Conference on Learning Representations (2021)

    Chu, X., Tian, Z., Zhang, B., Wang, X., Shen, C.: Conditional positional encodings for vision transformers. In: International Conference on Learning Representations (2021)

    2021

  9. [10]

    Donoho,D.:Compressedsensing.IEEETransactionsonInformationTheory52(4), 1289–1306 (2006).https://doi.org/10.1109/TIT.2006.871582

    work page doi:10.1109/tit.2006.871582 2006

  10. [11]

    ICLR (2021)

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., Houlsby, N.: An image is worth 16x16 words: Transformers for image recognition at scale. ICLR (2021)

    2021

  11. [12]

    Nature516(7529), 74–77 (Dec 2014).https://doi.org/10.1038/nature14005

    Gao, L., Liang, J., Li, C., Wang, L.V.: Single-shot compressed ultrafast photog- raphy at one hundred billion frames per second. Nature516(7529), 74–77 (Dec 2014).https://doi.org/10.1038/nature14005

    work page doi:10.1038/nature14005 2014

  12. [13]

    In: 2011 International Conference on Computer Vision

    Hitomi, Y., Gu, J., Gupta, M., Mitsunaga, T., Nayar, S.K.: Video from a sin- gle coded exposure photograph using a learned over-complete dictionary. In: 2011 International Conference on Computer Vision. pp. 287–294 (2011).https: //doi.org/10.1109/ICCV.2011.6126254 16 M. Zhang et al

    work page doi:10.1109/iccv.2011.6126254 2011

  13. [14]

    CoRR abs/1412.6980(2014)

    Kingma, D.P., Ba, J.: Adam: A method for stochastic optimization. CoRR abs/1412.6980(2014)

    Pith/arXiv arXiv 2014

  14. [15]

    IEEE Transactions on Pattern Analysis and Machine Intelli- gence41(12), 2990–3006 (2019).https://doi.org/10.1109/TPAMI.2018.2873587

    Liu, Y., Yuan, X., Suo, J., Brady, D.J., Dai, Q.: Rank minimization for snapshot compressive imaging. IEEE Transactions on Pattern Analysis and Machine Intelli- gence41(12), 2990–3006 (2019).https://doi.org/10.1109/TPAMI.2018.2873587

    work page doi:10.1109/tpami.2018.2873587 2019

  15. [16]

    Optics Express21(9), 10526 (Apr 2013).https://doi.org/10.1364/oe.21.010526

    Llull, P., Liao, X., Yuan, X., Yang, J., Kittle, D., Carin, L., Sapiro, G., Brady, D.J.: Coded aperture compressive temporal imaging. Optics Express21(9), 10526 (Apr 2013).https://doi.org/10.1364/oe.21.010526

    work page doi:10.1364/oe.21.010526 2013

  16. [17]

    arXiv: Learning (2016)

    Loshchilov, I., Hutter, F.: Sgdr: Stochastic gradient descent with warm restarts. arXiv: Learning (2016)

    2016

  17. [18]

    Lu,R.,Cheng,Z.,Chen,B.,Yuan,X.:Motion-awaredynamicgraphneuralnetwork forvideocompressivesensing.IEEETransactionsonPatternAnalysisandMachine Intelligence46(12), 7850–7866 (2024).https://doi.org/10.1109/TPAMI.2024. 3395804

    work page doi:10.1109/tpami.2024 2024

  18. [19]

    In: 2019 IEEE/CVF International Conference on Computer Vision (ICCV)

    Ma, J., Liu, X.Y., Shou, Z., Yuan, X.: Deep tensor admm-net for snapshot compres- sive imaging. In: 2019 IEEE/CVF International Conference on Computer Vision (ICCV). pp. 10222–10231 (2019).https://doi.org/10.1109/ICCV.2019.01032

    work page doi:10.1109/iccv.2019.01032 2019

  19. [20]

    International Journal of Computer Vision131(11), 2933–2958 (Nov 2023).https: //doi.org/10.1007/s11263-023-01844-4

    Meng, Z., Yuan, X., Jalali, S.: Deep Unfolding for Snapshot Compressive Imaging. International Journal of Computer Vision131(11), 2933–2958 (Nov 2023).https: //doi.org/10.1007/s11263-023-01844-4

    work page doi:10.1007/s11263-023-01844-4 2023

  20. [21]

    ArXiv abs/1704.00675(2017)

    Pont-Tuset, J., Perazzi, F., Caelles, S., Arbeláez, P., Sorkine-Hornung, A., Gool, L.V.: The 2017 davis challenge on video object segmentation. ArXiv abs/1704.00675(2017)

    Pith/arXiv arXiv 2017

  21. [22]

    APL Photonics5(3), 030801 (Mar 2020).https://doi.org/10.1063/1.5140721

    Qiao, M., Meng, Z., Ma, J., Yuan, X.: Deep learning for video compressive sensing. APL Photonics5(3), 030801 (Mar 2020).https://doi.org/10.1063/1.5140721

    work page doi:10.1063/1.5140721 2020

  22. [23]

    In: CVPR 2011

    Reddy, D., Veeraraghavan, A., Chellappa, R.: P2c2: Programmable pixel com- pressive camera for high speed imaging. In: CVPR 2011. pp. 329–336 (2011). https://doi.org/10.1109/CVPR.2011.5995542

    work page doi:10.1109/cvpr.2011.5995542 2011

  23. [24]

    Sun, Y., Yuan, X., Pang, S.: Compressive high-speed stereo imaging. Opt. Express 25(15), 18182–18190 (Jul 2017).https://doi.org/10.1364/OE.25.018182

    work page doi:10.1364/oe.25.018182 2017

  24. [25]

    In: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Wang, L., Cao, M., Yuan, X.: Efficientsci: Densely connected network with space- time factorization for large-scale video snapshot compressive imaging. In: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 18477–18486 (2023).https://doi.org/10.1109/CVPR52729.2023.01772

    work page doi:10.1109/cvpr52729.2023.01772 2023

  25. [27]

    doi: 10.1109/ICCV51070.2023.00649

    Wang, P., Wang, L., Yuan, X.: Deep optics for video snapshot compressive imaging. In: 2023 IEEE/CVF International Conference on Computer Vision (ICCV). pp. 10612–10622 (2023).https://doi.org/10.1109/ICCV51070.2023.00977

    work page doi:10.1109/iccv51070.2023.00977 2023

  26. [28]

    In: Leonardis, A., Ricci, E., Roth, S., Rus- sakovsky, O., Sattler, T., Varol, G

    Wang, P., Zhang, Y., Wang, L., Yuan, X.: Hierarchical separable video transformer for snapshot compressive imaging. In: Leonardis, A., Ricci, E., Roth, S., Rus- sakovsky, O., Sattler, T., Varol, G. (eds.) Computer Vision – ECCV 2024. pp. 104–122. Springer Nature Switzerland, Cham (2025)

    2024

  27. [29]

    2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) pp

    Wang, Z., Zhang, H., Cheng, Z., Chen, B., Yuan, X.: Metasci: Scalable and adap- tive reconstruction for video compressive sensing. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) pp. 2083–2092 (2021)

    2021

  28. [30]

    Image quality assessment: from error visibility to structural similarity

    Wang, Z., Bovik, A., Sheikh, H., Simoncelli, E.: Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing 13(4), 600–612 (2004).https://doi.org/10.1109/TIP.2003.819861 Differential Unfolding 17

    work page doi:10.1109/tip.2003.819861 2004

  29. [31]

    In: 2021 IEEE/CVF International Conference on Computer Vision (ICCV)

    Wt, Z., Zhangt, J., Mou, C.: Dense deep unfolding network with 3d-cnn prior for snapshot compressive imaging. In: 2021 IEEE/CVF International Conference on Computer Vision (ICCV). pp. 4872–4881 (2021).https://doi.org/10.1109/ ICCV48922.2021.00485

    arXiv 2021

  30. [32]

    Induced and reduced unbounded operator algebras

    Wu, Z., Zhang, Z., Song, J., Zhang, M.: Spatial-temporal synergic prior driven unfolding network for snapshot compressive imaging. In: 2021 IEEE International Conference on Multimedia and Expo (ICME). pp. 1–6 (2021).https://doi.org/ 10.1109/ICME51207.2021.9428320

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1109/icme51207.2021.9428320 2021

  31. [33]

    Xie, S., Huang, Y., Qu, G., Ge, Y.: Sampling-priors-augmented deep unfolding net- workforrobustvideocompressivesensing.JournaloftheFranklinInstitute362(3), 107545 (2025).https://doi.org/https://doi.org/10.1016/j.jfranklin.2025. 107545

    work page doi:10.1016/j.jfranklin.2025 2025

  32. [34]

    arXiv e-prints arXiv:2201.10419 (Jan 2022)

    Yang, C., Zhang, S., Yuan, X.: Ensemble learning priors unfolding for scal- able Snapshot Compressive Sensing. arXiv e-prints arXiv:2201.10419 (Jan 2022). https://doi.org/10.48550/arXiv.2201.10419

    work page doi:10.48550/arxiv.2201.10419 2022

  33. [35]

    IEEE Transactions on Image Processing24(1), 106–119 (2015).https://doi.org/10

    Yang, J., Liao, X., Yuan, X., Llull, P., Brady, D.J., Sapiro, G., Carin, L.: Com- pressive sensing by learning a gaussian mixture model from measurements. IEEE Transactions on Image Processing24(1), 106–119 (2015).https://doi.org/10. 1109/TIP.2014.2365720

    arXiv 2015

  34. [36]

    IEEE Transactions on Image Pro- cessing23(11), 4863–4878 (2014).https://doi.org/10.1109/TIP.2014.2344294

    Yang,J.,Yuan,X.,Liao,X.,Llull,P.,Brady,D.J.,Sapiro,G.,Carin,L.:Videocom- pressive sensing using gaussian mixture models. IEEE Transactions on Image Pro- cessing23(11), 4863–4878 (2014).https://doi.org/10.1109/TIP.2014.2344294

    work page doi:10.1109/tip.2014.2344294 2014

  35. [37]

    Selective review of offline change point detection methods,

    Yin, J., Wang, N., Hu, B., Wang, Y., Wang, Q.: Degradation-aware deep unfolding network with transformer prior for video compressive imaging. Signal Processing 227, 109660 (2025).https://doi.org/https://doi.org/10.1016/j.sigpro. 2024.109660

    work page doi:10.1016/j.sigpro 2025

  36. [38]

    In: 2016 IEEE International Conference on Image Processing (ICIP)

    Yuan, X.: Generalized alternating projection based total variation minimization for compressive sensing. In: 2016 IEEE International Conference on Image Processing (ICIP). pp. 2539–2543 (2016).https://doi.org/10.1109/ICIP.2016.7532817

    work page doi:10.1109/icip.2016.7532817 2016

  37. [39]

    Snapshot Compressive Imaging: Theory, Algorithms, and Applications,

    Yuan, X., Brady, D.J., Katsaggelos, A.K.: Snapshot compressive imaging: The- ory, algorithms, and applications. IEEE Signal Processing Magazine38(2), 65–88 (2021).https://doi.org/10.1109/MSP.2020.3023869

    work page doi:10.1109/msp.2020.3023869 2021

  38. [40]

    In: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Yuan, X., Liu, Y., Suo, J., Dai, Q.: Plug-and-play algorithms for large-scale snap- shot compressive imaging. In: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 1444–1454 (2020).https://doi.org/10. 1109/CVPR42600.2020.00152

    arXiv 2020

  39. [41]

    Evangelidis and Emmanouil Z

    Yuan, X., Liu, Y., Suo, J., Durand, F., Dai, Q.: Plug-and-play algorithms for video snapshot compressive imaging. IEEE Transactions on Pattern Analysis and Ma- chine Intelligence44(10), 7093–7111 (2022).https://doi.org/10.1109/TPAMI. 2021.3099035

    work page doi:10.1109/tpami 2022

  40. [42]

    doi: 10.1109/ICCV51070.2023.00649

    Zheng, S., Yuan, X.: Unfolding framework with prior of convolution-transformer mixture and uncertainty estimation for video snapshot compressive imaging. In: 2023IEEE/CVFInternationalConferenceonComputerVision(ICCV).pp.12692– 12703 (2023).https://doi.org/10.1109/ICCV51070.2023.01170

    work page doi:10.1109/iccv51070.2023.01170 2023