Pith. sign in

REVIEW 3 major objections 6 minor 75 references

Manifold Modeling in Embedded Space: A Perspective for Interpreting Deep Image Prior

T0 review · 3 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Splitting convolution into embedding plus transformation yields a simple network that matches deep image prior, supporting the view that DIP works by enforcing a low-dimensional patch-manifold prior.

desk verdict A simple Hankelization-plus-autoencoder model matches DIP on four restoration tasks, and the claim that DIP is a low-dimensional patch-manifold prior is plausible but not proven—the authors mostly say so themselves. read the letter →

arxiv 1908.02995 v2 pith:LKPCGUAZ submitted 2019-08-08 cs.CV cs.LG

Tatsuya Yokota , Hidekata Hontani , Qibin Zhao , Andrzej Cichocki This is my paper · ORCID
classification cs.CVcs.LG
keywords deepimagepriormanifoldmodelingdelayembeddingHankelizationdenoisingauto-encodertensorcompletionrestorationself-similarity
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to answer why the deep image prior (DIP) works by decomposing convolution into two parts: delay-embedding, which collects local patches, and a transformation, which maps them into a low-dimensional space. The authors build a much simpler model, manifold modeling in embedded space (MMES), from multiway delay embedding plus a denoising auto-encoder, and show it matches DIP across image completion, super-resolution, deconvolution, and denoising. Because MMES reproduces DIP's key behaviors, including its noise impedance, the paper argues that DIP's convolutional architecture is essentially enforcing a low-dimensional patch-manifold prior rather than something unique to convolution.

What carries the argument

The load-bearing mechanism is the multiway delay-embedding transform (Hankelization) sandwiched around a denoising auto-encoder. The transform $\mathcal{H}$ slides a window of size $\tau$ over the tensor, arranging all overlapping patches as columns of a matrix; the auto-encoder, with encoder $\phi_r: \mathbb{R}^D \to \mathbb{R}^r$ and decoder $\psi_r$, maps those columns into an $r$-dimensional patch manifold; the pseudo-inverse $\mathcal{H}^\dagger$ folds the auto-encoded columns back into an image. The objective combines a reconstruction loss $\|Y - F(\mathcal{H}^\dagger A_r \mathcal{H}(Z))\|_F^2$ with an auto-encoding loss $\|\mathcal{H}(Z) - A_r \mathcal{H}(Z)\|_F^2$, trained with additive noise. This structure is what enforces the low-dimensional patch-manifold prior and produces the impedance-like behavior.

What would settle it

Take a natural image and a version with identical patch statistics but randomly reordered patch positions, then optimize both DIP and MMES while recording mean squared error over time. The paper's account predicts both methods will show the same ordering of convergence rates because both are governed by the patch manifold, whereas a spatial-locality account predicts DIP resists the shuffled image much more than MMES; that divergence would falsify the interpretation.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the successful behavior of DIP does not depend on deep convolutional filters: replacing every convolution by the combination of a Hankelization (delay-embedding) step, an auto-encoder that compresses each patch to r dimensions, and the pseudo-inverse embedding yields comparable or better restorations on color-image completion up to 99% missing pixels, 3D MRI completion, ×4 and ×8 super-resolution, deconvolution, and denoising. The same noise impedance phenomenon—natural images optimize fastest, noisy images slower, shuffled and uniform patterns slowest—appears in MMES. The proposed interpretation is that DIP is a low-dimensional patch-manifold prior: convolutional layers constrain every local patch to lie near a low-dimensional manifold, and the kernel size and filter count correspond to the window size τ and latent dimension r in MMES.

Load-bearing premise

The interpretation stands on the assumption that the low-dimensional patch structure learned by the auto-encoder from the corrupted image is the same prior that makes DIP's convolutional network effective; if the resemblance comes instead from a shared tendency to fit smooth or repetitive content first, the paper's explanation of DIP does not follow.

Editorial extensions

If this is right

  • MMES reaches DIP-level quality on completion, super-resolution, deconvolution, and denoising, so the convolutional structure of DIP is not the sole source of its image prior.
  • Because MMES also exhibits noise impedance, the paper's explanation of impedance transfers: a denoising auto-encoder maps noisy patches toward higher-density regions on the patch manifold, pulling reconstructions away from noise.
  • DIP's architecture can be described in plain terms: local patch statistics are confined to a low-dimensional manifold, with kernel size and filter count playing the roles of window size τ and latent dimension r.
  • MMES is computationally lighter than DIP on 3D data in the reported experiments, making it a practical surrogate for unsupervised restoration when GPU memory or time is limited.
  • Patch size becomes a central hyperparameter: too large a window creates too many patch variations, while too small a window loses the self-similarity information needed for heavily corrupted inputs.

Reading between the lines

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

  • If the paper is right, the large literature on denoising auto-encoders—training noise as Tikhonov regularization, entropy reduction, and mapping toward high-density regions—becomes a quantitative toolkit for predicting when DIP will succeed or fail.
  • If the paper is right, behaviors that DIP and MMES share cannot be attributed to spatially extended convolutional kernels, since MMES uses only pointwise transforms after embedding; remaining differences would isolate what genuine convolution adds.
  • If the paper is right, a natural extension is to swap the denoising auto-encoder for PCA, a variational auto-encoder, or an adversarial auto-encoder; if DIP-like behavior persists only with denoising training, that would pin the prior on the noise-reconstruction objective rather than on manifold dimensionality alone.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes Manifold Modeling in Embedded Space (MMES), an unsupervised image/tensor restoration model that replaces convolution with a multiway delay-embedding (Hankelization) step, a nonlinear encoder-decoder autoencoder, and the pseudo-inverse embedding (Eqs. (5)-(8)). The authors frame this as a minimal translation of the convolutional generator used in Deep Image Prior (DIP), with patch size tau and latent dimension r playing the roles of kernel size and filter size. They report experiments across tensor completion, super-resolution, deconvolution, and denoising, showing that MMES produces results 'quite similar even competitive to DIP', and they interpret the similarity as evidence that DIP's implicit prior is a low-dimensional patch-manifold prior. The paper also reproduces DIP's 'noise impedance' phenomenon with MMES in a toy experiment.

Significance. If the central claim were established, this would be a valuable conceptual contribution: it would translate the opaque CNN prior of DIP into a checkable classical image-modeling vocabulary, namely low-dimensional patch manifolds, and it would connect DIP to dynamical-systems delay embedding and self-similarity. The paper's empirical body is a genuine strength: MMES is simple and precisely described, and the four-task comparison with numerous baselines is unusually broad for an interpretability paper. The toy manifold visualization and the noise-impedance comparison are nice illustrations. However, the evidence supports only a sufficiency statement, not the necessity/causal claim made in the title and abstract, and the comparison protocol is not fully symmetric. With the causal gap addressed or the claim appropriately weakened, the paper would be a solid empirical contribution.

major comments (3)
  1. [Abstract; Section V.B] The central interpretive claim that DIP's success is explained by an implicit low-dimensional patch-manifold prior is not secured by the evidence. What the experiments show is a sufficiency result: MMES, which explicitly imposes this prior through Eq. (5) and Eq. (8), produces restoration results and noise-impedance curves similar to DIP. This does not establish that the manifold constraint is the operative mechanism in DIP, because other priors favoring smooth or low-frequency content could produce the same behavioral signature. The manuscript never varies the manifold constraint independently (for example by changing r or tau while holding the rest of the model fixed and measuring whether DIP's behavior tracks the change), nor does it test whether DIP's internal patch representations lie near the learned manifold. The authors' own language in Section V.B, using 'prospect', 'rough explanation', and 'we believe', is more cautious than the title and abstract, and the claim should either be supported by such a causal or diagnostic experiment or explicitly downgraded to a conjecture.
  2. [Section IV.D.1 and Table I] The empirical claim that MMES is 'quite similar even competitive to DIP' is weakened by an asymmetric comparison protocol. MMES hyperparameters (tau, r) are tuned per image and per missing rate for best PSNR/SSIM (Table I), while DIP is run only with the default architecture and no search over kernel sizes, filter sizes, or depth. Conversely, DIP is given oracle early stopping, with the best iteration chosen by PSNR, while MMES is stopped at a fixed 20,000 iterations in the completion task. These asymmetries cut in different directions and make the aggregate curves in Fig. 16 difficult to interpret as a head-to-head test. Please either standardize the protocols, by giving both methods comparable per-image tuning or fixed hyperparameters, or present the comparison as illustrative rather than as evidence of competitiveness.
  3. [Section IV.D.2] The 3D MRI comparison is also affected by a capacity mismatch: the authors state that DIP's 3D filter counts were 'slightly reduced' because of GPU memory, and they attribute part of DIP's degradation to this. Since this experiment is used to claim that MMES outperforms DIP at low missing rates, the result should either be repeated with matched capacity and comparable computational cost or explicitly labeled as preliminary. As written, the sentence 'the proposed MMES outperformed DIP in low-rate missing cases' overstates what the experiment can support.
minor comments (6)
  1. [Table I] The entry 'saiboat' appears to be a typo and should read 'sailboat'.
  2. [Fig. 12] The hyperparameter sensitivity discussion refers only to visual quality; reporting PSNR/SSIM for each (r, sigma) and (tau, r) cell would make the claims about appropriate parameter ranges quantitatively checkable.
  3. [Algorithm 1] The stopping criterion 'until converge' is not defined; please specify a concrete termination condition, such as a maximum iteration count or a tolerance on the loss change.
  4. [Section II] There is a typo: 'point could' should be 'point cloud'; in addition, the sentence 'the MMES is originally/essentially has only decoder' is grammatically awkward and should be rewritten.
  5. [Section IV.D.4] The deconvolution comparison is reported only as per-image PSNR/SSIM values in Fig. 20; please include average values over the nine cases to support the statement that the methods are similar 'qualitatively and quantitatively'.
  6. [Footnotes] Several footnotes contain hard line breaks inside URLs (e.g., footnotes 7 and 8); please repair the line breaks so the links are readable.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: MMES is independently constructed, and the paper explicitly frames the DIP interpretation as a prospective explanation rather than a derivation.

full rationale

The derivation chain is self-contained. Equations (1)-(8) construct MMES from an explicit low-dimensional patch-manifold assumption: multi-way delay embedding H, a denoising auto-encoder A_r defining the manifold, and reconstruction through H† A_r H(Z). This is a genuine model, not a restatement of DIP. The paper's central claim is an empirical-interpretive one: MMES, which explicitly imposes the patch-manifold prior, produces restoration quality and noise-impedance behavior similar to DIP, and this similarity 'would help us for reinterpreting/characterizing the DIP' (abstract). The paper itself marks the explanation as tentative: Section V.B calls it 'a prospect' and 'our rough explanation,' and Section VI says 'we believe' it gives insight. No equation in the paper reduces DIP's output or optimization behavior to MMES by construction; DIP is an external baseline (Ulyanov et al.) that is simulated but not derived from the MMES equations. The only self-citation used structurally, Yokota et al. [66] for the multiway-delay embedding transform, is not load-bearing because the transform is fully defined in Eqs. (2)-(4) of this paper and can be checked independently. The skeptical concern that the low-dimensional patch-manifold constraint is never isolated as the causal mechanism inside DIP is a limitation of evidence for an interpretive hypothesis, not a circular reduction. Accordingly, no fitted parameter is relabeled as a prediction and no claimed result is equivalent to its inputs by definition.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The method introduces no new particles or physical entities. The free parameters are standard hyperparameters of patch-based and autoencoder methods, but they are tuned per image, which makes the comparison optimistic. The key axiom is the equivalence between DIP's implicit prior and the patch-manifold constraint, which is assumed for the interpretation rather than demonstrated.

free parameters (5)
  • tau (patch size) = varies per task and image, e.g., (6,6) to (16,16) for completion (Table I), (4,4,4) for MRI, 6 for super-resolution…
    Chosen by manual tuning to maximize PSNR/SSIM. The paper explicitly states hyperparameters of all methods were tuned manually to perform best PSNR/SSIM.
  • r (latent dimension of autoencoder) = varies per task and image, e.g., 1, 4, 8, 16, 32, 48, 64 in Table I and Fig.
    Controls the dimensionality of the patch manifold; tuned per image/task for best result.
  • sigma (noise std for denoising autoencoder) = 0.05 in completion, MRI, and denoising; 0.1 in super-resolution; 0.01 in deconvolution
    Noise level of DAE is a hyperparameter tuned per task; the sensitivity analysis in Fig. 12 shows performance depends on it.
  • lambda (trade-off parameter) = initialized to 5.0, then adapted multiplicatively (1.1 or 0.99) based on Lrec < LAE
    Balances reconstruction and auto-encoding losses; the adaptation schedule is a design choice rather than derived.
  • AE architecture (hidden layer sizes) = [8*tau^2, r, 8*tau^2] for super-resolution and denoising; [32*tau^2, r, 32*tau^2] for deconvolution; Fig.
    The number and width of hidden layers are hand-chosen per task, affecting representation ability.
assumptions (4)
  • standard math The pseudo-inverse H-dagger satisfies H-dagger(H) is the identity on the image space (Eq. 4).
    Used to derive Eq. (6)-(7); standard property of Moore-Penrose pseudo-inverse for the Hankelization operator with padding.
  • domain assumption Natural image patches lie on a low-dimensional manifold in patch space.
    This is the core prior; inherited from patch-manifold models [42], [39] and sparse-land models; assumed throughout, not proven.
  • domain assumption A denoising autoencoder trained on corrupted patches maps inputs toward higher-density regions of the data distribution.
    Invoked in Section V-B to explain noise impedance, citing [1].
  • ad hoc to paper The low-dimensional patch-manifold constraint is equivalent, in essence, to the implicit prior of DIP's convolutional architecture.
    This is the paper's interpretive claim, asserted in Section V and the abstract, but not derived or independently tested. The paper itself concedes the models are not completely equivalent.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Manifold Modeling in Embedded Space: A Perspective for Interpreting Deep Image Prior." pith.science (2026). https://pith.science/paper/LKPCGUAZ

@misc{pith2026190802995,
  author       = {Pith},
  title        = {Pith review of: Manifold Modeling in Embedded Space: A Perspective for Interpreting Deep Image Prior},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LKPCGUAZ}},
  note         = {Machine review of arXiv:1908.02995}
}
read the original abstract

Deep image prior (DIP), which utilizes a deep convolutional network (ConvNet) structure itself as an image prior, has attracted attentions in computer vision and machine learning communities. It empirically shows the effectiveness of ConvNet structure for various image restoration applications. However, why the DIP works so well is still unknown, and why convolution operation is useful for image reconstruction or enhancement is not very clear. In this study, we tackle these questions. The proposed approach is dividing the convolution into ``delay-embedding'' and ``transformation (\ie encoder-decoder)'', and proposing a simple, but essential, image/tensor modeling method which is closely related to dynamical systems and self-similarity. The proposed method named as manifold modeling in embedded space (MMES) is implemented by using a novel denoising-auto-encoder in combination with multi-way delay-embedding transform. In spite of its simplicity, the image/tensor completion, super-resolution, deconvolution, and denoising results of MMES are quite similar even competitive to DIP in our extensive experiments, and these results would help us for reinterpreting/characterizing the DIP from a perspective of ``low-dimensional patch-manifold prior''.

Figures

Figures reproduced from arXiv: 1908.02995 by the authors.

Figure 1
Figure 1. Conceptual illustration of deep image prior and the proposed manifold [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Comparison of typical auto-encoder ConvNet and the proposed MMES [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figure 4
Figure 4. Flow of multiway-delay-embedding operation ( [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (15 more)
Figure 5
Figure 5. Figure 5: Duplication matrix. In case that we have [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 7
Figure 7. Figure 7: An example of architecture of auto-encoder. [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]
Figure 8
Figure 8. Figure 8: Generator network in a case of color-image recovery. [PITH_FULL_IMAGE:figures/full_fig_p005_8.png]
Figure 10
Figure 10. Figure 10: Two-dimensional (8,8)-patch manifold learned from a 50% missing [PITH_FULL_IMAGE:figures/full_fig_p006_10.png]
Figure 11
Figure 11. Figure 11: Optimization behavior. The difference from the non-local similarity based approach is that the manifold modeling is “global” rather than “non￾local” which finds similar patches of the target patch from its neighborhood area. 1) Optimization behavior: For this experime…
Figure 12
Figure 12. Figure 12: Performance of reconstruction of color image of ‘Lena’ with 99% [PITH_FULL_IMAGE:figures/full_fig_p007_12.png]
Figure 13
Figure 13. Figure 13: Reconstruction of ‘home’ image by training with/without noise in [PITH_FULL_IMAGE:figures/full_fig_p007_13.png]
Figure 14
Figure 14. Figure 14: Reconstruction of ‘Lena’ image for various patch sizes [PITH_FULL_IMAGE:figures/full_fig_p007_14.png]
Figure 15
Figure 15. Figure 15: Optimization behaviors of DIP and MMES for different target images. (a) Four target images. (b) and (c) show optimization behaviors of DIP and [PITH_FULL_IMAGE:figures/full_fig_p008_15.png]
Figure 16
Figure 16. Figure 16: Comparison of averages of PSNR and SSIM for eight color image [PITH_FULL_IMAGE:figures/full_fig_p008_16.png]
Figure 17
Figure 17. Figure 17: Completion results from images with 99% missing pixels by HaLTRC [ [PITH_FULL_IMAGE:figures/full_fig_p009_17.png]
Figure 18
Figure 18. Figure 18: Results of MRI completion: (a) Illustration of MRI reconstructed from 90% missing tensor, and (b) optimization behaviors of PSNR with final values [PITH_FULL_IMAGE:figures/full_fig_p009_18.png]
Figure 19
Figure 19. Figure 19: Super-resolution results: The first and second lines ‘Starfish’ and ‘Leaves’ were up-scaled from (64,64,3) to (256,256,3), the third line ‘Monarch’ [PITH_FULL_IMAGE:figures/full_fig_p010_19.png]
Figure 20
Figure 20. Figure 20: Comparison of our approach with DIP for deconvolution/deblurring task. Three color images were blurred by three Gaussian windows of different [PITH_FULL_IMAGE:figures/full_fig_p011_20.png]
Figure 21
Figure 21. Figure 21: Denoising results in case of that the noise standard deviation is 40. [PITH_FULL_IMAGE:figures/full_fig_p012_21.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

75 extracted references · 67 canonical work pages

  1. [1]

    Alain and Y

    G. Alain and Y . Bengio. What regularized auto-encoders learn from the data-generating distribution. The Journal of Machine Learning Research, 15(1):3563–3593, 2014. 12

    work page 2014

  2. [2]

    Batson and L

    J. Batson and L. Royer. Noise2self: Blind denoising by self-supervision. In Proceedings of ICML , pages 524–533, 2019. 3

    work page 2019

  3. [3]

    C. M. Bishop. Training with noise is equivalent to Tikhonov regular- ization. Neural Computation, 7(1):108–116, 1995. 5

    work page 1995

  4. [4]

    Buades, B

    A. Buades, B. Coll, and J.-M. Morel. A non-local algorithm for image denoising. In Proceedings of CVPR , volume 2, pages 60–65. IEEE,

    work page

  5. [5]

    S. Cha, T. Park, and T. Moon. Gan2gan: Generative noise learning for blind image denoising with single noisy images. arXiv preprint arXiv:1905.10488, 2019. 3

    arXiv 1905

  6. [6]

    Cichocki, R

    A. Cichocki, R. Zdunek, A. H. Phan, and S.-i. Amari. Nonnegative Matrix and Tensor Factorizations: Applications to Exploratory Multi- way Data Analysis and Blind Source Separation . John Wiley & Sons,

    work page

  7. [7]

    Creswell and A

    A. Creswell and A. A. Bharath. Denoising adversarial autoen- coders. IEEE Transactions on Neural Networks and Learning Systems , 30(4):968–984, 2018. 5

    work page 2018

  8. [8]

    Dabov, A

    K. Dabov, A. Foi, V . Katkovnik, and K. Egiazarian. Color image denoising via sparse 3D collaborative filtering with grouping constraint in luminance-chrominance space. In Proceedings of ICIP , volume 1, pages I–313. IEEE, 2007. 10

    work page 2007

Show all 75 references
  1. [9]

    Dabov, A

    K. Dabov, A. Foi, V . Katkovnik, and K. Egiazarian. Image denoising by sparse 3-d transform-domain collaborative filtering. IEEE Transactions on Image Processing , 16(8), 2007. 1, 6, 7

    2007

  2. [10]

    M. W. Diederik P Kingma. Auto-encoding variational bayes. In Proceedings of ICLR , 2014. 5

    2014

  3. [11]

    T. Ding, M. Sznaier, and O. I. Camps. A rank minimization approach to video inpainting. In Proceedings of ICCV , pages 1–8. IEEE, 2007. 12

    2007

  4. [12]

    Elad and M

    M. Elad and M. Aharon. Image denoising via learned dictionaries and sparse representation. In Proceedings of CVPR , volume 1, pages 895–

  5. [13]

    Elad and M

    M. Elad and M. Aharon. Image denoising via sparse and redundant representations over learned dictionaries. IEEE Transactions on Image processing, 15(12):3736–3745, 2006. 11

    2006

  6. [14]

    Gal and Z

    Y . Gal and Z. Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. In Proceedings of ICML, pages 1050–1059, 2016. 5

    2016

  7. [15]

    Gandelsman, A

    Y . Gandelsman, A. Shocher, and M. Irani. ”double-dip”: Unsupervised image decomposition via coupled deep-image-priors. In Proceedings of CVPR, June 2019. 12

    2019

  8. [16]

    K. Gong, C. Catana, J. Qi, and Q. Li. PET image reconstruction using deep image prior. IEEE Transactions on Medical Imaging , 2018. 12

    2018

  9. [17]

    Goodfellow, Y

    I. Goodfellow, Y . Bengio, and A. Courville. Deep Learning. MIT press,

  10. [18]

    W. E. L. Grimson. From Images to Surfaces: A Computational Study of the Human Early Visual System . MIT Press, 1981. 1

    1981

  11. [19]

    S. Gu, L. Zhang, W. Zuo, and X. Feng. Weighted nuclear norm minimization with application to image denoising. In Proceedings of CVPR, pages 2862–2869, 2014. 6, 7

    2014

  12. [20]

    Guichard and F

    F. Guichard and F. Malgouyres. Total variation based interpolation. In Proceedings of EUSIPCO , pages 1–4. IEEE, 1998. 1

    1998

  13. [21]

    Heckel and P

    R. Heckel and P. Hand. Deep decoder: Concise image represen- tations from untrained non-convolutional networks. arXiv preprint arXiv:1810.03982, 2018. 3

    2018 arXiv

  14. [22]

    G. E. Hinton and R. R. Salakhutdinov. Reducing the dimensionality of data with neural networks. Science, 313(5786):504–507, 2006. 4

    2006

  15. [23]

    F. L. Hitchcock. The expression of a tensor or a polyadic as a sum of products. Journal of Mathematics and Physics , 6(1):164–189, 1927. 1

    1927

  16. [24]

    Hotelling

    H. Hotelling. Analysis of a complex of statistical variables into principal components. Journal of Educational Psychology , 24(6):417, 1933. 1

    1933

  17. [25]

    W. Hu, D. Tao, W. Zhang, Y . Xie, and Y . Yang. The twist tensor nuclear norm for video completion. IEEE Transactions on Neural Networks and Learning Systems, 28(12):2961–2973, 2016. 1

    2016

  18. [26]

    Hyvarinen, J

    A. Hyvarinen, J. Karhunen, and E. Oja. Independent Component Analysis, volume 46. John Wiley & Sons, 2004. 1

    2004

  19. [27]

    H. Ji, C. Liu, Z. Shen, and Y . Xu. Robust video denoising using low rank matrix completion. In Proceedings of CVPR , pages 1791–1798. IEEE, 2010. 1

    2010

  20. [28]

    D. P. Kingma and J. Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 , 2014. 5

    2014 arXiv

  21. [29]

    Krull, T.-O

    A. Krull, T.-O. Buchholz, and F. Jug. Noise2void-learning denoising from single noisy images. In Proceedings of CVPR , pages 2129–2137,

  22. [30]

    Laine, J

    S. Laine, J. Lehtinen, and T. Aila. Self-supervised deep image denoising. arXiv preprint arXiv:1901.10277 , 2019. 3

    1901 arXiv

  23. [31]

    D. D. Lee and H. S. Seung. Learning the parts of objects by non-negative matrix factorization. Nature, 401(6755):788, 1999. 1

    1999

  24. [32]

    Lehtinen, J

    J. Lehtinen, J. Munkberg, J. Hasselgren, S. Laine, T. Karras, M. Aittala, and T. Aila. Noise2noise: Learning image restoration without clean data. In Proceedings of ICML , pages 2971–2980, 2018. 3

    2018

  25. [33]

    S. Z. Li. Markov random field models in computer vision. In Proceedings of ECCV , pages 361–370. Springer, 1994. 1

    1994

  26. [34]

    Y . Li, K. R. Liu, and J. Razavilar. A parameter estimation scheme for damped sinusoidal signals based on low-rank Hankel approximation. IEEE Transactions on Signal Processing , 45(2):481–486, 1997. 12

    1997

  27. [35]

    J. Liu, P. Musialski, P. Wonka, and J. Ye. Tensor completion for estimating missing values in visual data. IEEE Transactions on Pattern Analysis and Machine Intelligence , 35(1):208–220, 2013. 1, 7, 9

    2013

  28. [36]

    Majumdar

    A. Majumdar. Blind denoising autoencoder. IEEE Transactions on Neural Networks and Learning Systems , 30(1):312–317, 2018. 5

    2018

  29. [37]

    Makhzani, J

    A. Makhzani, J. Shlens, N. Jaitly, I. Goodfellow, and B. Frey. Adversarial autoencoders. arXiv preprint arXiv:1511.05644 , 2015. 5

    2015 arXiv

  30. [38]

    Markovsky

    I. Markovsky. Structured low-rank approximation and its applications. Automatica, 44(4):891–909, 2008. 12

    2008

  31. [39]

    Osher, Z

    S. Osher, Z. Shi, and W. Zhu. Low dimensional manifold model for image processing. SIAM Journal on Imaging Sciences , 10(4):1669– 1690, 2017. 2

    2017

  32. [40]

    N. H. Packard, J. P. Crutchfield, J. D. Farmer, and R. S. Shaw. Geometry from a time series. Physical Review Letters , 45(9):712, 1980. 12

    1980

  33. [41]

    K. Pearson. LIII. On lines and planes of closest fit to systems of points in space. The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science , 2(11):559–572, 1901. 1

    1901

  34. [42]

    G. Peyre. Manifold models for signals and images. Computer Vision and Image Understanding , 113(2):249–260, 2009. 2

    2009

  35. [43]

    Poggio, V

    T. Poggio, V . Torre, and C. Koch. Computational vision and regulariza- tion theory. Nature, 317:26, 1985. 1

    1985

  36. [44]

    Rosca, B

    M. Rosca, B. Lakshminarayanan, D. Warde-Farley, and S. Mohamed. Variational approaches for auto-encoding generative adversarial net- works. arXiv preprint arXiv:1706.04987 , 2017. 5

    2017 arXiv

  37. [45]

    T. R. Shaham, T. Dekel, and T. Michaeli. Singan: Learning a generative model from a single natural image. In Proceedings of ICCV , pages 4570–4580, 2019. 3

    2019

  38. [46]

    Shi, Y .-M

    Q. Shi, Y .-M. Cheung, Q. Zhao, and H. Lu. Feature extraction for incomplete data via low-rank tensor decomposition with feature regularization. IEEE Transactions on Neural Networks and Learning Systems, 30(6):1803–1817, 2018. 1

    2018

  39. [47]

    Shocher, S

    A. Shocher, S. Bagon, P. Isola, and M. Irani. Ingan: Capturing and retargeting the DNA of a natural image. In Proceedings of ICCV, pages 4492–4501, 2019. 3

    2019

  40. [48]

    Shocher, N

    A. Shocher, N. Cohen, and M. Irani. Zero-shot super-resolution using deep internal learning. In Proceedings of CVPR, pages 3118–3126, 2018. 3

    2018

  41. [49]

    Sonoda and N

    S. Sonoda and N. Murata. Transportation analysis of denoising au- toencoders: A novel method for analyzing deep neural networks. arXiv preprint arXiv:1712.04145, 2017. 5

    2017 arXiv

  42. [50]

    Szummer and R

    M. Szummer and R. W. Picard. Temporal texture modeling. In Proceedings of ICIP , volume 3, pages 823–826. IEEE, 1996. 12

    1996

  43. [51]

    Tibshirani

    R. Tibshirani. Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society: Series B (Methodological) , 58(1):267– 288, 1996. 1

    1996

  44. [52]

    L. R. Tucker. Some mathematical notes on three-mode factor analysis. Psychometrika, 31(3):279–311, 1966. 1

    1966

  45. [53]

    Ulyanov, A

    D. Ulyanov, A. Vedaldi, and V . Lempitsky. Deep image prior. In Proceedings of CVPR , pages 9446–9454, 2018. 1, 7, 8, 9, 10, 11

    2018

  46. [54]

    Van Overschee and B

    P. Van Overschee and B. De Moor. Subspace algorithms for the stochastic identification problem. In Proceedings of IEEE Conference on Decision and Control , pages 1321–1326. IEEE, 1991. 12

    1991

  47. [55]

    Van Veen, A

    D. Van Veen, A. Jalal, M. Soltanolkotabi, E. Price, S. Vishwanath, and A. G. Dimakis. Compressed sensing with deep image prior and learned regularization. arXiv preprint arXiv:1806.06438 , 2018. 12

    2018 arXiv

  48. [56]

    Vincent, H

    P. Vincent, H. Larochelle, Y . Bengio, and P.-A. Manzagol. Extracting and composing robust features with denoising autoencoders. In Proceedings of ICML, pages 1096–1103, 2008. 5

    2008

  49. [57]

    C. R. V ogel and M. E. Oman. Fast, robust total variation-based reconstruction of noisy, blurred images. IEEE Transactions on Image YOKOTA et al.: MANIFOLD MODELING IN EMBEDDED SPACE: A PERSPECTIVE FOR INTERPRETING DEEP IMAGE PRIOR 14 Processing, 7(6):813–824, 1998. 1

    1998

  50. [58]

    W. Wang, V . Aggarwal, and S. Aeron. Efficient low rank tensor ring completion. In Proceedings of ICCV , pages 5697–5705, 2017. 1

    2017

  51. [59]

    Williams, T

    F. Williams, T. Schneider, C. Silva, D. Zorin, J. Bruna, and D. Panozzo. Deep geometric prior for surface reconstruction. In Proceedings of CVPR, pages 10130–10139, 2019. 3

    2019

  52. [60]

    Y . Wu, H. Tan, Y . Li, J. Zhang, and X. Chen. A fused cp factorization method for incomplete tensors. IEEE Transactions on Neural Networks and Learning Systems , 30(3):751–764, 2018. 1

    2018

  53. [61]

    J. Xu, Y . Huang, L. Liu, F. Zhu, X. Hou, and L. Shao. Noisy-as- clean: Learning unsupervised denoising from the corrupted image. arXiv preprint arXiv:1906.06878, 2019. 3

    1906 arXiv

  54. [62]

    J. Xu, L. Zhang, D. Zhang, and X. Feng. Multi-channel weighted nuclear norm minimization for real color image denoising. In Proceedings of ICCV, pages 1096–1104, 2017. 10

    2017

  55. [63]

    Y . Xu, R. Hao, W. Yin, and Z. Su. Parallel matrix factorization for low-rank tensor completion. Inverse Problems & Imaging , 9(2), 2015. 7, 9

    2015

  56. [64]

    J. Xue, Y . Zhao, W. Liao, J. C.-W. Chan, and S. G. Kong. Enhanced sparsity prior model for low-rank tensor completion. IEEE Transactions on Neural Networks and Learning Systems , 2019. 1

    2019

  57. [65]

    Yair and T

    N. Yair and T. Michaeli. Multi-scale weighted nuclear norm image restoration. In Proceedings of CVPR , pages 3165–3174, 2018. 7

    2018

  58. [66]

    Yokota, B

    T. Yokota, B. Erem, S. Guler, S. K. Warfield, and H. Hontani. Missing slice recovery for tensors using a low-rank model in embedded space. In Proceedings of CVPR , pages 8251–8259, 2018. 2, 3, 6, 7, 9

    2018

  59. [67]

    Yokota and H

    T. Yokota and H. Hontani. Simultaneous visual data completion and denoising based on tensor rank and total variation minimization and its primal-dual splitting algorithm. In Proceedings of CVPR , pages 3732– 3740, 2017. 7, 9

    2017

  60. [68]

    Yokota and H

    T. Yokota and H. Hontani. Simultaneous tensor completion and denoising by noise inequality constrained convex optimization. IEEE Access, 2019. 7

    2019

  61. [69]

    Yokota, K

    T. Yokota, K. Kawai, M. Sakata, Y . Kimura, and H. Hontani. Dy- namic PET image reconstruction using nonnegative matrix factorization incorporated with deep image prior. In Proceedings of ICCV , pages 3126–3135, 2019. 12

    2019

  62. [70]

    Yokota, Q

    T. Yokota, Q. Zhao, and A. Cichocki. Smooth PARAFAC decomposi- tion for tensor completion. IEEE Transactions on Signal Processing , 64(20):5423–5436, 2016. 7, 9

    2016

  63. [71]

    Zhang, D

    J. Zhang, D. Zhao, and W. Gao. Group-based sparse representation for image restoration. IEEE Transactions on Image Processing, 23(8):3336– 3351, 2014. 2, 6, 7, 9

    2014

  64. [72]

    Zhang, W

    L. Zhang, W. Wei, Q. Shi, C. Shen, A. van den Hengel, and Y . Zhang. Accurate tensor completion via adaptive low-rank representation. IEEE Transactions on Neural Networks and Learning Systems , 2019. 1

    2019

  65. [73]

    X. Zhang. A nonconvex relaxation approach to low-rank tensor com- pletion. IEEE Transactions on Neural Networks and Learning Systems , 30(6):1659–1671, 2018. 1

    2018

  66. [74]

    Zhang, G

    Z. Zhang, G. Ely, S. Aeron, N. Hao, and M. Kilmer. Novel methods for multilinear data completion and de-noising based on tensor-SVD. In Proceedings of CVPR , pages 3842–3849, 2014. 7, 9

    2014

  67. [75]

    Q. Zhao, L. Zhang, and A. Cichocki. Bayesian CP factorization of in- complete tensors with automatic rank determination. IEEE Transactions on Pattern Analysis and Machine Intelligence , 37(9):1751–1763, 2015. 1

    2015

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.