Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

DiffNMR3: Advancing NMR Resolution Beyond Instrumental Limits

T0 review · 3 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read A single conditional diffusion model reconstructs low-field NMR spectra into high-field-quality spectra across 14 upscaling factors, with lower error and fewer artifacts than per-factor baselines.

desk verdict A competent diffusion-based super-resolution pipeline for 2D NMR spectra, but the central claim of overcoming instrumental limits is unsupported because all low-resolution inputs are synthetic Gaussian blurs of high-field spectra. read the letter →

arxiv 2502.06845 v1 pith:HNC6RF5W submitted 2025-02-06 physics.ins-det cs.AIcs.LG

Sen Yan , Etienne Goffinet , Fabrizio Gabellieri , Ryan Young , Lydia Gkoura , Laurence Jennings , Filippo Castiglione , Thomas Launey This is my paper · ORCID
classification physics.ins-detcs.AIcs.LG
keywords NMRspectroscopydiffusionmodelsuper-resolutionmulti-scaleconditionalUNetlow-fieldspectralreconstructionGaussiandegradation
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

This paper tries to establish that a single conditional diffusion model, called MSSR, can take a low-resolution 2D NMR spectrum and reconstruct the spectrum a higher-field instrument would have produced, without changing the acquisition hardware. The authors argue that this makes high-field-quality molecular characterization accessible on affordable lower-field instruments. The model is trained on protein spectra from the ARTINA dataset that are artificially degraded by Gaussian blur and noise, and it is conditioned on a discrete upscaling factor so one network covers 14 field-strength ratios between 400 and 900 MHz. On held-out spectra, MSSR is reported to beat 14 task-specific baseline models on global and peak-level metrics, including lower hallucination and missed-peak ratios.

What carries the argument

The load-bearing mechanism is the conditional UNet of a denoising diffusion probabilistic model. The original high-field spectrum $x_0$ is corrupted over $T=2000$ diffusion steps; at inference the UNet denoises pure noise conditioned on three inputs: the low-resolution spectrum $x_{LR}$, the current time step $t$ via a sinusoidal embedding, and a one-hot class embedding of the upscaling factor $f$. The low-resolution spectrum is generated in training by Eq. (4): Gaussian convolution with standard deviation $\sigma_g = 1/f$ plus zero-mean Gaussian noise with standard deviation $0.01$. The class embedding is what provides the multi-scale behaviour, because changing $f$ changes the blur that must be inverted.

What would settle it

Run the trained MSSR on paired real low-field and high-field spectra of identical compounds, for example the same sample measured at 400 MHz and 900 MHz, and compare reconstructed peak positions and intensities against the high-field measurement. If the reconstructed spectra systematically miss or shift peaks that the high-field instrument resolves, or if using the true instrument point-spread function instead of Eq. (4) degrades reconstruction quality, the claim of high-field comparability fails.

Watch

Extended reading notes

Core claim

The central claim is that a diffusion model can invert the resolution loss of a low-field NMR measurement well enough that the output is comparable to a high-field spectrum. The paper demonstrates this for 2D protein NMR spectra by training a conditional denoising network to predict the noise added to a high-resolution spectrum, using the blurred low-resolution spectrum, a time step, and an upscaling factor class as conditions. Because the upscaling factor is encoded as a class label, the same trained network can generate spectra at any of the 14 ratios listed in Table 1 rather than needing one model per target field strength. On the ARTINA test split, the reconstruction improves over separate per-factor baseline models on global fit (MSE, $R^2$) and on peak-focused metrics, with hallucination ratios below 3 percent.

Load-bearing premise

The load-bearing premise is that a real low-field NMR spectrum is well approximated by convolving the corresponding high-field spectrum with a Gaussian kernel of width $\sigma_g = 1/f$ and adding Gaussian noise with standard deviation $0.01$; if real line shapes, field-dependent chemical shifts, or artifacts differ, the trained model will not transfer to actual low-field instruments.

Editorial extensions

If this is right

  • A single MSSR network replaces 14 field-pair-specific super-resolution models and is reported to improve on all of them on the paper's global and local metrics.
  • Low-field NMR users could post-process existing spectra to resolve finer splittings and peaks without buying a higher-field magnet.
  • Because the upscaling factor is a condition, the model can produce intermediate field-strength reconstructions (for example, 500, 600, 700, or 800 MHz targets from a 400 MHz input), not just one fixed target.
  • The method is designed to combine with other NMR acceleration techniques such as non-uniform sampling, which the authors name as a direction for further work.

Reading between the lines

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

  • If the Gaussian-degradation model of Eq. (4) matches a particular benchtop instrument, the same training recipe should transfer to that instrument, but the noise level $\sigma_{LR}=0.01$ and the Gaussian kernel would need to be re-estimated; the paper does not test this transfer.
  • The class embedding may interpolate to upscaling factors not in the discrete table (for instance, 550 MHz from 400 MHz), but the paper only evaluates the 14 listed ratios, so interpolation behaviour is an untested extension.
  • Because the training data are protein 2D spectra recorded at 600 to 950 MHz, the method's performance on small-molecule or 1D NMR, where line shapes and peak densities differ, is not established; testing on such data would clarify the scope.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes MSSR, a conditional diffusion model for 2D NMR super-resolution. Given a low-resolution NMR spectrum and an upscaling factor f, it reconstructs a high-resolution spectrum. Low-resolution inputs are generated synthetically by convolving high-resolution spectra from the ARTINA dataset with a Gaussian kernel of standard deviation sigma_g = 1/f and adding Gaussian noise. The model is conditioned on the upscaling factor f as a class embedding and is compared against per-factor baseline models. Reported global and local metrics show that MSSR outperforms these baselines on the synthetic degradations. The paper claims that reconstructions are comparable to high-field instruments and that the method overcomes the limitations of instrument field strength.

Significance. If validated on real low-field instruments, the approach could be significant, offering post-acquisition resolution enhancement without hardware upgrades. The use of a public protein NMR dataset, a unified multi-scale conditioning mechanism, and a clearly described diffusion pipeline are commendable. However, because the evaluation is entirely synthetic and based on the authors' own degradation operator, the central claim is not supported. The method as demonstrated is a deblurring of Gaussian-convolved spectra, not a demonstration of NMR super-resolution beyond instrumental limits. No comparison to existing NMR super-resolution methods is provided, and the quantitative results lack error bars, making the significance of the claimed improvements difficult to assess.

major comments (3)
  1. [Section 3.3 and Section 4.1] The low-resolution input x_LR is generated from the ground-truth high-resolution spectrum x_0 by Gaussian convolution with sigma_g = 1/f plus Gaussian noise (Eq. 4), and Section 4.1 states that this simulated spectrum is used as the low-resolution spectrum. Consequently, the model is trained and evaluated on inverting the authors' own blur operator. The abstract claims reconstructions are 'comparable to those obtained from high-field instruments' and that the method overcomes 'limitations of instrument field strength.' These claims require validation on real low-field NMR data, which is absent. Real low-field spectra differ from high-field spectra in line shapes, field-dependent chemical shifts, strong-coupling effects, shimming, and instrumental artifacts; no evidence is provided that the Gaussian degradation model is an adequate proxy.
  2. [Section 4.1 and Table 1] The same upscaling factor f used in Eq. (4) to set the Gaussian kernel width sigma_g = 1/f is also used as the class conditioning input. This design is internally consistent, but it means the model is asked to invert exactly the blur that generated its input. The paper does not test any other degradation mechanism, such as Lorentzian line shapes, field-dependent chemical shifts, or artifacts. Without real low-field data or at least a sensitivity analysis to mismatched blur kernels, the claim of 'beyond instrumental limits' is unsupported.
  3. [Section 4.4 and Figures 4–5] The quantitative evaluation reports global and local metrics without error bars or confidence intervals, and the number of test samples and variance across the test set are not described. Moreover, the peak-based metrics depend on the parameters of the peak detection algorithm (alpha, sigma, n in Algorithm 1) and on a distance threshold in peak matching, none of which are reported. These omissions make it impossible to assess the statistical significance of the claimed improvements or to reproduce the local metrics.
minor comments (5)
  1. [Title and header] The title and running header contain typographical errors: 'A DVANCING R ESOLUTION' should be 'Advancing Resolution'.
  2. [Section 4.1 and Table 1] Table 1 contains 15 entries, but the text states that there are n = 14 upscaling factors. The stated duplicate is also incorrect: 600 MHz/400 MHz gives 1.5 and 900 MHz/400 MHz gives 2.25, not the same value; the actual duplicate is 600/400 and 900/600, both equal to 1.5.
  3. [Section 4.5] The text says the baseline is 'unstable when the upscaling factor is smaller than 0.57', but the upscaling factors in Table 1 are all greater than 1. If the horizontal axis of Figures 4 and 5 plots the reciprocal of the upscaling factor, this should be stated explicitly.
  4. [Section 4.4.2 and Algorithm 1] The peak detection parameters alpha, sigma, and n, as well as the distance threshold used for peak matching, are not reported anywhere in the paper; these values are needed to reproduce the local metrics.
  5. [Introduction] The footnote in the Introduction contains a typo: 'brucker.com' should be 'bruker.com'.

Circularity Check

2 steps flagged · score 6.0 of 10

The central claim reduces to synthetic Gaussian deblurring: low-field inputs are generated from high-field targets (Eq. 4) and all metrics compare reconstructions against those same targets.

  1. self definitional [Section 3.3, 'Low-Resolution Spectrum' (Eq. 4)]
    "The original spectrum (high-resolution) x0 ∈ Rh×w undergoes downscaling by expanding the peaks’ widths based on a given upscaling factor f. ... the input spectrum is convolved with the Gaussian window g. Then the noise eϵ is added to simulate the low-resolution spectrum. xLR = convg(x0) + eϵ (4)"

    The low-field input is defined by the paper as a Gaussian blur of the very high-field spectrum that serves as ground truth, with Gaussian width sigma_g = 1/f. Thus the low-field-to-high-field relation is not measured or independently modeled; it is imposed by the authors' degradation operator. Reconstructing x0 from xLR therefore demonstrates inversion of that operator, not recovery from real low-field NMR data, so the abstract's claim of 'reconstruct high-field spectra from low-field NMR data' is equivalent to the input definition by construction.

  2. fitted input called prediction [Section 4.1, 'Dataset and experiment configuration']
    "Considering that the standard practice in NMR spectroscopy has been to train models on simulated spectra [42, 43, 44, 45], we employ the simulated spectrum mentioned in Section 3.3 as the low-resolution spectrum. Validated by the lab expert, we set σLR = 0.01."

    The only low-field data used are simulated via Eq. 4, with the noise floor set by the authors. All reported global and local metrics in Figs. 4-5 compare the reconstructed spectrum to the original x0 from which xLR was generated. The evaluation therefore measures the model's ability to invert the authors' own synthetic degradation, while the paper's central claim is about real instruments; the simulated low-field input is being presented as a prediction target without external validation.

full rationale

The paper is self-contained in the narrow sense that it does not rely on self-citation chains or imported uniqueness theorems: the diffusion architecture is standard and the ARTINA dataset is external. The circularity is in the validation of the central claim. Eq. 4 defines xLR as convg(x0) plus noise, so the model is trained and tested on pairs (x0, xLR) that are generated from the same x0 used as the reconstruction target. Section 4.1 explicitly chooses this simulated spectrum as the 'low-resolution spectrum.' Consequently, the reported 'comparable to high-field instruments' reconstructions are an in-sample synthetic deblurring result, not evidence that the method overcomes real instrumental limits. The upscaling factor f also controls both the blur width and the class conditioning, making the multi-scale task a conditioned deconvolution of the paper's own kernel. This is partial circularity rather than a mathematical tautology because the network still learns a nontrivial inverse on held-out proteins; a score of 6 reflects that the headline claim reduces by construction to synthetic deblurring.

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

The central claim rests on a synthetic degradation model (Gaussian blur plus noise) with hand-set noise level, an assumed inverse relation between field ratio and linewidth, and the validity of ARTINA high-field spectra plus 2D projections as ground truth. No new physical entities are introduced.

free parameters (5)
  • sigma_LR (low-resolution noise std) = 0.01
    Chosen by 'validated by the lab expert' (Section 4.1); controls the simulated low-resolution spectrum and is hand-set, not derived from instrument physics.
  • sigma_g = 1/f (Gaussian kernel width) = 1/f for each upscaling factor f
    Defines the synthetic degradation in Eq. (4); the functional relationship between field ratio and spectral line broadening is assumed, not measured.
  • inference_timestep T_inf = 500
    Chosen from Fig. 6 empirical study (Section 4.1, Appendix) rather than a principled criterion; tuned on the same data.
  • inference_repeat = 5
    Chosen from Fig. 6 empirical study; results are averages of 5 reconstructions.
  • peak detection parameters (alpha, sigma, n) = not reported
    Algorithm 1 requires a threshold factor, smoothing sigma, and neighborhood size; values are not given, affecting the hallucination and missed peak ratios.
assumptions (4)
  • ad hoc to paper Gaussian convolution plus Gaussian noise approximates the relation between low-field and high-field NMR spectra of the same sample (Eq. 4).
    This is the core degradation model; no physical derivation or real paired data support it.
  • ad hoc to paper A single scale parameter sigma_g = 1/f captures multi-field resolution differences.
    Assumed inverse relation between upscaling factor and line width; field-dependent effects (relaxation, second-order shifts, coupling) are ignored.
  • domain assumption ARTINA spectra (600 to 950 MHz) are valid high-field ground truth, and projecting 3D/4D spectra to 2D preserves signal integrity.
    Section 4.1; no validation that projections preserve quantitative peak intensities for training.
  • standard math DDPM with MSE loss learns the conditional distribution p(x0|xLR) for the synthetic paired data.
    Standard diffusion model theory; not the source of failure.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DiffNMR3: Advancing NMR Resolution Beyond Instrumental Limits." pith.science (2026). https://pith.science/paper/HNC6RF5W

@misc{pith2026250206845,
  author       = {Pith},
  title        = {Pith review of: DiffNMR3: Advancing NMR Resolution Beyond Instrumental Limits},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HNC6RF5W}},
  note         = {Machine review of arXiv:2502.06845}
}
read the original abstract

Nuclear Magnetic Resonance (NMR) spectroscopy is a crucial analytical technique used for molecular structure elucidation, with applications spanning chemistry, biology, materials science, and medicine. However, the frequency resolution of NMR spectra is limited by the "field strength" of the instrument. High-field NMR instruments provide high-resolution spectra but are prohibitively expensive, whereas lower-field instruments offer more accessible, but lower-resolution, results. This paper introduces an AI-driven approach that not only enhances the frequency resolution of NMR spectra through super-resolution techniques but also provides multi-scale functionality. By leveraging a diffusion model, our method can reconstruct high-field spectra from low-field NMR data, offering flexibility in generating spectra at varying magnetic field strengths. These reconstructions are comparable to those obtained from high-field instruments, enabling finer spectral details and improving molecular characterization. To date, our approach is one of the first to overcome the limitations of instrument field strength, achieving NMR super-resolution through AI. This cost-effective solution makes high-resolution analysis accessible to more researchers and industries, without the need for multimillion-dollar equipment.

Figures

Figures reproduced from arXiv: 2502.06845 by the authors.

Figure 1
Figure 1. The backbone of our MSSR approach. The UNet architecture [ [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Our MSSR pipeline. In the diffusion process, the original NMR spectrum [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. A: Original spectrum. B: Low-resolution spectrum (with upscaling factor [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Global metrics. We investigate the Mean Squared Error (MSE) and the Coefficient of Determination ( [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Local metrics. We list the peak-focused metrics since the peaks are critical for identifying compound features. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Different metrics plotted against the inference timesteps for different repeat values. Considering the consuming [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Reading Radio from Camera: Visually-Grounded, Lightweight, and Interpretable RSSI Prediction

    eess.SP 2025-10 reject novelty 5.0 of 10

    A MobileNet-based model that predicts path loss from GPS and a shadow-fading residual from images and boxes reaches 0.70 dB RSSI RMSE on DeepSense 6G, but the claimed advantage over prior work is confounded by extra inputs.

Reference graph

Works this paper leans on

44 extracted references · 22 canonical work pages · cited by 1 Pith paper

  1. [1]

    An introduction to biological nmr spectroscopy

    Dominique Marion. An introduction to biological nmr spectroscopy. Molecular & Cellular Proteomics , 12(11):3006–3025, 2013

    work page 2013

  2. [2]

    NMR spectroscopy: basic principles, concepts and applications in chemistry

    Harald Günther. NMR spectroscopy: basic principles, concepts and applications in chemistry . John Wiley & Sons, 2013

    work page 2013

  3. [3]

    Carbohydrate structural determination by nmr spectroscopy: modern methods and limitations

    Jens Ø Duus, Charlotte H Gotfredsen, and Klaus Bock. Carbohydrate structural determination by nmr spectroscopy: modern methods and limitations. Chemical reviews, 100(12):4589–4614, 2000

    work page 2000

  4. [4]

    NMR in molecular biology

    Oleg Jardetzky and Gordon Carl Kenmure Roberts. NMR in molecular biology. Academic Press, 2013

    work page 2013

  5. [5]

    Recent advances in application of 27al nmr spectroscopy to materials science

    Mohamed Haouas, Francis Taulelle, and Charlotte Martineau. Recent advances in application of 27al nmr spectroscopy to materials science. Progress in nuclear magnetic resonance spectroscopy, 94:11–36, 2016

    work page 2016

  6. [6]

    MRI: basic principles and applications

    Mark A Brown and Richard C Semelka. MRI: basic principles and applications. John Wiley & Sons, 2011

    work page 2011

  7. [7]

    High resolution NMR: theory and chemical applications

    Edwin D Becker. High resolution NMR: theory and chemical applications. Elsevier, 1999

    work page 1999

  8. [8]

    Low-field and benchtop nmr

    Bernhard Blümich. Low-field and benchtop nmr. Journal of Magnetic Resonance, 306:27–35, 2019

    work page 2019

Show all 44 references
  1. [9]

    Super-resolution image reconstruction: a technical overview

    Sung Cheol Park, Min Kyu Park, and Moon Gi Kang. Super-resolution image reconstruction: a technical overview. IEEE signal processing magazine, 20(3):21–36, 2003

    2003

  2. [10]

    Diffusion models, image super-resolution and everything: A survey

    Brian B Moser, Arundhati S Shanbhag, Federico Raue, Stanislav Frolov, Sebastian Palacio, and Andreas Dengel. Diffusion models, image super-resolution and everything: A survey. arXiv preprint arXiv:2401.00736, 2024

    2024 arXiv

  3. [11]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020

    2020

  4. [12]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part ...

    2015

  5. [13]

    Attention u-net: Learning where to look for the pancreas

    Ozan Oktay, Jo Schlemper, Loic Le Folgoc, Matthew Lee, Mattias Heinrich, Kazunari Misawa, Kensaku Mori, Steven McDonagh, Nils Y Hammerla, Bernhard Kainz, et al. Attention u-net: Learning where to look for the pancreas. arXiv preprint arXiv:1804.03999, 2018

    2018 arXiv

  6. [14]

    Ganimation: Anatomically-aware facial animation from a single image

    Albert Pumarola, Antonio Agudo, Aleix M Martinez, Alberto Sanfeliu, and Francesc Moreno-Noguer. Ganimation: Anatomically-aware facial animation from a single image. InProceedings of the European conference on computer vision (ECCV), pages 818–833, 2018

    2018

  7. [15]

    Combining gan with reverse correlation to construct personalized facial expressions

    Sen Yan, Catherine Soladié, Jean-Julien Aucouturier, and Renaud Seguier. Combining gan with reverse correlation to construct personalized facial expressions. Plos one, 18(8):e0290612, 2023

    2023

  8. [16]

    Exploring mental prototypes by an efficient interdisciplinary approach: Interactive microbial genetic algorithm

    Sen Yan, Catherine Soladié, and Renaud Seguier. Exploring mental prototypes by an efficient interdisciplinary approach: Interactive microbial genetic algorithm. In 2023 IEEE 17th International Conference on Automatic Face and Gesture Recognition (FG), pages 1–8. IEEE, 2023. 11...

    2023

  9. [17]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018

    2018 arXiv

  10. [18]

    Llama: Open and efficient foundation language models

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023

    2023 arXiv

  11. [19]

    Wavenet: A generative model for raw audio

    Aaron Van Den Oord, Sander Dieleman, Heiga Zen, Karen Simonyan, Oriol Vinyals, Alex Graves, Nal Kalch- brenner, Andrew Senior, Koray Kavukcuoglu, et al. Wavenet: A generative model for raw audio. arXiv preprint arXiv:1609.03499, 12, 2016

    2016 arXiv

  12. [20]

    Diffwave: A versatile diffusion model for audio synthesis

    Zhifeng Kong, Wei Ping, Jiaji Huang, Kexin Zhao, and Bryan Catanzaro. Diffwave: A versatile diffusion model for audio synthesis. arXiv preprint arXiv:2009.09761, 2020

    2009 arXiv

  13. [22]

    Srdiff: Single image super-resolution with diffusion probabilistic models

    Haoying Li, Yifan Yang, Meng Chang, Shiqi Chen, Huajun Feng, Zhihai Xu, Qi Li, and Yueting Chen. Srdiff: Single image super-resolution with diffusion probabilistic models. Neurocomputing, 479:47–59, 2022

    2022

  14. [23]

    A style-based generator architecture for generative adversarial networks

    Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 4401–4410, 2019

    2019

  15. [24]

    Generative adversarial networks

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks. Communications of the ACM, 63(11):139–144, 2020

    2020

  16. [25]

    Auto-encoding variational bayes

    Diederik P Kingma. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013

    2013 arXiv

  17. [26]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pa...

    2021

  18. [27]

    Repaint: Inpainting using denoising diffusion probabilistic models

    Andreas Lugmayr, Martin Danelljan, Andres Romero, Fisher Yu, Radu Timofte, and Luc Van Gool. Repaint: Inpainting using denoising diffusion probabilistic models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11461–11471, 2022

    2022

  19. [28]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022

    2022

  20. [29]

    Glide: Towards photorealistic image generation and editing with text-guided diffusion models

    Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya Sutskever, and Mark Chen. Glide: Towards photorealistic image generation and editing with text-guided diffusion models. arXiv preprint arXiv:2112.10741, 2021

    2021 arXiv

  21. [30]

    Hierarchical text-conditional image generation with clip latents

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image generation with clip latents. arXiv preprint arXiv:2204.06125, 1(2):3, 2022

    2022 arXiv

  22. [31]

    Diffusion models beat gans on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. Advances in neural information processing systems, 34:8780–8794, 2021

    2021

  23. [32]

    Image super-resolution via iterative refinement

    Chitwan Saharia, Jonathan Ho, William Chan, Tim Salimans, David J Fleet, and Mohammad Norouzi. Image super-resolution via iterative refinement. IEEE transactions on pattern analysis and machine intelligence , 45(4):4713–4726, 2022

    2022

  24. [33]

    Attention is all you need

    A Vaswani. Attention is all you need. Advances in Neural Information Processing Systems, 2017

    2017

  25. [34]

    Image restoration with mean-reverting stochastic differential equations

    Ziwei Luo, Fredrik K Gustafsson, Zheng Zhao, Jens Sjölund, and Thomas B Schön. Image restoration with mean-reverting stochastic differential equations. arXiv preprint arXiv:2301.11699, 2023

    2023 arXiv

  26. [35]

    Hierarchical integration diffusion model for realistic image deblurring

    Zheng Chen, Yulun Zhang, Ding Liu, Jinjin Gu, Linghe Kong, Xin Yuan, et al. Hierarchical integration diffusion model for realistic image deblurring. Advances in neural information processing systems, 36, 2024

    2024

  27. [36]

    Super-resolved nuclear magnetic resonance spectroscopy

    Satish Mulleti, Amrinder Singh, Varsha P Brahmkhatri, Kousik Chandra, Tahseen Raza, Sulakshana P Mukherjee, Chandra Sekhar Seelamantula, and Hanudatta S Atreya. Super-resolved nuclear magnetic resonance spectroscopy. Scientific reports, 7(1):9651, 2017

    2017

  28. [37]

    Sampling signals with finite rate of innovation

    Martin Vetterli, Pina Marziliano, and Thierry Blu. Sampling signals with finite rate of innovation. IEEE transactions on Signal Processing, 50(6):1417–1428, 2002

    2002

  29. [38]

    Super-resolution nmr spectroscopy

    Luca Wenchel, Olivia Gampp, and Roland Riek. Super-resolution nmr spectroscopy. Journal of Magnetic Resonance, 366:107746, 2024. 12 DiffNMR3

    2024

  30. [39]

    Laion-400m: Open dataset of clip-filtered 400 million image-text pairs

    Christoph Schuhmann, Richard Vencu, Romain Beaumont, Robert Kaczmarczyk, Clayton Mullis, Aarush Katta, Theo Coombes, Jenia Jitsev, and Aran Komatsuzaki. Laion-400m: Open dataset of clip-filtered 400 million image-text pairs. arXiv preprint arXiv:2111.02114, 2021

    2021 arXiv

  31. [40]

    Rapid protein assignments and structures from raw nmr spectra with the deep learning technique artina

    Piotr Klukowski, Roland Riek, and Peter Güntert. Rapid protein assignments and structures from raw nmr spectra with the deep learning technique artina. Nature Communications, 13(1):6151, 2022

    2022

  32. [41]

    Announcing the worldwide protein data bank

    Helen Berman, Kim Henrick, and Haruki Nakamura. Announcing the worldwide protein data bank. Nature structural & molecular biology, 10(12):980–980, 2003

    2003

  33. [42]

    Fast pure shift nmr spectroscopy using attention-assisted deep neural network

    Haolin Zhan, Jiawei Liu, Qiyuan Fang, Xinyu Chen, Yang Ni, and Lingling Zhou. Fast pure shift nmr spectroscopy using attention-assisted deep neural network. Advanced Science, page 2309810, 2024

    2024

  34. [43]

    Accelerated nuclear magnetic resonance spectroscopy with deep learning

    Xiaobo Qu, Yihui Huang, Hengfa Lu, Tianyu Qiu, Di Guo, Tatiana Agback, Vladislav Orekhov, and Zhong Chen. Accelerated nuclear magnetic resonance spectroscopy with deep learning. Angewandte Chemie, 132(26):10383– 10386, 2020

    2020

  35. [44]

    Fid-net: A versatile deep neural network architecture for nmr spectral reconstruction and virtual decoupling

    Gogulan Karunanithy and D Flemming Hansen. Fid-net: A versatile deep neural network architecture for nmr spectral reconstruction and virtual decoupling. Journal of biomolecular NMR, 75(4):179–191, 2021

    2021

  36. [45]

    Fast acquisition of high-quality nuclear magnetic resonance pure shift spectroscopy via a deep neural network

    Xiaoxu Zheng, Zhengxian Yang, Chuang Yang, Xiaoqi Shi, Yao Luo, Jie Luo, Qing Zeng, Yanqin Lin, and Zhong Chen. Fast acquisition of high-quality nuclear magnetic resonance pure shift spectroscopy via a deep neural network. The Journal of Physical Chemistry Letters, 13(9):2101–...

    2022

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.