Astrophotography turbulence mitigation via generative models

Joonyeoup Kim; Stanley Chan; Xijun Wang; Xingguang Zhang; Yu Yuan

REVIEW 4 major objections 6 minor 31 references

Astrophotography turbulence mitigation via generative models

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read AstroDiff, a dual-branch generative model, claims to beat existing learning-based methods on severe astronomical turbulence by fusing a diffusion prior with a conditional restoration branch via stochastic gradient Langevin dynamics.

desk verdict The datasets and LPIPS gains are real, but the SOTA claim overreaches: head-to-head numbers are synthetic-only, real-data has no baselines, and the tables disagree. read the letter →

arxiv 2506.02981 v1 pith:G2KDYOEF submitted 2025-06-03 cs.CV eess.IV

Joonyeoup Kim , Yu Yuan , Xingguang Zhang , Xijun Wang , Stanley Chan This is my paper

classification cs.CVeess.IV

keywords atmosphericturbulencemitigationastronomicalimagerestorationdiffusionmodelsBayesianstochasticgradientLangevindynamicsperceptualqualityluckyimaginggenerativepriors

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 proposes AstroDiff, a single-image generative method for removing atmospheric turbulence from astronomical photographs. It combines a diffusion prior trained only on clean planet images with a diffusion restoration branch trained on synthetic turbulence pairs, then fuses the two through stochastic gradient Langevin dynamics. The authors claim this dual-branch design preserves global structure under severe turbulence where deterministic methods over-smooth or create artifacts, and that it achieves better perceptual quality (LPIPS) and no-reference quality (BRISQUE) than existing learning-based baselines, with a mixed PSNR that is lower in low and medium turbulence. The motivation is to replace the heavy data acquisition and manual processing of lucky imaging with an automated single-frame restoration pipeline.

What carries the argument

The load-bearing mechanism is the SGLD-based fusion of two separately pre-trained diffusion branches within a Bayesian diffusion formulation. The prior branch is a U-Net (a convolutional encoder–decoder) denoising diffusion model trained on 22,512 clean astronomical images, so it knows what planets should look like; the restoration branch is a conditional diffusion model trained on 110,745 paired clean/turbulent images, so it knows how to invert the degradation. At inference, each SGLD step combines the restoration branch's data-driven gradient and the prior branch's generative gradient with additive Langevin noise, producing a sample from the joint posterior rather than a deterministic regression.

What would settle it

A field test would settle it: record a long high-frame-rate burst of a bright planet, build a lucky-imaging reference from the sharpest frames, then run AstroDiff on single frames from the same burst; if the restored single frames do not at least match the lucky-imaging reference in no-reference quality and resolved detail, the claim that AstroDiff is the best turbulence mitigation model for astrophotography is falsified.

Watch

Extended reading notes

Core claim

In its own terms, AstroDiff is a Bayesian diffusion restoration model that writes the posterior as p(y|x) ∝ p^γ(y|x) p(y), with a conditional restoration branch supplying the likelihood and an unconditional generative prior branch supplying the prior over astronomical scenes. Inference alternates gradient descent on both terms with injected noise via SGLD, so the output must simultaneously fit the observed turbulent frame and look like a plausible planet or moon. Evaluated on the AstroEVA benchmark of 1,907 real telescope captures plus simulated turbulence at nine $C_n^{2}$ levels, the paper reports that AstroDiff achieves the lowest LPIPS in every turbulence regime (e.g., 0.07 in high turbulence vs. 0.11–0.18 for the baselines) and improves BRISQUE from 95.64 to 22.78 on real data. PSNR is mixed—lower than the best baseline in low and medium regimes, highest in high turbulence—which the authors attribute to the usual perceptual-versus-pixel-accuracy trade-off.

Load-bearing premise

Every synthetic training pair and the AstroEVA turbulence benchmark are generated with the simulator of reference [4]; if that simulator does not faithfully reproduce real atmospheric seeing, the reported LPIPS and BRISQUE gains may not appear on real telescope images.

Editorial extensions

If this is right

Turbulence mitigation becomes a single-image, automated operation: no hundreds of high-frame-rate captures, no lucky-frame selection, no manual registration.
In severe turbulence, AstroDiff retains global coherence and structural detail, where the paper shows deterministic baselines producing artifacts or over-smoothed reconstructions.
The perceptual and no-reference gains are strongest by LPIPS and BRISQUE; applications that need pixel-level fidelity (e.g., astrometry) face a lower PSNR in low and medium regimes.
The released datasets (PlanetSYN, TechSYN, AstroEVA) give future astronomical restoration work a common training and evaluation ground.

Reading between the lines

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

Editorial inference: the prior/restoration split is not astronomy-specific, so the same SGLD fusion could be applied to other degradations with scarce clean data, such as underwater imaging or low-light microscopy, though the paper does not test this.
Editorial inference: the real-data BRISQUE improvement does not establish that restored images preserve scientifically useful detail; a user study or a task such as measuring resolved surface features on Jupiter would be needed to know whether the perceptual gain carries astronomical information.
Editorial inference: since every synthetic pair comes from one turbulence simulator, the method's practical ceiling is set by how well that simulator captures anisoplanatic seeing; a direct single-frame versus lucky-imaging comparison on a real telescope with measured turbulence would be the natural next test.

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.

Desk Editor's Note

read the letter

Plainly: the paper's real contribution is three new datasets for astrophotography turbulence and the empirical observation that a dual-branch diffusion prior plus restoration model, fused with SGLD, gives better LPIPS and BRISQUE on simulated astronomical turbulence than deterministic baselines. That is worth a look. But the headline claim that it 'outperforms existing state-of-the-art' is not established. The only head-to-head quantitative comparison (Table 3) runs on synthetic images generated by the Chan et al. simulator, which is also the source of two of the three baselines. The real telescope captures are scored only with no-reference BRISQUE, with no baseline run, so transfer to real seeing is unverified. The internal numbers are also inconsistent: Table 3 reports AstroDiff at 30.22/30.86 PSNR for medium/high turbulence, while the same 'Two Step Model' in Table 4 shows 30.02/29.86. The abstract and conclusion should be qualified until that is resolved.

What is good: the datasets are concrete and could be reused; the method is simple to understand; the LPIPS improvements are consistent and the paper honestly acknowledges the PSNR trade-off. The architecture itself is a direct adaptation of Bayesian Diffusion Models, so the novelty is not in the math but in the domain-specific training and the empirical demonstration. That is a legitimate, if modest, contribution.

Soft spots beyond the table discrepancy: no error bars anywhere, so we cannot tell if the single-digit LPIPS differences are meaningful; key inference hyperparameters (fusion weights, SGLD step schedule, number of refinement steps, and the likelihood exponent gamma) are not reported; the evaluation set is narrow—one Celestron telescope, mostly Jupiter images—so generalizability is unclear. The authors also claim code is available at a Vercel page; that is not a stable archival link. The self-citation of the simulator is not a flaw by itself, but the evaluation is heavily dependent on that simulator's fidelity.

Who should read this: people working on turbulence mitigation or diffusion-based image restoration will find the datasets and the negative result (perceptual gains without pixel gains) informative. It deserves a serious referee, but it needs major revision: release code and data, add error bars, specify the inference details, fix the tables, and soften the claims to match the evidence. As it stands, the evidence supports 'competitive LPIPS on simulated turbulence,' not 'best model for astrophotography.'

Referee Report

4 major / 6 minor

Summary. The paper proposes AstroDiff, a two-branch generative framework for mitigating atmospheric turbulence in astrophotography. One branch is an unconditional diffusion prior trained on clean astronomical images; the other is a conditional diffusion restoration branch trained on turbulence-degraded/clean pairs. The two branches are combined at inference time through a stochastic-gradient-Langevin-dynamics (SGLD) fusion that blends generative and restoration gradients. The authors introduce three datasets: PlanetSYN and TechSYN for training and AstroEVA, a benchmark containing real telescope captures plus turbulence-simulated ground-truth pairs. Quantitative evaluation reports PSNR, LPIPS, and BRISQUE against DATUM, TMT, and ESTRNN, and the paper claims state-of-the-art perceptual quality and structural fidelity, especially under severe turbulence.

Significance. If the claims are supported, the paper makes a useful contribution: it is an early application of generative diffusion priors to astrophotography-specific turbulence mitigation, and the proposed PlanetSYN/TechSYN training datasets and the AstroEVA evaluation set could be valuable assets to the community. The SGLD-based fusion of a domain-specific prior with a restoration branch is a sensible idea, and the reported LPIPS and BRISQUE improvements over the compared baselines are encouraging. However, the significance is substantially tempered by the limited and internally inconsistent quantitative evidence, the absence of error bars or significance tests, and the fact that the only head-to-head SOTA comparison is conducted on synthetic turbulence generated by the same simulator used to build the benchmark. The real-telescope evaluation is purely no-reference and includes no baseline methods.

major comments (4)

[§4.3, §4.5, Tables 3 and 4] The same final model appears as 'AstroDiff (ours)' in Table 3 with medium/high PSNR 30.22/30.86 and as 'Two Step Model' in Table 4 with medium/high PSNR 30.02/29.86. This internal inconsistency means the reported quantitative results cannot both be correct as presented. Please resolve the discrepancy and report a single, consistent set of numbers, ideally with standard deviations or confidence intervals over the AstroEVA subsets.
[§4.2, §4.3, Table 2] All state-of-the-art comparisons in Table 3 are on AstroEVA images degraded by the turbulence simulator of Chan et al. [4]. The real-telescope evaluation in Table 2 reports only AstroDiff's BRISQUE before/after with no baseline methods. Therefore the abstract's unqualified claim that AstroDiff 'outperforms existing state-of-the-art learning-based methods' is not demonstrated on real atmospheric seeing, which is the target domain. Either run the baselines on the real captures with a no-reference metric or carefully restrict the claim to simulated turbulence.
[§3.3, Eq. (2)] The SGLD update in Eq. (2) is written as y_{t+1} = (eta_t/2)[-∇log ρ_γ(y_t) - ∇log ρ_γ(y_t | x)] + sqrt(eta_t) eta_t, which omits the current iterate y_t on the right-hand side; it is therefore not the standard SGLD update. In addition, the fusion weights, the step-size schedule eta_t, the number of refinement iterations, and the likelihood exponent gamma are never specified. These hyperparameters are essential for reproducibility and for the claim that the fusion procedure is principled. Please correct the equation and report all fusion hyperparameters.
[§3.2, Table 1] The text states that BRISQUE was used to categorize turbulence into weak, medium, and high levels, but Table 1 lists only subject/C_n^2 assignments and no BRISQUE thresholds, scores, or ranges. This makes the grouping used in Table 3 non-reproducible. Please clarify how the categories were obtained and provide the actual BRISQUE intervals for each category.

minor comments (6)

[Abstract and Conclusion] The conclusion's phrase 'making it the best turbulence mitigation model for astrophotography' is too strong given the limited benchmark coverage and the internal inconsistencies in the quantitative tables; please temper the claim to match the demonstrated scope.
[§3.1, Eq. (1)] The notation p^γ(y|x) on the right-hand side of Eq. (1) is confusing: in a posterior p(y|x) ∝ p^γ(x|y) p(y), the likelihood should be p(x|y), not p(y|x). Please clarify the notation.
[Throughout] There are numerous typographical inconsistencies: 'AstroEV A' appears with an inconsistent space, Table 1 uses 'AstroEv', the references contain 'V AEs' and 'F oundations', and Eq. (2) uses both ρ and p for the same distribution. A careful proofreading pass is needed.
[Fig. 2 and Fig. 4] The labels 'AstroDiff (Only Restoration Branch)' and 'AstroDiff (Final Architecture)' in Figure 4 are confusing because the restoration branch alone is not the full AstroDiff method. Please relabel the figures to distinguish the one-step restoration branch from the two-step fused model.
[§4.1] The implementation details omit the total number of parameters, training hardware, and inference time, which are useful for assessing practicality and reproducibility.
[Data availability] The code and results URL (https://web-six-kappa-66.vercel.app/) does not appear to be a standard repository link; please provide a stable GitHub or institutional repository address.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the method is an empirical combination of published diffusion/SGLD components, and its evaluation, though largely based on the authors' own simulator and baselines, does not reduce to a fitted input or self-cited theorem.

full rationale

The paper does not claim a first-principles derivation; it claims an empirical improvement. The two equations presented (Eq. 1 for the Bayesian posterior and Eq. 2 for the SGLD update) are standard published formulations from refs. [20] and [21], used as building blocks rather than derived from, or equivalent to, the paper's own outputs. The dual-branch architecture is not fitted to the evaluation target: the prior branch is trained on PlanetSYN, the restoration branch on TechSYN, and the fusion weights are fixed; no parameter is fitted on AstroEVA and then reported as a prediction. The state-of-the-art claim is a benchmark comparison computed with external metrics (PSNR, LPIPS, BRISQUE) on AstroEVA images degraded by the Chan et al. turbulence simulator [4]. That simulator is a prior published physical model with stated turbulence parameters (C_n^2 from 5e-16 to 3e-13), and its output does not encode AstroDiff's results, so the comparison is not circular. The overlap of the authors with the simulator [4] and with baselines [3] and [4] is a real external-validity concern, and the real-telescope BRISQUE evaluation in Table 2 lacks any baseline comparison, but that is a methodological limitation, not circular reasoning. The numeric inconsistency between the 'AstroDiff' rows of Table 3 and the 'Two Step Model' rows of Table 4 is a reporting inconsistency, not a circular step. No uniqueness theorem is imported from self-citations, no ansatz is smuggled in via citation, and no known empirical pattern is renamed as a new result. Accordingly, the derivation and central claim are self-contained rather than circular.

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

The paper introduces no new physical entities or conserved quantities. Its central claim rests on standard learning architecture and on domain assumptions about the validity of the turbulence simulator and quality metrics. The main unstated inputs are the fusion and inference hyperparameters, which are listed as free parameters.

free parameters (4)

Fusion weights for combining generative and restoration gradients
In Stage 2, the two gradient terms in Equation (2) are blended with 'fixed weights' but the values are never reported. The balance between generative prior and data fidelity directly controls output sharpness versus fidelity.
SGLD step size schedule eta_t
The update in Equation (2) depends on an unspecified noise and step-size schedule eta_t over iterations; this schedule affects convergence and final image quality, but is not described.
Number of SGLD refinement steps
The inference loop repeats 'until y_{t+1} transitions from noise to a coherent reconstruction', but no stopping criterion or step count is given, making the method's behavior depend on an unreported choice.
Likelihood exponent gamma in p_gamma(y|x)
Equation (1) introduces a power gamma on the likelihood, but its value is never defined. This exponent is a hand-chosen parameter governing how strongly the restored image must match the input.

assumptions (4)

domain assumption The turbulence simulator of Chan et al. (2024) accurately reproduces real atmospheric turbulence degradation for both training and evaluation.
All synthetic pairs and the AstroEVA benchmark rely on this simulator (Section 3.2). If the simulator is not faithful, the quantitative comparisons do not reflect real-world performance.
domain assumption Diffusion models trained on the constructed PlanetSYN and TechSYN datasets capture the distribution of real astronomical images needed for restoration.
The generative prior is pre-trained solely on synthetic and curated planet-like images (Section 3.3, Stage 1.1), and the paper assumes these transfer to real telescopes.
domain assumption BRISQUE and LPIPS are valid quality metrics for astronomical images.
These metrics are used to categorize turbulence levels and to claim higher perceptual quality (Sections 3.2, 4.2, 4.3). Both are trained on natural images and may not align with astronomical image fidelity.
standard math The Bayesian diffusion framework of Xu et al. (2024) is correctly applied in the proposed fusion.
The paper adopts the SGLD fusion from prior work without re-derivation; the update in Equation (2) assumes the score functions of the prior and likelihood are correctly combined.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Astrophotography turbulence mitigation via generative models." pith.science (2026). https://pith.science/paper/G2KDYOEF

@misc{pith2026250602981,
  author       = {Pith},
  title        = {Pith review of: Astrophotography turbulence mitigation via generative models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G2KDYOEF}},
  note         = {Machine review of arXiv:2506.02981}
}

read the original abstract

Photography is the cornerstone of modern astronomical and space research. However, most astronomical images captured by ground-based telescopes suffer from atmospheric turbulence, resulting in degraded imaging quality. While multi-frame strategies like lucky imaging can mitigate some effects, they involve intensive data acquisition and complex manual processing. In this paper, we propose AstroDiff, a generative restoration method that leverages both the high-quality generative priors and restoration capabilities of diffusion models to mitigate atmospheric turbulence. Extensive experiments demonstrate that AstroDiff outperforms existing state-of-the-art learning-based methods in astronomical image turbulence mitigation, providing higher perceptual quality and better structural fidelity under severe turbulence conditions. Our code and additional results are available at https://web-six-kappa-66.vercel.app/

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

31 extracted references · 25 canonical work pages

[4]

EXPERIMENTS InputReconstructedInputReconstructed Fig. 5: Sample images of Mars and Jupiter in AstroEV A (first column) vs enhanced Image of Mars and Jupiter using As- troDiff (second column) Method BRISQUE↓ Original Images 95.64 Enhanced Images 22.78 Table 2: BRISQUE evaluation on real data (lower is better). Reconstruction achieves76.2%improvement. 4.1. ...

work page
[1]

They are primar- ily due to changes in temperature and pressure, which cause the refractive index of the air to change [1, 2, 3, 4, 5]

INTRODUCTION Atmospheric turbulence is a phenomenon rooted from ran- dom fluctuations in the Earth’s atmosphere. They are primar- ily due to changes in temperature and pressure, which cause the refractive index of the air to change [1, 2, 3, 4, 5]. These fluctuations induce both blurring from high-frequency light scattering and image jitter from large-sca...

work page
[2]

RELATED WORK 2.1. Lucky Imaging for Astrophotography Turbulence Mitigation Lucky imaging is a well-established astrophotography tech- nique that leverages high-speed cameras to capture thousands of short-exposure frames, taking advantage of brief moments when atmospheric turbulence is minimal [6, 7, 10]. Dur- ing these optimal intervals, a subset of frame...

work page Pith review arXiv 2025
[3]

METHODS In this section, we will elaborate on the theoretical founda- tions, the constructed datasets, as well as the designed net- work with training methodology. 3.1. Preliminary In the following sections, we present a unique network in- spired by Bayesian diffusion models [20] that integrates a generative prior branch with a restoration branch tailored...

work page
[5]

AstroDiff is built by combining a gen- erative prior branch and a restoration branch, and fused with SGLD inference

CONCLUSION In this paper, We present AstroDiff, a new network designed for astrophotography. AstroDiff is built by combining a gen- erative prior branch and a restoration branch, and fused with SGLD inference. Experimental results demonstrate effective atmospheric turbulence removal and structural preservation of our architecture, making it the best turbu...

work page
[6]

The point spread function in lucky imaging and variations in seeing on short timescales,

J. E. Baldwin, P. J. Warner, and C. D. Mackay, “The point spread function in lucky imaging and variations in seeing on short timescales,”A&A, vol. 480, no. 2, pp. 589–597, Mar. 2008

work page 2008
[7]

Neutralizing the impact of atmospheric turbulence on complex scene imaging via deep learn- ing,

D. Jin, Y . Chen, Y . Lu, J. Chen, P. Wang, Z. Liu, S. Guo, and X. Bai, “Neutralizing the impact of atmospheric turbulence on complex scene imaging via deep learn- ing,”Nature Machine Intelligence, vol. 3, pp. 876 – 884, 2021

work page 2021
[8]

Physics-driven turbulence image restoration with stochastic refinement,

A. Jaiswal, X. Zhang, S. H. Chan, and Z. Wang, “Physics-driven turbulence image restoration with stochastic refinement,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2023, pp. 12 170–12 181

work page 2023

Show all 31 references

[9]

Imag- ing through the atmosphere using turbulence mitigation transformer,

X. Zhang, Z. Mao, N. Chimitt, and S. H. Chan, “Imag- ing through the atmosphere using turbulence mitigation transformer,”arXiv preprint arXiv: 2207.06465, 2023

2023 arXiv
[10]

Spatio-temporal turbulence mitigation: A translational perspective,

X. Zhang, N. Chimitt, Y . Chi, Z. Mao, and S. H. Chan, “Spatio-temporal turbulence mitigation: A translational perspective,”arXiv preprint arXiv: 2401.04244, 2024

2024 arXiv
[11]

often generate artifacts or overly smoothed reconstruc- tions. In contrast, AstroDiff’s dual branch design enforces both restoration and generative constraints: the learned prior steers outputs toward physically plausible structures, while the restoration branch ensures adhere...
[12]

Spatial correlation of zernike phase-expansion coefficients for atmospheric turbulence with finite outer scale,

N. Takato and I. Yamaguchi, “Spatial correlation of zernike phase-expansion coefficients for atmospheric turbulence with finite outer scale,”J. Opt. Soc. Am. A, vol. 12, no. 5, pp. 958–963, May 1995

1995
[13]

Seeing mt. rainier: Lucky imaging for multi-image denoising, sharpening, and haze removal,

N. Joshi and M. Cohen, “Seeing mt. rainier: Lucky imaging for multi-image denoising, sharpening, and haze removal,” in2010 IEEE International Conference on Computational Photography (ICCP), 2010

2010
[14]

Christopher, R

A. Christopher, R. Hari Kishan, and P. Sudeep,Image Reconstruction Using Deep Learning. John Wiley & Sons, Inc., 2023, pp. 65–87

2023
[15]

A survey of deep learning approaches to image restoration,

J. Su, B. Xu, and H. Yin, “A survey of deep learning approaches to image restoration,”Neurocomputing, vol. 487, pp. 46–65, 2022

2022
[16]

Removing atmospheric turbu- lence via space-invariant deconvolution,

X. Zhu and P. Milanfar, “Removing atmospheric turbu- lence via space-invariant deconvolution,”IEEE Trans- actions on Pattern Analysis and Machine Intelligence, vol. 35, no. 1, pp. 157–170, 2013

2013
[17]

Image recon- struction of static and dynamic scenes through anisopla- natic turbulence,

Z. Mao, N. Chimitt, and S. H. Chan, “Image recon- struction of static and dynamic scenes through anisopla- natic turbulence,”IEEE Transactions on Computational Imaging, vol. 6, pp. 1415–1428, 2020

2020
[18]

Luckycam,

“Luckycam,” https://www.not.iac.es/instruments/ luckycam/, accessed: February 05, 2025

2025
[19]

An introduction to vari- ational autoencoders,

D. P. Kingma and M. Welling, “An introduction to vari- ational autoencoders,”F oundations and Trends® in Ma- chine Learning, vol. 12, no. 4, p. 307–392, 2019

2019
[20]

Generative adversarial networks,

I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Ben- gio, “Generative adversarial networks,”arXiv preprint arXiv: 1406.2661, 2014

2014 arXiv
[21]

Deep unsupervised learning using nonequi- librium thermodynamics,

J. Sohl-Dickstein, E. Weiss, N. Maheswaranathan, and S. Ganguli, “Deep unsupervised learning using nonequi- librium thermodynamics,” inProceedings of the 32nd International Conference on Machine Learning, vol. 37, 2015, pp. 2256–2265

2015
[22]

Denoising diffusion prob- abilistic models,

J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion prob- abilistic models,”arXiv preprint arXiv: 2006.11239, 2020

2006 arXiv
[23]

Image super-resolution via iterative refinement,

C. Saharia, J. Ho, W. Chan, T. Salimans, D. J. Fleet, and M. Norouzi, “Image super-resolution via iterative refinement,”arXiv preprint arXiv: 2104.07636, 2021

2021 arXiv
[24]

Denois- ing diffusion restoration models,

B. Kawar, M. Elad, S. Ermon, and J. Song, “Denois- ing diffusion restoration models,”arXiv preprint arXiv: 2201.11793, 2022

2022 arXiv
[25]

Multiscale structure guided diffusion for im- age deblurring,

M. Ren, M. Delbracio, H. Talebi, G. Gerig, and P. Mi- lanfar, “Multiscale structure guided diffusion for im- age deblurring,” inProceedings of the IEEE/CVF In- ternational Conference on Computer Vision, 2023, pp. 10 721–10 733

2023
[26]

Bayesian diffusion models for 3d shape re- construction,

H. Xu, Y . Lei, Z. Chen, X. Zhang, Y . Zhao, Y . Wang, and Z. Tu, “Bayesian diffusion models for 3d shape re- construction,”arXiv preprint arXiv: 2403.06973, 2024

2024 arXiv
[27]

Bayesian learning via stochastic gradient langevin dynamics,

M. Welling and Y . W. Teh, “Bayesian learning via stochastic gradient langevin dynamics,” inInternational Conference on Machine Learning, 2011

2011
[28]

No- reference image quality assessment in the spatial do- main,

A. Mittal, A. K. Moorthy, and A. C. Bovik, “No- reference image quality assessment in the spatial do- main,”IEEE Transactions on Image Processing, vol. 21, pp. 4695–4708, 2012

2012
[29]

U-Net: Con- volutional networks for biomedical image segmenta- tion,

O. Ronneberger, P. Fischer, and T. Brox, “U-Net: Con- volutional networks for biomedical image segmenta- tion,”arXiv preprint arXiv: 1505.04597, 2015

2015 arXiv
[30]

Real-world video deblurring: A benchmark dataset and an efficient recurrent neural network,

Z. Zhong, Y . Gao, Y . Zheng, B. Zheng, and I. Sato, “Real-world video deblurring: A benchmark dataset and an efficient recurrent neural network,”arXiv preprint arXiv: 2106.16028, 2022

2022 arXiv
[31]

The unreasonable effectiveness of deep fea- tures as a perceptual metric,

R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep fea- tures as a perceptual metric,”arXiv preprint arXiv: 1801.03924, 2018

2018 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [4]

EXPERIMENTS InputReconstructedInputReconstructed Fig. 5: Sample images of Mars and Jupiter in AstroEV A (first column) vs enhanced Image of Mars and Jupiter using As- troDiff (second column) Method BRISQUE↓ Original Images 95.64 Enhanced Images 22.78 Table 2: BRISQUE evaluation on real data (lower is better). Reconstruction achieves76.2%improvement. 4.1. ...

work page

[2] [1]

They are primar- ily due to changes in temperature and pressure, which cause the refractive index of the air to change [1, 2, 3, 4, 5]

INTRODUCTION Atmospheric turbulence is a phenomenon rooted from ran- dom fluctuations in the Earth’s atmosphere. They are primar- ily due to changes in temperature and pressure, which cause the refractive index of the air to change [1, 2, 3, 4, 5]. These fluctuations induce both blurring from high-frequency light scattering and image jitter from large-sca...

work page

[3] [2]

RELATED WORK 2.1. Lucky Imaging for Astrophotography Turbulence Mitigation Lucky imaging is a well-established astrophotography tech- nique that leverages high-speed cameras to capture thousands of short-exposure frames, taking advantage of brief moments when atmospheric turbulence is minimal [6, 7, 10]. Dur- ing these optimal intervals, a subset of frame...

work page Pith review arXiv 2025

[4] [3]

METHODS In this section, we will elaborate on the theoretical founda- tions, the constructed datasets, as well as the designed net- work with training methodology. 3.1. Preliminary In the following sections, we present a unique network in- spired by Bayesian diffusion models [20] that integrates a generative prior branch with a restoration branch tailored...

work page

[5] [5]

AstroDiff is built by combining a gen- erative prior branch and a restoration branch, and fused with SGLD inference

CONCLUSION In this paper, We present AstroDiff, a new network designed for astrophotography. AstroDiff is built by combining a gen- erative prior branch and a restoration branch, and fused with SGLD inference. Experimental results demonstrate effective atmospheric turbulence removal and structural preservation of our architecture, making it the best turbu...

work page

[6] [6]

The point spread function in lucky imaging and variations in seeing on short timescales,

J. E. Baldwin, P. J. Warner, and C. D. Mackay, “The point spread function in lucky imaging and variations in seeing on short timescales,”A&A, vol. 480, no. 2, pp. 589–597, Mar. 2008

work page 2008

[7] [7]

Neutralizing the impact of atmospheric turbulence on complex scene imaging via deep learn- ing,

D. Jin, Y . Chen, Y . Lu, J. Chen, P. Wang, Z. Liu, S. Guo, and X. Bai, “Neutralizing the impact of atmospheric turbulence on complex scene imaging via deep learn- ing,”Nature Machine Intelligence, vol. 3, pp. 876 – 884, 2021

work page 2021

[8] [8]

Physics-driven turbulence image restoration with stochastic refinement,

A. Jaiswal, X. Zhang, S. H. Chan, and Z. Wang, “Physics-driven turbulence image restoration with stochastic refinement,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2023, pp. 12 170–12 181

work page 2023