$\gamma$-Bridge: A Look-Parametric Diffusion Bridge

Jilong Li; Tengxi Wang; Wufan Zhao; Xuran Hu; Yujie Zhu

REVIEW 4 major objections 5 minor 44 references

$\gamma$-Bridge: A Look-Parametric Diffusion Bridge

T0 review · 4 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read A single diffusion network, trained only on clean natural images corrupted with single-look synthetic Gamma speckle, can restore SAR images at any input and output look level, enabling zero-shot despeckling across multiple real radar sensor

desk verdict Solid bridge construction and honest experiments, but the full-grid claim rests on NFE=5 results and there is a table inconsistency that must be fixed before the numbers can be trusted. read the letter →

arxiv 2607.22719 v1 pith:FMN43CX7 submitted 2026-07-21 cs.CV

Xuran Hu , Yujie Zhu , Tengxi Wang , Jilong Li , Wufan Zhao This is my paper

classification cs.CV

keywords SARdespecklingmultiplicativeGammanoisediffusionbridgelooknumberzero-shotrestorationGamma–Lévydecompositionimageconsistencytraining

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

γ-Bridge makes the noise level of SAR imagery an explicit, physically meaningful control. By indexing a diffusion bridge's time axis with the equivalent number of looks L—the statistic that defines speckle strength—the paper derives a closed-form Gamma–Lévy reverse step that moves between exact multiplicative Gamma marginals. One network, trained only on clean natural images corrupted with single-look synthetic speckle, can then anchor itself at any input look and stop at any output look, covering the entire $(L_{\text{in}}, L_{\text{out}})$ grid without retraining. If correct, SAR despeckling would need no sensor-specific fine-tuning: an estimated look number plus the same checkpoint would serve multiple spaceborne and airborne sensors, which the paper demonstrates on six real SAR datasets. The paper also reports that deep reverse chains violate its own joint-law premise (ratio mean 0.947 and Kolmogorov–Smirnov p=1.3e-14 at 25 steps), so the headline grid capability is carried by shallow five-step inference.

What carries the argument

The engine is the look-parametric forward bridge, whose time axis $L(t)$ runs from $L_{\max}$ (clean limit) to $L_{\text{obs}}$ (observation). The key identity is Gamma additivity (two independent Gammas sharing a rate sum to a Gamma), which splits a bridge state into a scaled current state plus an independent Gamma increment, giving the closed-form reverse posterior in both stochastic and deterministic versions. The Gamma–Lévy coupling fixes the joint distribution across bridge times so that, with an oracle predictor, the reverse chain exactly inverts the forward chain (Prop. 3). The two-step consistency loss supplies a trajectory-level training signal that remains nonzero at the pointwise $L_1$ optimum, keepin

What would settle it

Run the released checkpoint's deterministic reverse on 64 BSDS500 single-look crops at NFE=25 and compute the residual ratio mean and KS p-value against $\mathrm{Gamma}(1,1)$; the paper's Table VI already reports 0.947 and 1.3e-14, so if one demands ratio mean within a few percent of 1 and p>0.05, the multi-step joint-law claim is falsified at 25 steps. A complementary check for the zero-shot grid claim: compare smart-start at $L_{\text{in}}=16$, $L_{\text{out}}=16$ against a model trained at $L_{\text{obs}}=16$; a substantial PSNR gap would indicate the grid capability is not truly zero-shot.

Watch

Extended reading notes

Core claim

The central discovery is that a single conditioned diffusion bridge whose forward marginals are exactly $x_t = (x_0 / L(t)) \cdot \mathrm{Gamma}(L(t), 1)$ can perform zero-shot restoration over the full admissible $(L_{\text{in}}, L_{\text{out}})$ grid after training only at $L_{\text{obs}} = 1$. Gamma additivity yields a closed-form reverse posterior in both stochastic and deterministic forms: the deterministic update $x_{t-1} = (L(t)/L(t-1)) x_t + (1 - L(t)/L(t-1)) \hat{x}_0$ is an affine mixture between the current state and the network's clean estimate, with the mixture coefficient set entirely by the look schedule. Because every intermediate state is a physically valid $L(t)$-look image, bridge time itself becomes an inference-time con

Load-bearing premise

The reverse chain must keep the state-conditioning pair inside the joint distribution seen in training, but the paper's own Table VI shows this premise fails at 25 steps (ratio mean 0.947, KS p=1.3e-14), so all headline grid results rest on five-step inference.

Editorial extensions

If this is right

A single L_obs=1-trained checkpoint can replace sensor-specific despeckling models: given an estimated look number, the same network serves any admissible (L_in, L_out) pair.
Smart-start is not a cosmetic choice: on synthetic inputs at L_in=4 it beats naive-start by 12.5 dB, and the empirical ratio variance tracks 1/L_in across six look levels.
Every intermediate bridge state is a physically valid L(t)-look image, so the model offers controllable partial despeckling—stopping at L_out=2, 4, 16, etc.—rather than a single fixed endpoint.
The ratio-statistic checks (mean near 1, variance vs. 1/L_in, Beta-coupled reference for target-L) provide a ground-truth-free way to audit intermediate Gamma marginals, which is exactly what is needed for real SAR where clean references do not exist.

Reading between the lines

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

The joint-law mismatch visible at NFE=25 (ratio mean 0.947, KS p=1.3e-14) implies the demonstrated grid capability is bounded in depth; extending to deeper chains would require the randomized-L_obs training the paper defers to future work.
The construction is not specific to Gamma noise: the only probabilistic ingredient is closure under convolution, so the same bridge logic should translate to Poisson noise by replacing Gamma additivity with Poisson additivity, a direction the paper notes but does not develop.
The smart-start in-distribution assumption presumes real SAR follows a single-L Gamma model; the mid-pack airborne results suggest that where real sensors exhibit texture or dark-tail statistics, the ENL estimate anchors the chain at the wrong bridge state, and a learned patch-adaptive look estimator could close the gap without fine-tuning.

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

The new thing here is real: parameterizing a diffusion bridge's time axis by the physical look number L, with independent (L_in, L_out) controls at inference. A single L_obs=1-trained model does appear to denoise across a wide range of input and output looks on synthetic data, and the zero-shot transfer to several real SAR sensors via an ENL estimator is a genuinely useful capability. The math is mostly clean: Prop. 1-2 and Theorem 1 check out, and the Gamma-Levy decomposition is a nice repurposing of the Gamma additivity / Beta-thinning identity that the paper credits to Xie et al. The consistency-loss insight—that pointwise L1 training gives zero gradient at the optimum while a trajectory-level loss still has signal—is well explained and empirically supported by the ablation in Table X. Code is released and limitations are disclosed with unusual candor.

The soft spots are in proportion. First, Table IX reports 21.21 dB for deterministic NFE=5 on the same protocol where Tables VI, VIII, and X all report 22.55-22.56 dB. That is not a minor typo; it cuts into the credibility of every multi-step number until reconciled. Second, the headline "full grid" capability is demonstrated only at NFE=5. At NFE=25 the authors themselves show the ratio mean drifting to 0.947, KS p-value collapsing to 1.3e-14, and a -0.51 dB regression. They attribute this to train-inference joint mismatch, which is plausible and honestly stated, but it means the smart-start / target-L grid is only established at shallow depth. The theory for joint-law preservation (Prop. 3) is true for the coupling they define under an oracle predictor, so it does not rescue the trained sampler.

Also, the smart-start claim that a real input is "by construction" in-distribution is shaky for airborne SAR, and the paper's own mid-pack results on those sensors confirm the gap. The abstract's "leading results" and the Beta-ratio validation bullet are a bit stronger than the body supports. Missing seeds, error bars, and details on post-hoc normalization also make the real-SAR table hard to interpret.

That said, this is a serious piece of work. The core idea is novel and demonstrated, the theory is coherent, and the limitations are disclosed rather than hidden. I would send it to peer review with a request that the authors reconcile the table discrepancy and report the NFE=5 vs NFE=25 grid performance explicitly. For anyone working on SAR despeckling or non-Gaussian diffusion bridges, this is worth reading now.

Referee Report

4 major / 5 minor

Summary. The paper introduces γ-Bridge, a diffusion bridge for multiplicative Gamma noise with SAR despeckling as the main application. Bridge time is identified with the physical look number L(t), the forward marginals are exactly x_t | x_0 ~ x_0·Gamma(L(t), L(t)), and a Gamma–Lévy decomposition yields closed-form stochastic and deterministic reverse steps. A single network trained only at L_obs=1 on natural images is claimed to support zero-shot restoration over the full (L_in, L_out) grid through target-L stopping and smart-start input control; real-SAR deployment uses a homogeneous-patch ENL estimator. Experiments report synthetic benchmarks, multi-step analysis, runtime look-number controls, ablations, and no-reference results on six spaceborne/airborne SAR sensors.

Significance. If the central claim holds, the contribution is significant: a single synthetic-trained checkpoint with physically interpretable input/output look-number controls would be a practical and conceptually appealing advance for SAR despeckling. The elementary Gamma-marginal derivations (Prop. 1–2, Theorem 1, and the Beta-coupled variance formula in Section IV-D) are clean and correctly assembled, and the paper is unusually candid in Sections III-E and V about marginal-versus-joint correctness and train–inference mismatch. Code is released. However, the headline zero-shot grid claim is currently demonstrated only at NFE=5, the deeper-chain mismatch is acknowledged but unresolved, and there is an internal numerical inconsistency between Table IX and Tables VI/X. The significance is therefore conditional on resolving these issues; the core construction is not invalidated.

major comments (4)

[IV-C, Tables VI and IX/X] Table VI reports deterministic NFE=5 PSNR = 22.56 dB and NFE=25 = 22.05 dB on the 64-crop L_obs=1 protocol, and Table X repeats the full-model NFE=5/25 values (22.56/22.05). Table IX, described as the same checkpoint and the same 64 BSDS500 crops, gives deterministic NFE=5 PSNR = 21.21 dB and NFE=25 = 20.71 dB. The roughly 1.35 dB discrepancy at both depths is not explainable by rounding, and it affects the multi-step analysis and the ablation baseline. The protocol must be reconciled or the tables corrected before the quantitative claims are reproducible.
[III-E, Eq. (9), Prop. 3, Remark 4] The theoretical joint-law guarantee is weaker than the narrative suggests. Eq. (9) defines the forward joint q(x_{T-1},...,x_0|x_0) as the product of the reverse kernels of Eq. (6); Prop. 3 then proves that the oracle reverse chain initialised at x_obs has this law. That is true by construction and does not compare the learned sampler with an independent forward coupling or any other joint law. The text immediately after Prop. 3 and Remark 4 concede the limitation: L_cons is defined on the deterministic trajectory and does not formally entail stochastic-chain joint-law fidelity. Thus the load-bearing guarantee behind smart-start/full-grid restoration is an empirical claim, currently demonstrated only at NFE=5. Please state this explicitly and either supply a real trajectory-level test at the depths used by the headline grid experiments or narrow the claim.
[V, Table VI] The paper's own train-inference mismatch analysis shows that the iterated reverse chain deviates from the training-time joint at greater depth: at NFE=25 the ratio mean drifts to 0.947 and the KS p-value collapses to 1.3e-14, with a -0.51 dB regression versus NFE=5. All headline grid results (Tables VII–VIII, Fig. 9) use NFE=5. Consequently, the claim of zero-shot restoration over the full admissible (L_in, L_out) grid is substantiated only at shallow depth; cells requiring longer reverse chains inherit an unresolved mismatch. The authors should either provide a scheme that keeps the ratio statistics calibrated at larger NFE, or explicitly scope the zero-shot claim to the validated NFE regime.
[III-H, IV-B] Smart-start asserts that a real input y with estimated look L̂_in is 'by construction an in-distribution sample of q(x_{t*}|x_0)'. This holds only under the single-L Gamma observation model. The paper's own real-SAR results show a substantial domain gap on airborne sensors (Section IV-B: γ-Bridge is mid-pack on miniSAR/FARAD, with excess fine texture and dark-tail intensities), so the in-distribution anchoring is not established for those sensors. The claim should be qualified to 'under the Gamma model,' and the homogeneous-patch estimator should be validated (e.g., sensitivity to window size/percentile, or calibration on synthetic data) before extending the zero-shot full-grid claim to heterogeneous sensors.

minor comments (5)

[IV-B] The 'post-hoc normalization' used in real-SAR evaluation is not defined. If it rescales outputs to match the input mean, the mean-preservation claims in Tables IV–V should be interpreted accordingly; please clarify.
[III-H] The homogeneous-patch ENL estimator has several free hyperparameters (window size 32, stride 16, 90th percentile) that are not ablated. A small sensitivity study would help establish that smart-start is robust to L̂_in errors.
[IV-C and Fig. 8] The deterministic reverse uses a T=100 exponential schedule but NFE ∈ {1,5,25}. Please specify how the NFE steps are selected from the 100-step schedule (uniform subsampling, geometric, or otherwise) for reproducibility.
[Table I] Several entries in Table I appear without a visible separator (e.g., '48.1026.14' and '23.1161.95'), making the PSNR/SSIM columns ambiguous. This is likely a formatting artifact but should be fixed.
[III-E, Remark 3] The dual use of x_0 for both the deterministic clean reflectivity and the random terminal bridge state is a recurring source of confusion. A distinct symbol for the terminal random variable would improve readability.

Circularity Check

1 steps flagged · score 2.0 of 10

Most of the derivation is a valid construction, but Proposition 3 is a self-consistency tautology: Eq. (9) defines the forward joint as the product of the reverse kernels, so claiming the reverse chain has that joint law is true by construction. The headline empirical grid and zero-shot results are externally benchmarked and not circular.

self definitional [Section III-E, Eq. (9) and Proposition 3]
"Under this coupling the forward process is a Markov chain in the reverse-time index and admits the joint distribution q(x_{T−1}, . . . , x0 |x0) = q(x_{T−1} |x0) ∏ q(x_{t−1} |x_t, x0), whose kernels are given exactly by Eq. 6. Any other coupling ... we work with the Gamma–Lévy coupling because it is the one implicitly realised by the reverse sampler."

The forward joint in Eq. (9) is defined as the product of the reverse kernels of Eq. (6). Proposition 3 then proves that the reverse chain initialised at x_{T−1}=x_obs has law equal to this forward joint — an identity that holds by construction, not an independent first-principles result. Corollary 1 inherits this definitional status. The paper itself restricts the claim to the oracle x̂0=x0 and Remark 4 concedes it does not formally entail stochastic-chain fidelity for the learned predictor, so the tautology is not load-bearing for the empirical benchmark claims.

full rationale

The paper's central empirical contributions are self-contained against external data: synthetic PSNR/SSIM comparisons on Set12, McMaster, and Kodak24 against 12 independent baselines, and real-SAR zero-shot ENL/EPI/SQI/Mean results across 33 tiles from six sensors, all without sensor-specific fine-tuning. These measurements do not reduce to fitted parameters or to the paper's own definitions. The forward marginals in Prop. 1 restate the defining Eq. (3); Prop. 2 and Theorem 1 are standard Gamma additivity in closed form. The only by-construction item is Prop. 3/Cor. 1, where Eq. (9) is defined as the reverse-chain product; this is a self-consistency statement, not a falsifiable prediction about real data. The paper candidly documents the train-inference joint mismatch in Section V and Table VI (ratio mean 0.947, KS p=1.3e-14, −0.51 dB at NFE=25), which is a limitation/correctness concern rather than circularity. Self-citations ([42], [44]) are used only as baseline/protocol references and are not load-bearing. Accordingly, no significant circularity is found beyond the definitional joint-law statement, and the score is 2.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The construction is mostly self-contained: Gamma additivity is standard, and the reverse posterior follows from it by algebra. The extra load-bearing items are domain assumptions about SAR noise (single-L Gamma, flat-patch ENL), a coupling the paper defines for itself, and the undemonstrated mean-median identification. No new physical entities are introduced.

free parameters (5)

Look schedule L(t) = T=100, L_max=10^4, log-linear spacing
Hand-chosen; the ablation shows switching to a linear schedule costs 4.5 dB, so the schedule shape is load-bearing for performance, though it is not fitted to match theory.
Loss weights (lambda_rec, lambda_ratio, lambda_cons) = (10, 1, 5)
Hand-set and held fixed; no sensitivity analysis reported.
Stability clip tau = 5
Bounding the log-residual; the paper states the bound is never approached on converged checkpoints, so it is not load-bearing.
ENL estimator hyperparameters = 32x32 windows, stride 16, 90th percentile
Hand-chosen heuristic that determines L_in for real SAR smart-start; the paper notes it over-shoots on repetitive fine texture.
Effective default inference depth = NFE=5 (deterministic)
The paper's default; deeper chains degrade, so the reported gains are specific to this depth.

assumptions (6)

standard math Lemma 2: independent Gammas with common rate add to a Gamma with summed shapes
Proved in Appendix A via characteristic functions; underpins Prop. 2's Levy decomposition.
standard math Lemma 1: Gamma scaling
Used throughout to convert between rate-1 draws and L-look marginals.
domain assumption Eq. (2): real SAR intensity follows a single-L multiplicative unit-mean Gamma model
The zero-shot real-SAR claim — smart-start anchoring and the look estimator — both presuppose this. Violations (non-Gamma clutter, varying ENL) show up as the paper's mid-pack airborne results.
ad hoc to paper Gamma-Levy coupling is the operative forward joint
Eq. (3) fixes only the marginals; the paper defines the joint as the product of reverse kernels (Eq. 9), so Prop. 3's joint-law guarantee holds for a coupling the authors constructed rather than observed.
domain assumption Posterior mean approximately equals median for the multiplicative-Gamma posterior; the l1-trained predictor tracks the posterior mean to within a few percent
Stated without derivation in Section III-E ('Relation to the network predictor'); the deterministic-reverse mean-tracking argument depends on it.
domain assumption Top-decile ENL patches are speckle-dominated (the flattest regions)
Section III-H; underpins the L_in estimator used for real SAR.

how reviews work

0 comments

Cite this review

Pith. "Pith review of $\gamma$-Bridge: A Look-Parametric Diffusion Bridge." pith.science (2026). https://pith.science/paper/FMN43CX7

@misc{pith2026260722719,
  author       = {Pith},
  title        = {Pith review of: $\gamma$-Bridge: A Look-Parametric Diffusion Bridge},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FMN43CX7}},
  note         = {Machine review of arXiv:2607.22719}
}

abstract

Multiplicative Gamma noise is a signal-dependent degradation in coherent imaging; synthetic aperture radar (SAR) despeckling is its most prominent real-world instance. Existing diffusion denoisers parameterize their forward process by abstract signal-to-noise schedules rather than by the physical look number $L$, so different deployment scenarios typically require separately trained models, and transfer from synthetic Gamma training to real SAR remains challenging without clean ground truth. We introduce $\gamma$-Bridge, a look-parametric bridge whose schedule $L(t)$ connects the noisy observation at $L_{obs}$ to the clean limit through exact multiplicative Gamma marginals. Its closed-form Gamma--L\'evy reverse posterior admits both stochastic and deterministic processes, while observation conditioning and a two-step consistency loss stabilize multi-step inference in the low-SNR single-look regime. Because bridge time directly represents $L$, one conditioned network can smart-start from any admissible input look and stop at a target look number. These two orthogonal controls enable zero-shot restoration over the full admissible grid after training only at $L_{obs} = 1$ on natural images with synthetic Gamma corruption. Combined with a homogeneous-patch look estimator, $\gamma$-Bridge processes data from six spaceborne and airborne SAR sensors without sensor-specific fine-tuning, achieving leading results on standard synthetic benchmarks while providing physically interpretable input and output controls absent from prior denoisers. Codes are released \href{https://github.com/Teriri1999/GammaBridge}{here}.

Figures

Figures reproduced from arXiv: 2607.22719 by the authors.

**Figure 1.** Overview of γ-Bridge: the bridge time indexes the physical look number L(t), connecting the clean limit to the single-look observation through exact Gamma marginals. Proposition 2 (Levy decomposition) ´ . Let Gt−1 := Gt + Yt→t−1. Then Gt−1 ∼ Gamma(L(t−1), 1), and xt−1 = αt xt + x0 L(t−1) Yt→t−1, αt := L(t) L(t−1). (5) Proof. Yt→t−1 ∼ Gamma(L(t−1)−L(t), 1) and Gt ∼ Gamma(L(t), 1) share rate 1 and are independent by c… view at source ↗

**Figure 2.** Training (top): Gθ maps a random bridge state xt and the observation xobs to xˆ0 via a log-residual head, supervised by Lrec, Lratio, and Lcons. Inference (bottom): the estimated look number Lˆ triggers smart-start at the matching bridge step, and target-L stopping produces outputs at any Lout. Applying the chain rule q(xT −1, . . . , x0 | x0) = q(xT −1 | x0) Q t q(xt−1 | xt, x0) concludes. Corollary 1 (Distribution… view at source ↗

**Figure 3.** Qualitative synthetic despeckling on two natural scenes. Left: clean [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (6 more)

**Figure 4.** Figure 4: Qualitative comparison on spaceborne SAR sensors. Thirteen methods are split into two rows of seven columns each; [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

**Figure 5.** Figure 5: Qualitative comparison on airborne SAR sensors. Eleven methods are split into three rows; [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

**Figure 6.** Figure 6: Qualitative comparison on a FARAD X-band scene (parking lot with vehicles). Thirteen methods split across two rows; [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]

**Figure 7.** Figure 7: Target-L output control from a single model. From left: reference x0, observation, and outputs at Lout ∈ {2, 4, 8, 16, 66, 266, 10000}. 𝒙𝟎 𝒙𝒐𝒃𝒔 𝑺𝒕𝒆𝒑 − 𝟏 𝑺𝒕𝒆𝒑 − 𝟓 𝑺𝒕𝒆𝒑 − 𝟐𝟓 [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]

**Figure 8.** Figure 8: Multi-step reverse at Lobs=1 on three BSDS500 crops. Left to right: clean x0, noisy xobs, and outputs at NFE ∈ {1, 5, 25} (deterministic). At 25 steps, PSNR drops by 0.51 dB relative to 5 steps and the sample-mean ratio falls to 0.947, revealing a systematic ≈ 5% over-…

**Figure 9.** Figure 9: Runtime look-number control (error bars: [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

44 extracted references · 4 linked inside Pith

[1]

Multiplicative noise removal using variable splitting and constrained optimization,

J. M. Bioucas-Dias and M. A. T. Figueiredo, “Multiplicative noise removal using variable splitting and constrained optimization,”IEEE Transactions on Image Processing, vol. 19, no. 7, pp. 1720–1730, 2010

2010
[2]

Dynamical threshold-based fractional anisotropic diffusion for speckle noise removal,

J. Wei and X. Liao, “Dynamical threshold-based fractional anisotropic diffusion for speckle noise removal,”IEEE Transactions on Image Processing, vol. 34, pp. 2826–2839, 2025

2025
[3]

Fractional-order cross-diffusion system for multiplicative noise removal,

J. Gao, J. Sun, and S. Shi, “Fractional-order cross-diffusion system for multiplicative noise removal,”Computers & Mathematics with Applica- tions, vol. 164, pp. 1–11, 2024

2024
[4]

Sar despeckling using a denoising diffusion probabilistic model,

M. V . Perera, N. G. Nair, W. G. C. Bandara, and V . M. Patel, “Sar despeckling using a denoising diffusion probabilistic model,”IEEE Geoscience and Remote Sensing Letters, vol. 20, pp. 1–5, 2023

2023
[5]

Sddpm: Speckle denoising diffusion proba- bilistic models,

S. Guha and S. T. Acton, “Sddpm: Speckle denoising diffusion proba- bilistic models,”arXiv preprint arXiv:2311.10868, 2023

arXiv 2023
[6]

Self-supervised score-based despeckling for sar imagery via log-domain transformation,

J. Heo, “Self-supervised score-based despeckling for sar imagery via log-domain transformation,”arXiv preprint arXiv:2601.14334, 2026

arXiv 2026
[7]

Denoising diffusion gamma models,

E. Nachmani, R. S. Roman, and L. Wolf, “Denoising diffusion gamma models,”arXiv preprint arXiv:2110.05948, 2021

arXiv 2021
[8]

Diffusion model for generative image denoising,

Y . Xie, M. Yuan, B. Dong, and Q. Li, “Diffusion model for generative image denoising,”arXiv preprint arXiv:2302.02398, 2023

arXiv 2023

Show all 44 references

[9]

Dispeckle: Diffusion model that unwinds speckle formation with off-the-shelf gaussian denoisers,

D. Lu, C. Liu, J. Yin, and J. Yang, “Dispeckle: Diffusion model that unwinds speckle formation with off-the-shelf gaussian denoisers,”IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1–22, 2025

2025
[10]

A tunable despeckling neural network stabilized via diffusion equation,

Y . Ran, Z. Guo, J. Li, Y . Li, M. Burger, and B. Wu, “A tunable despeckling neural network stabilized via diffusion equation,”Signal Processing, vol. 239, p. 110324, 2026

2026
[11]

Sar image despeckling through convolutional neural networks,

G. Chierchia, D. Cozzolino, G. Poggi, and L. Verdoliva, “Sar image despeckling through convolutional neural networks,” in2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 2017, pp. 5438–5441

2017
[12]

Sar image despeckling using a convolutional neural network,

P. Wang, H. Zhang, and V . M. Patel, “Sar image despeckling using a convolutional neural network,”IEEE Signal Processing Letters, vol. 24, no. 12, pp. 1763–1767, 2017

2017
[13]

Speckle-driven unsupervised despeckling for sar images,

F. Bo, X. Ma, S. Hu, G. An, Y . Li, and Y . Cen, “Speckle-driven unsupervised despeckling for sar images,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 18, pp. 13 023–13 034, 2025

2025
[14]

Sar2sar: A semi-supervised despeckling algorithm for sar images,

E. Dalsasso, L. Denis, and F. Tupin, “Sar2sar: A semi-supervised despeckling algorithm for sar images,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 14, pp. 4321– 4329, 2021

2021
[15]

Speckle2void: Deep self-supervised sar despeckling with blind-spot convolutional neu- ral networks,

A. B. Molini, D. Valsesia, G. Fracastoro, and E. Magli, “Speckle2void: Deep self-supervised sar despeckling with blind-spot convolutional neu- ral networks,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–17, 2021

2021
[16]

Deep image prior,

D. Ulyanov, A. Vedaldi, and V . Lempitsky, “Deep image prior,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2018

2018
[17]

Self-supervised sar despeckling using deep image prior,

C. Albisani, D. Baracchi, A. Piva, and F. Argenti, “Self-supervised sar despeckling using deep image prior,”Pattern Recognition Letters, vol. 190, pp. 169–176, 2025

2025
[18]

I 2SB: Image-to-image schr ¨odinger bridge,

G.-H. Liu, A. Vahdat, D.-A. Huang, E. Theodorou, W. Nie, and A. Anandkumar, “I 2SB: Image-to-image schr ¨odinger bridge,” inPro- ceedings of the 40th International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, A. Krause, E. Brunskill, K. Cho, B...

2023
[19]

Da-physdiff: A dual- aware physics-informed diffusion model for sar image despeckling,

X. Cai, L. Wang, Y . Zhang, M. Wang, and Y . Li, “Da-physdiff: A dual- aware physics-informed diffusion model for sar image despeckling,” IEEE Geoscience and Remote Sensing Letters, vol. 23, pp. 1–5, 2026

2026
[20]

Star-shaped denoising diffusion probabilistic models,

A. Okhotin, D. Molchanov, A. Vladimir, G. Bartosh, V . Ohanesian, A. Alanov, and D. Vetrov, “Star-shaped denoising diffusion probabilistic models,” inAdvances in Neural Information Processing Systems, A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, Eds., v...

2023
[21]

Denoising diffusion bridge models,

L. Zhou, A. Lou, S. Khanna, and S. Ermon, “Denoising diffusion bridge models,” inInternational Conference on Learning Representations, B. Kim, Y . Yue, S. Chaudhuri, K. Fragkiadaki, M. Khan, and Y . Sun, Eds., vol. 2024, 2024, pp. 8160–8171

2024
[22]

Unpaired image-to-image translation via neural schr ¨odinger bridge,

B. Kim, G. Kwon, K. Kim, and J. C. YE, “Unpaired image-to-image translation via neural schr ¨odinger bridge,” inInternational Conference on Learning Representations, B. Kim, Y . Yue, S. Chaudhuri, K. Fragki- adaki, M. Khan, and Y . Sun, Eds., vol. 2024, 2024, pp. 19 312–19 331

2024
[23]

Diffusion bridge implicit models,

K. Zheng, G. He, J. Chen, F. Bao, and J. Zhu, “Diffusion bridge implicit models,” inInternational Conference on Learning Representations, Y . Yue, A. Garg, N. Peng, F. Sha, and R. Yu, Eds., vol. 2025, 2025, pp. 81 857–81 884

2025
[24]

Residual diffusion bridge model for image restoration,

H. Wang, J. Zhang, H. Chen, H. Guo, D. Wang, J. Ma, and B. Du, “Residual diffusion bridge model for image restoration,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, 2026, pp. 8375–8386

2026
[25]

Consistency trajectory models: Learning probability flow ode trajectory of diffusion,

D. Kim, C.-H. Lai, W.-H. Liao, N. Murata, Y . Takida, T. Uesaka, Y . He, Y . Mitsufuji, and S. Ermon, “Consistency trajectory models: Learning probability flow ode trajectory of diffusion,” 2024

2024
[26]

Generalized consistency trajectory models for image manipulation,

B. Kim, J. Kim, J. Kim, and J. C. Ye, “Generalized consistency trajectory models for image manipulation,” 2025, Conference paper, p. 101753 – 101775

2025
[27]

Digital image enhancement and noise filtering by use of local statistics,

J.-S. Lee, “Digital image enhancement and noise filtering by use of local statistics,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. PAMI-2, no. 2, pp. 165–168, 1980

1980
[28]

A model for radar images and its application to adaptive digital filtering of multiplicative noise,

V . S. Frost, J. A. Stiles, K. S. Shanmugan, and J. C. Holtzman, “A model for radar images and its application to adaptive digital filtering of multiplicative noise,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. PAMI-4, no. 2, pp. 157–166, 1982

1982
[29]

Iterative weighted maximum likelihood denoising with probabilistic patch-based weights,

C.-A. Deledalle, L. Denis, and F. Tupin, “Iterative weighted maximum likelihood denoising with probabilistic patch-based weights,”IEEE Transactions on Image Processing, vol. 18, no. 12, pp. 2661–2672, 2009

2009
[30]

A nonlocal sar image denoising algorithm based on llmmse wavelet shrinkage,

S. Parrilli, M. Poderico, C. V . Angelino, and L. Verdoliva, “A nonlocal sar image denoising algorithm based on llmmse wavelet shrinkage,” IEEE Transactions on Geoscience and Remote Sensing, vol. 50, no. 2, pp. 606–616, 2012

2012
[31]

Nl- sar: A unified nonlocal framework for resolution-preserving (pol)(in)sar denoising,

C.-A. Deledalle, L. Denis, F. Tupin, A. Reigber, and M. J ¨ager, “Nl- sar: A unified nonlocal framework for resolution-preserving (pol)(in)sar denoising,”IEEE Transactions on Geoscience and Remote Sensing, vol. 53, no. 4, pp. 2021–2038, 2015

2021
[32]

Alternating minimization algorithm for speckle reduction with a shifting technique,

H. Woo and S. Yun, “Alternating minimization algorithm for speckle reduction with a shifting technique,”IEEE Transactions on Image Processing, vol. 21, no. 4, pp. 1701–1714, 2012

2012
[33]

Mulog, or how to apply gaussian denoisers to multi-channel sar speckle reduction?

C.-A. Deledalle, L. Denis, S. Tabti, and F. Tupin, “Mulog, or how to apply gaussian denoisers to multi-channel sar speckle reduction?”IEEE Transactions on Image Processing, vol. 26, no. 9, pp. 4389–4403, 2017

2017
[34]

As if by magic: Self-supervised training of deep despeckling networks with merlin,

E. Dalsasso, L. Denis, and F. Tupin, “As if by magic: Self-supervised training of deep despeckling networks with merlin,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–13, 2021

2021
[35]

Transformer-based sar image despeckling,

M. V . Perera, W. G. C. Bandara, J. M. J. Valanarasu, and V . M. Patel, “Transformer-based sar image despeckling,” inIGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium, 2022, pp. 751–754

2022
[36]

An optimized blockwise nonlocal means denoising filter for 3-d magnetic resonance images,

P. Coupe, P. Yger, S. Prima, P. Hellier, C. Kervrann, and C. Barillot, “An optimized blockwise nonlocal means denoising filter for 3-d magnetic resonance images,”IEEE Transactions on Medical Imaging, vol. 27, no. 4, pp. 425–441, 2008

2008
[37]

Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising,

K. Zhang, W. Zuo, Y . Chen, D. Meng, and L. Zhang, “Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising,”IEEE Transactions on Image Processing, vol. 26, no. 7, pp. 3142–3155, 2017

2017
[38]

Sar image despeckling using continuous attention module,

J. Ko and S. Lee, “Sar image despeckling using continuous attention module,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 15, pp. 3–19, 2021

2021
[39]

Agsdnet: Attention and gradient-based sar denoising network,

R. K. Thakur and S. K. Maji, “Agsdnet: Attention and gradient-based sar denoising network,”IEEE Geoscience and Remote Sensing Letters, vol. 19, pp. 1–5, 2022

2022
[40]

Sifsdnet: Sharp image feature based sar denoising network,

T. R. Kumar and M. S. Kumar, “Sifsdnet: Sharp image feature based sar denoising network,” inIGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium, 2022, pp. 3428–3431

2022
[41]

Towards deep unsupervised sar despeckling with blind-spot convolutional neural networks,

A. B. Molini, D. Valsesia, G. Fracastoro, and E. Magli, “Towards deep unsupervised sar despeckling with blind-spot convolutional neural networks,” inIGARSS 2020 - 2020 IEEE International Geoscience and Remote Sensing Symposium, 2020, pp. 2507–2510

2020
[42]

Sar despeckling via regional denoising diffusion probabilistic model,

X. Hu, Z. Xu, Z. Chen, Z. Feng, M. Zhu, and L. Stankovi ´c, “Sar despeckling via regional denoising diffusion probabilistic model,” in IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium, 2024, pp. 7226–7230

2024
[43]

Contrastive learn- ing for real sar image despeckling,

Y . Fang, R. Liu, Y . Peng, J. Guan, D. Li, and X. Tian, “Contrastive learn- ing for real sar image despeckling,”ISPRS Journal of Photogrammetry and Remote Sensing, vol. 218, pp. 376–391, 2024

2024
[44]

Sar despeckling via log-yeo-johnson transformation and sparse representa- tion,

X. Hu, M. Zhu, D. Stankovi ´c, Z. Feng, S. Mao, and L. Stankovi ´c, “Sar despeckling via log-yeo-johnson transformation and sparse representa- tion,”arXiv preprint arXiv:2412.18121, 2024

2024 arXiv

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.

[1] [1]

Multiplicative noise removal using variable splitting and constrained optimization,

J. M. Bioucas-Dias and M. A. T. Figueiredo, “Multiplicative noise removal using variable splitting and constrained optimization,”IEEE Transactions on Image Processing, vol. 19, no. 7, pp. 1720–1730, 2010

2010

[2] [2]

Dynamical threshold-based fractional anisotropic diffusion for speckle noise removal,

J. Wei and X. Liao, “Dynamical threshold-based fractional anisotropic diffusion for speckle noise removal,”IEEE Transactions on Image Processing, vol. 34, pp. 2826–2839, 2025

2025

[3] [3]

Fractional-order cross-diffusion system for multiplicative noise removal,

J. Gao, J. Sun, and S. Shi, “Fractional-order cross-diffusion system for multiplicative noise removal,”Computers & Mathematics with Applica- tions, vol. 164, pp. 1–11, 2024

2024

[4] [4]

Sar despeckling using a denoising diffusion probabilistic model,

M. V . Perera, N. G. Nair, W. G. C. Bandara, and V . M. Patel, “Sar despeckling using a denoising diffusion probabilistic model,”IEEE Geoscience and Remote Sensing Letters, vol. 20, pp. 1–5, 2023

2023

[5] [5]

Sddpm: Speckle denoising diffusion proba- bilistic models,

S. Guha and S. T. Acton, “Sddpm: Speckle denoising diffusion proba- bilistic models,”arXiv preprint arXiv:2311.10868, 2023

arXiv 2023

[6] [6]

Self-supervised score-based despeckling for sar imagery via log-domain transformation,

J. Heo, “Self-supervised score-based despeckling for sar imagery via log-domain transformation,”arXiv preprint arXiv:2601.14334, 2026

arXiv 2026

[7] [7]

Denoising diffusion gamma models,

E. Nachmani, R. S. Roman, and L. Wolf, “Denoising diffusion gamma models,”arXiv preprint arXiv:2110.05948, 2021

arXiv 2021

[8] [8]

Diffusion model for generative image denoising,

Y . Xie, M. Yuan, B. Dong, and Q. Li, “Diffusion model for generative image denoising,”arXiv preprint arXiv:2302.02398, 2023

arXiv 2023