Pith. sign in

REVIEW 4 major objections 5 minor 42 references

Null-space diffusion distillation turns a slow iterative diffusion solver for lensless cameras into a single-pass network that preserves measurement consistency and perceptual quality at roughly 600x lower latency.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-03 22:05 UTC pith:Y27U6PSL

load-bearing objection Useful and fast, but the consistency claim is not backed by any measurement — worth refereeing with a requested revision. the 4 major comments →

arxiv 2511.12024 v3 pith:Y27U6PSL submitted 2025-11-15 cs.CV

Null-Space Diffusion Distillation Unlocks Speed, Fidelity and Realism in Lensless Imaging

Jose Reinaldo Cunha Santos A V Silva Neto , Hodaka Kawachi , Yasushi Yagi , Tomoya Nakamura This is my paper
classification cs.CV
keywords lensless imaginginverse problemsdiffusion modelsknowledge distillationnull-space decompositionmeasurement consistencysingle-pass reconstructionperceptual quality
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Lensless cameras capture light through a coded mask, producing measurements so multiplexed that reconstructing the scene is a severely ill-posed inverse problem. The paper argues that existing strategies face a three-way trade-off: consistency-oriented classical methods are fast but perceptually poor, supervised deep models are fast and realistic but can violate physical constraints, and diffusion-prior methods are realistic and consistent but slow because they iterate. The paper's central claim is that this trade-off can be broken by distilling the null-space component of an iterative diffusion solver into a single-pass student network, conditioned on both the raw measurement and a range-space anchor. On a 200-image lensless face test set, the student reaches near-teacher perceptual quality (perceptual-distance score 0.4140 versus the teacher's 0.4068, lower is better) at 0.0342 seconds per image, roughly 600 times faster than the teacher's 20.89 seconds. If this holds, lensless cameras could produce photorealistic reconstructions without paired lensed supervision and at practical speeds.

Core claim

The central claim is that the range–null decomposition used by a training-free iterative diffusion solver is an ideal target for distillation. The solver enforces measurement consistency on the range space and lets a pretrained diffusion prior act only on the null space; the paper's student reconstructs an image as the sum of a fixed range-space anchor (a pseudo-inverse deconvolution of the measurement) and a learned null-space residual predicted by a fine-tuned diffusion U-Net. Because the anchor is fixed, every output is measurement-consistent by construction; because the residual is trained offline against cached teacher outputs, no lensed ground-truth images are needed. The paper reports

What carries the argument

The load-bearing mechanism is the null-space residual predictor: a compact 'input reducer' network compresses a six-channel input (the lensless measurement stacked with a pseudo-inverse range-space anchor) down to three channels, and a pretrained diffusion U-Net, fine-tuned at a fixed timestep, outputs the null-space correction. Adding that correction to the anchor gives the final reconstruction. This mirrors, in one forward pass, the alternating projection of the iterative teacher: the range-space component never changes, and the diffusion prior only supplies the missing null-space detail. The teacher's outputs are precomputed once with a fixed random seed, so the student can be trained wit

Load-bearing premise

The load-bearing premise, stated in the evaluation section, is that a perceptual-distance metric computed with no guarantees of pixel-level alignment or color balance between reference and reconstruction is still a valid measure of quality; if misalignment makes that metric misleading, the paper's quantitative evidence for near-teacher quality collapses.

What would settle it

Register every reconstruction to its reference (for example with optical flow) before computing perceptual distance, or evaluate on a synthetic lensless dataset where ground-truth alignment is exactly known; if the student's perceptual-distance gap to the teacher grows substantially, or its advantage over the classical baselines disappears, then the reported near-teacher quality is an artifact of the unresolved alignment problem.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Photorealistic lensless reconstruction becomes practical in speed: about 0.03 seconds per image instead of about 21 seconds, with near-teacher perceptual quality on a 200-image face test set.
  • The range–null conditioning makes measurement consistency structural: since the output is an anchor plus a null-space residual, the reconstruction cannot ignore the physics of the mask the way supervised regressors can.
  • Ablations indicate the decomposition itself matters: distilling the range–null structure generalizes better to unseen real scenes than distilling a full reconstruction, so the improvement is not just from having a good teacher.
  • The fixed-timestep result simplifies the architecture: the student's diffusive timestep can be frozen to a single value with no observed loss difference, removing a degree of freedom from the network.
  • The method is ground-truth-free in principle, so it can be applied to new lensless cameras without building a paired lensed-lensless capture rig.

Where Pith is reading between the lines

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

  • If the recipe transfers, the same offline null-space distillation could give fast, ground-truth-free reconstructions for other ill-posed inverse problems that use range-null projection, such as deblurring, inpainting, or super-resolution; the paper does not test this.
  • The fixed-timestep ablation suggests the student may not need the full pretrained diffusion U-Net at all; training a from-scratch lightweight residual network on the same cached targets could shrink the deployable model well below the current backbone.
  • Since the teacher is stochastic, each measurement currently yields one fixed-seed target; distilling an ensemble of seeds into the student would likely lower output variance and could be done at no extra inference cost.
  • A controlled comparison that isolates the value of the range-space anchor—same architecture and targets, but dropping the anchor from the student input—would directly test how much of the gain comes from the structure-aware conditioning rather than from the teacher targets.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes Null-Space Diffusion Distillation (NSDD), a single-pass student network distilled from an iterative DDNM+ solver for lensless imaging. The student is conditioned on the lensless measurement y and on a Wiener-filter range-space anchor A†y, and is trained with MSE to cached DDNM+ teacher reconstructions, avoiding paired lensless-lensed supervision. On 200 Lensless FFHQ test images, NSDD reports LPIPS 0.4140 at 0.0342 s/image versus the teacher's 0.4068 at 20.8908 s/image, and qualitative results are shown on PhlatCam. The authors claim that NSDD preserves measurement consistency and achieves photorealistic results at a fraction of the teacher's runtime and memory.

Significance. The reported speed advantage is substantial and the teacher-student distillation setup is standard and reproducible, with code provided. If the measurement-consistency claim were substantiated and the range-null structure were actually enforced, the paper would offer a practical route to fast, ground-truth-free lensless imaging. The main risk is that the central conceptual contribution — preserving measurement consistency and null-space structure — is not enforced by the architecture or loss and is not measured, and the quantitative quality claim rests on a single metric computed on 200 images without error bars. These issues are fixable and do not undermine the core speed/fidelity trade-off.

major comments (4)
  1. [Abstract; Sec. 4.1, Eq. (14)-(15)] The claim that NSDD 'preserves measurement consistency' is currently unsupported. Equation (14) defines xhat_s = A†y + xhat_null, but xhat_null is an unconstrained network output; nothing projects it onto the null space of A. The training loss in Eq. (15) is only MSE to teacher reconstructions and contains no data-consistency or physics term. Consistency is therefore neither enforced nor measured: Table 1 reports LPIPS and runtime only, with no residual metric such as ||y - A xhat_s|| / ||y||. Please either add a null-space projection or consistency loss, and report a consistency metric on the test set.
  2. [Abstract; Sec. 4.3, Fig. 6] The abstract states that 'ablation experiments show that distilling the range–null decomposition improves reconstruction quality and robustness over unstructured full-reconstruction distillation,' but the main text contains no such experiment. The only ablation reported is the fixed diffusion timestep comparison in Fig. 6. If this ablation exists in the supplementary material, it must be cited and summarized; otherwise the claim should be removed or the experiment added. This is load-bearing because it is the main evidence for the range-null structure being beneficial.
  3. [Sec. 4.2, Table 1] The quantitative evaluation uses LPIPS over 200 test images with no error bars or statistical significance testing. The paper itself acknowledges there are 'no guarantees about pixel-level alignment between reference and reconstructions' and that color balance differs. LPIPS is computed from spatially localized deep features and is sensitive to misalignment, so the reported 0.4140 (NSDD) vs 0.4068 (DDNM+) difference may not be meaningful. Please report standard errors/confidence intervals, consider paired tests, and include a robustness check (e.g., a distribution-level metric such as FID/KID, or consistency residuals).
  4. [Sec. 5 (Conclusion)] The conclusion claims 'the first ground-truth-free route to photorealistic lensless imaging,' but the paper cites [33], 'Self-supervised neural reconstructions for lensless imaging,' which appears to be a prior ground-truth-free reconstruction method. Please either substantiate the novelty claim by clarifying the distinction from [33], or soften the claim to avoid an overstatement.
minor comments (5)
  1. [Sec. 2.1, Eq. (1)-(2)] Notation is inconsistent: Eq. (1) uses x for the latent scene, while Eq. (2) switches to v. Also, 'R 1×Ny' and 'R 1×Nx' should be R^{Ny} and R^{Nx} or similar.
  2. [Table 1] The table header lists 'DDNM' but the paper refers to DDNM+ elsewhere; the asterisk next to NSDD is unexplained. Please align labels and add a footnote.
  3. [Abstract] The abstract claims memory reduction, but no memory measurements are reported in Table 1 or elsewhere. Either report memory usage or remove the claim from the abstract.
  4. [Sec. 4.1] Typo: 'although DDPM is stochastic, fixed seeds allow us to train...' should be 'DDPM is stochastic, but fixed seeds...'.
  5. [References] Reference [13] has a typo ('synthetis' should be 'synthesis'). References [40] and [41] appear to duplicate the same paper title with overlapping authors; please merge or distinguish.

Circularity Check

2 steps flagged

NSDD's near-teacher quality is the training objective itself; measurement-consistency preservation is asserted but not enforced or measured.

specific steps
  1. self definitional [Sec. 4.1, Eq. (14)-(15); Abstract]
    "The final reconstruction composes a range anchor and a learned null-space residual: xhat_s = A†y + xhat_null. (14) ... We choose to use the Mean Squared Error (MSE) objective between the student reconstruction for the i-th measurement xhat_s and its respective cached teacher estimate T(y_i): L_MSE = 1/N sum ||xhat_s - T(y_i)||_2^2. (15)"

    The student output is defined by minimizing MSE against cached DDNM+ teacher reconstructions (Eq. 15). The headline result 'near-teacher perceptual quality (second-best LPIPS, below DDNM+)' is therefore the value of the training objective on the test split, not an independent prediction. Because the residual in Eq. 14 is not projected onto the null space, the student's fidelity and consistency behavior are inherited only through mimicking T(y), which is exactly the fitted target.

  2. other [Sec. 4.1 (Eq. 14-15) and Sec. 4.2 Evaluation / Table 1]
    "No lensed references or physics residual terms are required; training stability is achieved by the fixed-seed teacher and the range anchor in (14). ... Table 1. Average inference time and LPIPS score over 200 reconstructions on the Lensless FFHQ test set."

    The abstract claims NSDD 'preserves measurement consistency', but Eq. 14 adds an unprojected residual to A†y and Eq. 15 contains no data-consistency or physics term. Table 1 reports only LPIPS and wall-clock time, not a measurement residual. Thus the consistency claim is not derived from the construction; it is asserted as inherited from the teacher, i.e., from the fitted target itself, making it unfalsifiable within the paper's reported evidence.

full rationale

The paper does not rely on a self-citation chain or an imported uniqueness theorem: the diffusion priors, DDNM+, and DPS are external citations, and the authors' own prior work is used mainly for the dataset and mask, which is normal. The genuine independent content is the speedup (0.0342 s vs 20.9 s/image) and the ablation showing that range-null distillation beats unstructured full-reconstruction distillation. However, the central claim of near-teacher perceptual quality is by construction the MSE fitting objective to cached teacher outputs, and the headline 'preserves measurement consistency' is asserted without a consistency metric or an enforcement term in the loss. This is partial circularity plus an unsupported claim, but not a fully forced derivation or a self-citation tautology, so a score of 4 is appropriate.

Axiom & Free-Parameter Ledger

2 free parameters · 5 axioms · 0 invented entities

The central claim rests on a linear, known forward model; a pretrained diffusion prior; the Wiener approximation of the pseudo-inverse; LPIPS as a valid metric under misalignment; and the fixed-seed teacher as a sufficient distillation target. The only hand-set number that directly tunes the teacher is σy=0.6, plus the fixed student timestep t=999.

free parameters (2)
  • DDNM+ teacher noise parameter σy = 0.6
    Chosen from parameter sweep in Sec. 3 as a balance between smoothing and noise; it sets the teacher targets for distillation and therefore the student's quality ceiling. No separate validation split reported.
  • Student fixed diffusion timestep t = 999
    The pretrained UNet's timestep is fixed during training and inference; ablation shows t=999 and t=59 are nearly indistinguishable, so the choice is not performance-critical but still a hand-set design constant.
axioms (5)
  • domain assumption Lensless forward model is linear and approximately shift-invariant: y = Ax + n (Eq. 1-2), with known system matrix A/PSF h.
    All reconstruction methods, teacher, and student rely on this physical model; errors in A transfer directly to range-space anchors.
  • domain assumption A pretrained unconditional diffusion model provides a valid prior for natural images (FFHQ/ImageNet).
    DDNM+ teacher and NSDD's student backbone inherit the prior quality; no retraining on lensless data.
  • domain assumption Wiener filter is an adequate approximation of the pseudo-inverse A† for the range-space anchor.
    Used in both teacher and student (Sec. 3 and 4.2, following PhoCoLens); if the approximation is poor, range-null updates are wrong.
  • domain assumption LPIPS computed against reference images remains meaningful despite acknowledged lack of pixel-level alignment and color balance differences.
    Explicitly stated in Sec. 4.2 Evaluation; LPIPS is not invariant to spatial misalignment, so this assumption is load-bearing for quantitative quality claims.
  • domain assumption A fixed initial noise xT for the teacher (fixed seed) makes the offline teacher targets reproducible and suitable for training a student without noise conditioning.
    Sec. 4.1; if teacher outputs depend strongly on the random seed, a single seed may not represent the teacher distribution, and the student inherits one sample's stochasticity.

pith-pipeline@v1.3.0-alltime-deepseek · 9890 in / 11711 out tokens · 104303 ms · 2026-08-03T22:05:45.102572+00:00 · methodology

0 comments
read the original abstract

Lensless imaging reconstructs scenes from highly multiplexed measurements, resulting in a severely ill-posed inverse problem. In this work, we identify a fundamental trade-off between measurement consistency, perceptual quality, and inference speed across lensless reconstruction paradigms. Traditional methods favor consistency but produce perceptually degraded results, supervised approaches achieve high-quality reconstructions with fast inference but may violate physical constraints, and diffusion-prior methods achieve high perceptual quality and consistency--particularly when structured constraints such as range-null decomposition are used--but remain slow due to iterative sampling. Motivated by this observation, we propose Null-Space Diffusion Distillation (NSDD), a single-pass reconstruction model that distills structured diffusion-prior inference into an efficient feed-forward network. NSDD learns to produce high-quality reconstructions that preserve measurement consistency while avoiding costly iterative sampling. Experimental results demonstrate that NSDD achieves perceptual quality and consistency competitive with diffusion-prior methods, while providing significantly faster inference and offering a favorable balance across all three objectives. Furthermore, ablation experiments show that distilling the range--null decomposition improves reconstruction quality and robustness over unstructured full-reconstruction distillation, including on unseen real scenes. These results highlight the potential of structure-aware distillation for efficient lensless imaging. Code is available at github.com/JRCSAVSN/NullSpaceDiffusionDistillation.

Figures

Figures reproduced from arXiv: 2511.12024 by Hodaka Kawachi, Jose Reinaldo Cunha Santos A V Silva Neto, Tomoya Nakamura, Yasushi Yagi.

Figure 1
Figure 1. Figure 1: Left: Displaying lensless captures, the reference (i.e., Ref.) image displayed to be captured, and the reconstructions using a [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Depiction of the two types of diffusion guidance ap [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Parameter sweep for DPS and DDNM+ algorithms for samples of the lensless FFHQ dataset. Here y ∈ R H×W×3 and A†y ∈ R H×W×3 are stacked along the channel dimension. The extracted range-space in￾formation (i.e., A † y) shrinks the large feasible set of pos￾sible reconstructions induced by the ill-posed, noisy and highly multiplexed lensless captures. Student predictor. We reuse a pretrained diffusion UNet gθ … view at source ↗
Figure 4
Figure 4. Figure 4: NSDD overview. Top: Precomputation of the DDNM [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative reconstructions for Lensless FFHQ and PhlatCam datasets. Left to right: measurement, Wiener, ADMM, DPS, [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Training and testing losses across the 100 training epochs for lensless FFHQ and PhlatCam dataset, for models with fixed [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

42 extracted references · 2 linked inside Pith

  1. [1]

    Anderson

    Brian D.O. Anderson. Reverse-time diffusion equation mod- els.Stochastic Processes and their Applications, 12(3):313– 326, 1982. 2

    1982

  2. [2]

    Barthelmy, Louis M

    Scott D. Barthelmy, Louis M. Barbier, Jay R. Cummings, Ed E. Fenimore, Neil Gehrels, Derek Hullinger, Hans A. Krimm, Craig B. Markwardt, David M. Palmer, Ann Par- sons, Goro Sato, Masaya Suzuki, Tadayuki Takahashi, Makota Tashiro, and Jack Tueller. The burst alert telescope (bat) on the swift midex mission.Space Sci Rev, 120:143– 164, 2005. 1

    2005

  3. [3]

    One-step offline distillation of diffusion- based models via koopman modeling, 2025

    Nimrod Berman, Ilan Naiman, Moshe Eliasof, Hedi Zisling, and Omri Azencot. One-step offline distillation of diffusion- based models via koopman modeling, 2025. 3

    2025

  4. [4]

    Coskun, and Aydo- gan Ozcan

    Waheb Bishara, Ting-Wei Su, Ahmet F. Coskun, and Aydo- gan Ozcan. Lensfree on-chip microscopy over a wide field- of-view using pixel super-resolution.Opt. Express, 18(11): 11181–11191, 2010. 1

    2010

  5. [5]

    Robinson, Laura Waller, and Ashok Veeraraghavan

    Vivek Boominathan, Jacob T. Robinson, Laura Waller, and Ashok Veeraraghavan. Recent advances in lensless imaging. Optica, 9(1):1–16, 2022. 1

    2022

  6. [6]

    Distributed optimization and statistical learning via the alternating direction method of multipliers

    Stephen Boyd, Neal Parikh, Eric Chu, Borja Peleato, and Jonathan Eckstein. Distributed optimization and statistical learning via the alternating direction method of multipliers. Found. Trends Mach. Learn., 3(1):1–122, 2011. 4

    2011

  7. [7]

    Phocolens: Photorealistic and consis- tent reconstruction in lensless imaging.NeurIPS, 37:12219– 12242, 2024

    Xin Cai, Zhiyuan You, Hailong Zhang, Jinwei Gu, Wentao Liu, and Tianfan Xue. Phocolens: Photorealistic and consis- tent reconstruction in lensless imaging.NeurIPS, 37:12219– 12242, 2024. 1, 4

    2024

  8. [8]

    Caroli, J

    E. Caroli, J. B. Stephen, G. Di Cocco, L. Natalucci, and A. Spizzichino. Coded aperture imaging in x- and gamma-ray astronomy.Space Sci Rev, 45:349–403, 1987. 1

    1987

  9. [9]

    Improving diffusion models for inverse prob- lems using manifold constraints

    Hyungjin Chung, Byeongsu Sim, Dohoon Ryu, and Jong Chul Ye. Improving diffusion models for inverse prob- lems using manifold constraints. InNeurIPS, 2022. 1, 3

    2022

  10. [10]

    Parallel diffusion models of operator and image for blind inverse problems.CVPR, 2023

    Hyungjin Chung, Jeongsol Kim, Sehui Kim, and Jong Chul Ye. Parallel diffusion models of operator and image for blind inverse problems.CVPR, 2023

    2023

  11. [11]

    Diffusion pos- terior sampling for general noisy inverse problems

    Hyungjin Chung, Jeongsol Kim, Michael Thompson Mc- cann, Marc Louis Klasky, and Jong Chul Ye. Diffusion pos- terior sampling for general noisy inverse problems. InICLR,

  12. [12]

    Direct diffusion bridge using data consistency for inverse problems

    Hyungjin Chung, Jeongsol Kim, and Jong Chul Ye. Direct diffusion bridge using data consistency for inverse problems. NeurIPS, 2023. 1, 3

    2023

  13. [13]

    Diffusion models beat gans on image synthetis

    Prafulla Dhariwal and Alex Nichol. Diffusion models beat gans on image synthetis. InNeurIPS, 2021. 5, 6

    2021

  14. [14]

    J. Dong, S. Roth, and B. Schiele. Deep wiener deconvolu- tion: wiener meets deep learning for image deblurring. In NeurIPS, pages 1048–1059, 2020. 4

    2020

  15. [15]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. InNeurIPS, pages 6840–6851. Curran Associates, Inc., 2020. 2

    2020

  16. [16]

    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. IEEE TPAMI, 43(12):4217–4228, 2021. 3

    2021

  17. [17]

    Flatnet: Towards photorealistic scene reconstruction from lensless measure- ments.IEEE TPAMI, 44(4):1934–1948, 2022

    Salman Siddique Khan, Varun Sundar, Vivek Boominathan, Ashok Veeraraghavan, and Kaushik Mitra. Flatnet: Towards photorealistic scene reconstruction from lensless measure- ments.IEEE TPAMI, 44(4):1934–1948, 2022. 1

    1934

  18. [18]

    Flowdps : Flow-driven posterior sampling for inverse prob- lems

    Jeongsol Kim, Bryan Sangwoo Kim, and Jong Chul Ye. Flowdps : Flow-driven posterior sampling for inverse prob- lems. InICCV, pages 12328–12337, 2025. 3

    2025

  19. [19]

    Unrolled primal-dual networks for lensless cameras

    Oliver Kingshott, Nick Antipa, Emrah Bostan, and Kaan Aks ¸it. Unrolled primal-dual networks for lensless cameras. Opt. Express, 30(26):46324–46335, 2022. 1

    2022

  20. [20]

    On-chip fluorescence microscopy with a random microlens diffuser.Opt

    Grace Kuo, Fanglin Linda Liu, Irene Grossrubatscher, Ren Ng, and Laura Waller. On-chip fluorescence microscopy with a random microlens diffuser.Opt. Express, 28(6):8384– 8399, 2020. 1

    2020

  21. [21]

    Video diffusion posterior sampling for seeing beyond dynamic scattering layers.IEEE TPAMI, 47(12):11152–11167, 2025

    Taesung Kwon, Gookho Song, Yoosun Kim, Jeongsol Kim, Jong Chul Ye, and Mooseok Jang. Video diffusion posterior sampling for seeing beyond dynamic scattering layers.IEEE TPAMI, 47(12):11152–11167, 2025. 3

    2025

  22. [22]

    Lens- less camera: Unraveling the breakthroughs and prospects

    Shuowen Li, Yunhui Gao, Jiachen Wu, Mingjie Wang, Zhangcheng Huang, Shumei Chen, and Liangcai Cao. Lens- less camera: Unraveling the breakthroughs and prospects. Fundamental Research, 2024. 1

    2024

  23. [23]

    Latent consistency models: Synthesizing high- resolution images with few-step inference, 2023

    Simian Luo, Yiqin Tan, Longbo Huang, Jian Li, and Hang Zhao. Latent consistency models: Synthesizing high- resolution images with few-step inference, 2023. 3

    2023

  24. [24]

    Lcm-lora: A universal stable-diffusion acceleration module.arXiv preprint arXiv:2311.05556, 2023

    Simian Luo, Yiqin Tan, Suraj Patil, Daniel Gu, Patrick von Platen, Apolin´ario Passos, Longbo Huang, Jian Li, and Hang Zhao. Lcm-lora: A universal stable-diffusion acceleration module.arXiv preprint arXiv:2311.05556, 2023. 3

    Pith/arXiv arXiv 2023

  25. [25]

    On distillation of guided diffusion models

    Chenlin Meng, Robin Rombach, Ruiqi Gao, Diederik Kingma, Stefano Ermon, Jonathan Ho, and Tim Saliman. On distillation of guided diffusion models. InCVPR, 2023. 3

    2023

  26. [26]

    Learned recon- structions for practical mask-based lensless imaging.Opt

    Kristina Monakhova, Joshua Yurtsever, Grace Kuo, Nick Antipa, Kyrollos Yanny, and Laura Waller. Learned recon- structions for practical mask-based lensless imaging.Opt. Express, 27(20):28075–28090, 2019. 1, 6

    2019

  27. [27]

    Improving dif- fusion models for inverse problems using optimal posterior covariance.arXiv preprint arXiv:2402.02149, 2024

    Xinyu Peng, Ziyang Zheng, Wenrui Dai, Nuoqian Xiao, Chenglin Li, Junni Zou, and Hongkai Xiong. Improving dif- fusion models for inverse problems using optimal posterior covariance.arXiv preprint arXiv:2402.02149, 2024. 1

    Pith/arXiv arXiv 2024

  28. [28]

    Solving linear inverse problems provably via posterior sampling with latent diffusion models

    Litu Rout, Negin Raoof, Giannis Daras, Constantine Cara- manis, Alex Dimakis, and Sanjay Shakkottai. Solving linear inverse problems provably via posterior sampling with latent diffusion models. InNeurIPS, 2023. 3

    2023

  29. [29]

    Rudin, Stanley Osher, and Emad Fatemi

    Leonid I. Rudin, Stanley Osher, and Emad Fatemi. Nonlinear total variation based noise removal algorithms.Physica D: Nonlinear Phenomena, 60(1):259–268, 1992. 2

    1992

  30. [30]

    Progressive distillation for fast sampling of diffusion models

    Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models. InICLR, 2022. 3

    2022

  31. [31]

    Adversarial diffusion distillation

    Axel Sauer, Dominik Lorenz, Andreas Blattmann, and Robin Rombach. Adversarial diffusion distillation. InECCV, 2024. 3

    2024

  32. [32]

    Jos ´e Reinaldo Cunha Santos A. V . Silva Neto, Tomoya Naka- mura, Yasushi Makihara, and Yasushi Yagi. Extended depth- of-field lensless imaging using an optimized radial mask. IEEE TCI, 9:857–868, 2023. 3

    2023

  33. [33]

    Jos ´e Reinaldo Cunha Santos A. V . Silva Neto, Hodaka Kawachi, Yasushi Yagi, and Tomoya Nakamura. Self- supervised neural reconstructions for lensless imaging.Op- tical Review, 32:664–668, 2025. 2, 3

    2025

  34. [34]

    Pseudoinverse-guided diffusion models for inverse problems

    Jiaming Song, Arash Vahdat, Morteza Mardani, and Jan Kautz. Pseudoinverse-guided diffusion models for inverse problems. InICLR, pages 1–30, 2023. 1, 3

    2023

  35. [35]

    Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole

    Yang Song, Jascha Sohl-Dickstein, Diederik P. Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions. InICLR, 2021. 2

    2021

  36. [36]

    Consistency models

    Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models. InInt. Conf. Mach. Learn.,

  37. [37]

    Ubertini, F

    P. Ubertini, F. Lebrun, G. Di Cocco, A. Bazzano, A. J. Bird, K. Broenstad, A. Goldwurm, G. La Rosa, C. Labanti, P. Lau- rent, I. F. Mirabel, E. M. Quadrini, B. Ramsey, V . Reglero, L. Sabau, B. Sacco, R. Staubert, L. Vigroux, M. C. Weisskopf, and A. A. Zdziarski. Ibis: The imager on-board integral. Astronomy and Astrophysics, 411:131–139, 2003. 1

    2003

  38. [38]

    Zero-shot im- age restoration using denoising diffusion null-space model

    Yinhuai Wang, Jiwen Yu, and Jian Zhang. Zero-shot im- age restoration using denoising diffusion null-space model. ICLR, 2023. 1, 2, 3

    2023

  39. [39]

    Lensless digital holo- graphic microscopy and its applications in biomedicine and environmental monitoring.Methods, 136:4–16, 2018

    Yichen Wu and Aydogan Ozcan. Lensless digital holo- graphic microscopy and its applications in biomedicine and environmental monitoring.Methods, 136:4–16, 2018. 1

    2018

  40. [40]

    One-step dif- fusion with distribution matching distillation

    Tianwei Yin, Michael Gharbi, Richard Zhang, Eli Shecht- man, Fredo Durand, and William Freeman. One-step dif- fusion with distribution matching distillation. InNeurIPS,

  41. [41]

    One-step diffusion with distribution matching distillation

    Tianwei Yin, Michael Gharbi, Richard Zhang, Eli Shecht- man, Fredo Durand, William Freeman, and Taesung Park. One-step diffusion with distribution matching distillation. In CVPR, 2024. 3

    2024

  42. [42]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InCVPR, pages 586–595,