Pith. sign in

REVIEW 1 major objections 6 minor 70 references

Synthetic lens blur from wave optics beats real-captured training data

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 · glm-5.2

2026-07-08 22:47 UTC pith:TKRFELLB

load-bearing objection Cross-device generalization claim rests on no-reference metrics while full-reference metrics show the opposite trend; the reliability of no-reference metrics for defocus deblurring evaluation is not independently validated. the 1 major comments →

arxiv 2607.05837 v1 pith:TKRFELLB submitted 2026-07-07 cs.CV

Realistic Compound-Lens Defocus Blur Synthesis

Yunkyu Lee , Woohyeok Kim , Sunghyun Cho This is my paper
classification cs.CV
keywords defocus deblurringpoint spread functionwave opticscompound lensDebye formulationchirp Z-transformsynthetic datasetimage restoration
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.

This paper argues that physically accurate, lens-diverse defocus blur can be synthesized at scale by computing wave-optics point spread functions (PSFs) for real compound lens designs using the Debye formulation accelerated by the chirp Z-transform (CZT), then rendering depth-aware blur in radiometrically linear space with camera ISP simulation. The authors show that this pipeline is both physically grounded and computationally efficient — the Debye CZT computes a PSF in 0.018 seconds where a Rayleigh-Sommerfeld integral takes 45 seconds for comparable accuracy. Using 700 filtered photographic lens designs, they generate CLDefocus, a dataset of 42,000 blurred-sharp image pairs. The central experimental claim is that deblurring models trained on CLDefocus generalize better across unseen real cameras and lenses than models trained on either a real-captured dataset (DPDD, limited to one camera) or a simplified synthetic dataset (SYNDOF, using Gaussian blur kernels). The authors also demonstrate that imperfections in real-captured ground-truth images — geometric misalignment, photometric drift, residual blur — bias pixel-wise evaluation metrics toward blur-preserving predictions, which helps explain why CLDefocus-trained models score lower on PSNR against flawed ground truth while achieving higher perceptual quality.

Core claim

The Debye CZT provides an explicit sampling criterion (N > 4NA²√(n²_t − NA²)|z|/λ) that determines the minimum grid resolution for aliasing-free PSF computation, eliminating the empirical sampling tuning required by Huygens-principle methods. This makes it feasible to compute physically accurate, lens-specific PSFs across hundreds of compound lens designs and depth configurations, producing a synthetic dataset whose optical diversity exceeds what real capture can achieve. When used to train deblurring networks, this dataset yields measurably better cross-device generalization on no-reference perceptual metrics across four benchmark datasets and smartphone images, while also improving depth-估

What carries the argument

Debye CZT

Load-bearing premise

The Debye formulation with scalar diffraction accurately models the blur produced by real photographic compound lenses. This is least reliable for severe off-axis fields, where the Debye approximation and imperfect ray-clipping masks can produce inaccurate PSFs.

What would settle it

If the computed PSFs deviate significantly from real lens blur — particularly for off-axis fields or lenses with complex pupil shapes — the synthetic dataset would not improve cross-device generalization, and the performance gains over simpler blur models would disappear.

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

Share X Bluesky LinkedIn Reddit HN

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

1 major / 6 minor

Summary. This paper proposes a pipeline for synthesizing realistic defocus blur datasets for compound lenses. The pipeline integrates three components: (1) efficient wave-optics PSF computation via the Debye CZT formulation with explicit sampling criteria, (2) depth-aware defocus rendering with occlusion handling via layered compositing, and (3) blur synthesis in radiometrically linear space with camera ISP simulation. Using this pipeline, the authors generate CLDefocus, a dataset of 40,000 training pairs spanning 700 lens designs. Experiments compare models trained on CLDefocus against those trained on DPDD (real-captured) and SYNDOF (synthetic with simplified blur) across multiple real benchmarks (RTF, RealDOF, DPDD) and four deblurring architectures. The central claim is that CLDefocus training improves cross-device generalization. The paper also analyzes imperfections in real-captured ground truth that bias full-reference metrics.

Significance. The paper makes a solid contribution to defocus deblurring dataset synthesis. Strengths include: (1) a parameter-free sampling criterion (Eq. 2) for stable PSF computation, addressing a practical bottleneck in prior wave-optics approaches; (2) a 2500x speedup over Rayleigh-Sommerfeld (Sec. 5.3, Fig. 5) with explicit aliasing control; (3) validation across four deblurring architectures (NRKNet, Restormer, INIKNet, NAFNet) in Table S4; (4) reproducible code and dataset publicly available; (5) the smartphone evaluation (Sec. S6) and downstream task results (Sec. S7) provide additional evidence beyond standard benchmarks. The analysis of real-GT imperfections (Sec. 5.2) is a useful contribution to the evaluation methodology discussion. The lens diversity analysis (Fig. S6) provides evidence that the 700-lens collection spans a meaningful range of optical properties.

major comments (1)
  1. Sec. 5.1, Tables 1-2: The central claim that CLDefocus improves cross-device generalization is supported primarily by no-reference metrics (NIQE, MUSIQ, TOPIQ), while full-reference metrics (PSNR, SSIM, LPIPS) on real benchmarks (RTF, RealDOF, DPDD) generally favor the DPDD-trained model. The paper argues in Sec. 5.2 that real GT imperfections bias pixel-wise metrics toward blur-preserving outputs. This argument is plausible and supported by qualitative examples (Fig. 4), but the converse — that no-reference metrics reliably measure deblurring fidelity — is not independently established. NIQE, MUSIQ, and TOPIQ measure general image quality, not deblurring accuracy; a model that oversharpened or introduced high-frequency artifacts could score higher without producing more faithful deblurring. The downstream task results (Tables S2-S3) provide only marginal corroboration (RMSE 0.246 vs 0.0
minor comments (6)
  1. Sec. 4.2: The depth estimation relies on Depth Pro, a monocular estimator. The impact of depth estimation errors on synthesis quality is acknowledged in Sec. 6 but not quantified. A brief sensitivity analysis or discussion of failure modes would strengthen the paper.
  2. Table 2: On the DPDD test set, the DPDD-trained model wins on MUSIQ and TOPIQ, showing that no-reference metrics do not uniformly favor CLDefocus. This is actually informative for the reader and could be discussed more explicitly to characterize when each training set is advantageous.
  3. Sec. S4.2: The noise coefficients beta_1_ref = beta_2_ref = 1e-5 are described as much smaller than prior work. A brief justification or reference for this choice would help reproducibility.
  4. Fig. 5: The runtime comparison between Debye CZT and Rayleigh-Sommerfeld is informative but the N values for R-S are only inline in the text. A small table would improve readability.
  5. Sec. 5.1: The training protocol matches total iterations across datasets (350,000 for NRKNet), but datasets differ in size (40,000 vs 350 pairs). The interaction between dataset size and training duration deserves brief discussion.
  6. The term 'photorealistic' is used in the abstract and throughout, but photorealism is not directly validated via human evaluation. Consider softening to 'physically grounded' or similar.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful and constructive review. The referee's assessment is accurate, and the recommendation of minor revision is appropriate. We address the single major comment below.

read point-by-point responses

  1. Referee: Sec. 5.1, Tables 1-2: The central claim that CLDefocus improves cross-device generalization is supported primarily by no-reference metrics (NIQE, MUSIQ, TOPIQ), while full-reference metrics (PSNR, SSIM, LPIPS) on real benchmarks (RTF, RealDOF, DPDD) generally favor the DPDD-trained model. The paper argues in Sec. 5.2 that real GT imperfections bias pixel-wise metrics toward blur-preserving outputs. This argument is plausible and supported by qualitative examples (Fig. 4), but the converse — that no-reference metrics reliably measure deblurring fidelity — is not independently established. NIQE, MUSIQ, and TOPIQ measure general image quality, not deblurring accuracy; a model that oversharpened or introduced high-frequency artifacts could score higher without producing more faithful deblurring. The downstream task results (Tables S2-S3) provide only marginal corroboration (RMSE 0.246 vs 0.2

    Authors: The referee raises a valid and important concern. We agree that no-reference metrics alone do not conclusively establish deblurring fidelity, and the current manuscript does not adequately address the risk that these metrics could reward oversharpening or artifact introduction rather than faithful deblurring. We will revise the manuscript to strengthen the argument on multiple fronts. First, we will add explicit discussion acknowledging the limitation of no-reference metrics for evaluating deblurring specifically, including the oversharpening concern the referee identifies. Second, we will expand the downstream task results discussion to emphasize that these provide task-level evidence independent of image quality metrics: the depth estimation RMSE improvement (0.246 vs. 0.248, Table S2) and segmentation IoU improvement (0.871 vs. 0.867, Table S3) show that CLDefocus-trained models produce restorations that are more useful for downstream vision, which would not hold if the model were merely introducing artifacts. We acknowledge these margins are modest, and we will state this honestly. Third, we will add a brief discussion pointing to the qualitative results (Figs. 3, S12–S14) and the smartphone evaluation (Sec. S6) as additional evidence that the improvements reflect genuine deblurring rather than artifact introduction, since the smartphone PSFs differ substantially from the training distribution and artifact-only gains would not be expected to generalize. We do not claim to fully resolve the fundamental difficulty of evaluating deblurring without reliable ground truth — this is an open problem — but we will present the evidence more completely and temper the central claim accordingly. revision: partial

Circularity Check

0 steps flagged

No significant circularity found

full rationale

The paper's derivation chain is built from externally established optical and image-processing principles. The Debye CZT formulation (Eq. 1, Eq. S4–S9) is cited to Leutenegger et al. [30] and Hu et al. [22], both external. The off-axis extension cites Cai et al. [9], external. The ISP model follows RSBlur [49], external. The layered compositing follows Hasinoff & Kutulakos [19] and Kraus & Strengert [25], external. The Zernike wavefront fitting uses standard basis functions [41]. No load-bearing step reduces to a self-citation by the present authors. The dataset is generated from an external lens design collection [48] and external source images (DPDD RAW [2]). Evaluation is conducted on external benchmarks (RTF [14], RealDOF [29], DPDD [2]) using standard metrics. The CLDefocus test set is generated by the same pipeline used for training data, so the strong in-domain results (Table 1, top row) are expected by construction, but the paper does not present these as the central cross-device generalization claim — the central claim rests on the external benchmarks. The skeptic's concern about no-reference vs. full-reference metric reliability is an evaluation-validity concern, not a circularity concern: the paper does not define its evaluation metrics in terms of its own outputs, nor does it fit parameters to the test data and then predict them. The downstream task evaluations (Tables S2, S3) use external models (Depth Anything V2 [64], SAM 2 [47]) on the external RealDOF dataset. No step in the derivation chain reduces to its inputs by construction. Score: 0.

Axiom & Free-Parameter Ledger

6 free parameters · 4 axioms · 0 invented entities

The paper does not invent new physical entities or particles. It uses existing optical models and computational tools. The free parameters are practical implementation choices for the synthesis pipeline, not fundamental constants fitted to data.

free parameters (6)
  • N (sampling number) = 2 * N_inf
    Set to twice the theoretical lower bound to suppress aliasing (Sec. 3.1).
  • u (upsampling factor) = 5
    Chosen to remove observable aliasing in all tested configurations (Sec. 3.2).
  • CoC_limit = 1/4 * k * delta_x
    Restricts CoC to lie within half the kernel viewport (Sec. S3.2).
  • N_limit = 1536
    Restricts maximum sampling number for computational feasibility (Sec. S3.2).
  • beta_1_ref, beta_2_ref (noise coefficients) = 1e-5
    Set smaller than prior work to avoid decreasing deblurring performance (Sec. S4.2).
  • alpha (saturation power) = U(0, 4)
    Sampled from a uniform distribution for saturation synthesis (Sec. S4.2).
axioms (4)
  • domain assumption Scalar diffraction approximation is valid for photographic lenses.
    The paper treats wave fields as scalar-valued, justified by low NA and negligible polarization effects (Sec. 3.1).
  • domain assumption Debye formulation accurately models diffraction for the lens designs considered.
    The pipeline relies on the Debye formulation for PSF computation, which may be less accurate for severe off-axis fields (Sec. 6).
  • domain assumption Monocular depth estimation provides sufficiently accurate depth maps for synthesis.
    The pipeline uses Depth Pro to obtain depth maps for rendering; imprecise depth maps could produce implausible results (Sec. 6).
  • standard math Layered compositing with occlusion handling accurately models depth-of-field imaging.
    The paper adopts layered compositing formulation from prior work [19, 25] for blur synthesis (Sec. 4.3).

pith-pipeline@v1.1.0-glm · 34589 in / 2224 out tokens · 521815 ms · 2026-07-08T22:47:17.373392+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Realistic Compound-Lens Defocus Blur Synthesis." pith.science (2026). https://pith.science/paper/TKRFELLB

@misc{pith2026260705837,
  author       = {Pith},
  title        = {Pith review of: Realistic Compound-Lens Defocus Blur Synthesis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TKRFELLB}},
  note         = {Machine review of arXiv:2607.05837}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Defocus blur degrades fine image structures and limits visual perception, which can adversely affect downstream vision tasks. Although recent deep learning deblurring methods have achieved strong performance, their effectiveness depends on training data and often degrades across cameras and lenses due to limited optical diversity and realism in existing datasets. In this paper, we propose a pipeline for synthesizing realistic defocus deblurring datasets for diverse compound lenses. It integrates efficient wave-optics PSF computation via Debye CZT propagation, depth-aware defocus rendering with occlusion handling, and blur synthesis in the radiometrically linear space with camera ISP simulation. This unified pipeline enables the scalable generation of photorealistic defocus datasets with diverse lens characteristics. Using our pipeline, we generate CLDefocus, a large-scale synthetic dataset containing lens-diverse defocus image pairs. We further analyze the limitations of real-captured defocus datasets and show that such imperfections can bias full-reference evaluation. Extensive experiments demonstrate that models trained on CLDefocus achieve improved cross-device generalization compared to models trained on existing real and synthetic datasets.

Figures

Figures reproduced from arXiv: 2607.05837 by Sunghyun Cho, Woohyeok Kim, Yunkyu Lee.

Figure 1
Figure 1. Figure 1: Overview of our PSF computation process. The left shows the physical setup of the Debye formulation. A point source P1 emits light, which passes through the lens and forms a wavefront W. S is the exit pupil sphere that is centered at a focal point P2 and passes through the exit pupil center. We model W as a converging spherical wave on S with deviations (aberrations). The Debye formulation integrates all w… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of our data synthesis pipeline. From a sharp RAW image, we first sample a RAW patch and convert it to linear RGB via a partial ISP. Using the cor￾responding depth map, we sample a lens and a focusing distance that satisfy our constraints on blur size and practical PSF computation. We then quantize the depths while minimizing defocus discontinuity and build a multi-layered representation. Next, we … view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparisons on the RealDOF dataset. sity of our dataset. On real benchmarks, the CLDefocus-trained model generally yields lower full-reference scores than the DPDD-trained model. This behavior is attributable to imperfections in real ground-truth images, including residual de￾focus, brightness inconsistencies, and spatial misalignment, which bias pixel-wise measures such as PSNR and SSIM. We an… view at source ↗
Figure 4
Figure 4. Figure 4: Limitations of ground truth in real-captured defocus datasets. Geometric mis￾alignment caused by dynamic objects and photometric misalignment introduced by aperture changes lead to spatial and color inconsistencies, as reflected in the RGB his￾tograms. These imperfections result in a mismatch between full-reference metrics and visual quality, where perceptually sharper restorations may receive lower metric… view at source ↗
Figure 5
Figure 5. Figure 5: Comparison between the Debye CZT and the Rayleigh–Sommerfeld. The Debye CZT determines the sampling density and produces a stable, aliasing-free PSF without empirical tuning. In contrast, the Rayleigh–Sommerfeld provides no explicit sampling criterion, so we adjust the sampling density manually. Insufficient sampling leads to severe PSF warping and geometric distortion. Only sufficiently dense sampling yie… view at source ↗
Figure 6
Figure 6. Figure 6: PSF examples of various lens designs (column) and field positions (row), com￾puted by the Debye CZT. 5.5 Ablation Study We conduct an ablation study to evaluate the contributions of three components in our synthesis pipeline: (1) realistic lens PSFs, (2) the imaging pipeline, and (3) depth-varying scenes. For each experiment, we maintain all other components, including consistent patch sampling and augment… view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative ablation results illustrating the impact of each component. 6 Conclusion In this paper, we proposed a framework for synthesizing realistic defocus de￾blurring datasets for compound lenses. It integrates efficient wave-optics PSF computation, depth-aware defocus rendering with occlusion handling, and ISP￾aware blur synthesis in the radiometrically linear space. This unified framework enables sca… 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

70 extracted references · 70 canonical work pages

  1. [1]

    In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR)

    Abdelhamed, A., Lin, S., Brown, M.S.: A high-quality denoising dataset for smartphone cameras. In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR). pp. 1692–1700 (2018).https: //doi.org/10.1109/CVPR.2018.00182

    work page doi:10.1109/cvpr.2018.00182 2018

  2. [2]

    In: Vedaldi, A., Bischof, H., Brox, T., Frahm, J.M

    Abuolaim, A., Brown, M.S.: Defocus deblurring using dual-pixel data. In: Vedaldi, A., Bischof, H., Brox, T., Frahm, J.M. (eds.) Computer Vision – ECCV 2020. Lecture Notes in Computer Science, vol. 12355, pp. 111–126. Springer (2020). https://doi.org/10.1007/978-3-030-58607-2_7

    work page doi:10.1007/978-3-030-58607-2_7 2020

  3. [3]

    In: 15th Pacific Conference on Computer Graphics and Applications (PG)

    Bando, Y., Nishita, T.: Towards digital refocusing from a single photograph. In: 15th Pacific Conference on Computer Graphics and Applications (PG). pp. 363–

    work page

  4. [4]

    IEEE (2007).https://doi.org/10.1109/PG.2007.22

    work page doi:10.1109/pg.2007.22 2007

  5. [5]

    Advanced Optical Technologies10(3), 145–232 (2021).https://doi.org/ 10.1515/aot-2021-0023

    Blahnik, V., Schindelbeck, O.: Smartphone imaging technology and its applica- tions. Advanced Optical Technologies10(3), 145–232 (2021).https://doi.org/ 10.1515/aot-2021-0023

    work page doi:10.1515/aot-2021-0023 2021

  6. [6]

    IEEE Transactions on Audio and Electroacoustics18(4), 451–455 (1970).https://doi.org/10.1109/TAU.1970.1162132

    Bluestein, L.: A linear filtering approach to the computation of discrete Fourier transform. IEEE Transactions on Audio and Electroacoustics18(4), 451–455 (1970).https://doi.org/10.1109/TAU.1970.1162132

    work page doi:10.1109/tau.1970.1162132 1970

  7. [7]

    In: International Conference on Learning Representations (ICLR) (2025) Realistic Compound-Lens Defocus Blur Synthesis 37

    Bochkovskii, A., Delaunoy, A., Germain, H., Santos, M., Zhou, Y., Richter, S.R., Koltun, V.: Depth Pro: Sharp monocular metric depth in less than a second. In: International Conference on Learning Representations (ICLR) (2025) Realistic Compound-Lens Defocus Blur Synthesis 37

    work page 2025

  8. [8]

    Progress in Optics51, 349–468 (2008)

    Braat, J.J.M., van Haver, S., Janssen, A.J.E.M., Dirksen, P.: Assessment of optical systems by means of point-spread functions. Progress in Optics51, 349–468 (2008). https://doi.org/10.1016/S0079-6638(07)51006-1

    work page doi:10.1016/s0079-6638(07)51006-1 2008

  9. [9]

    Bradbury, J., Frostig, R., Hawkins, P., Johnson, M.J., Leary, C., Maclaurin, D., Necula, G., Paszke, A., VanderPlas, J., Wanderman-Milne, S., Zhang, Q.: JAX: Composable transformations of Python+Numpy programs.https://github.com/ jax-ml/jax(2018)

    work page 2018

  10. [10]

    Optics Communications450, 329–334 (2019).https://doi.org/10.1016/j.optcom.2019.06.022

    Cai, Y., Wang, Z., Liang, Y., Ren, F., Yao, B., Lei, M., Yan, S.: Direct calculation of tightly focused field in an arbitrary plane. Optics Communications450, 329–334 (2019).https://doi.org/10.1016/j.optcom.2019.06.022

    work page doi:10.1016/j.optcom.2019.06.022 2019

  11. [11]

    IEEE Transactions on Image Processing33, 2404–2418 (2024).https://doi.org/ 10.1109/TIP.2024.3378466

    Chen, C., Mo, J., Hou, J., Wu, H., Liao, L., Sun, W., Yan, Q., Lin, W.: TOPIQ: A top-down approach from semantics to distortions for image quality assessment. IEEE Transactions on Image Processing33, 2404–2418 (2024).https://doi.org/ 10.1109/TIP.2024.3378466

    work page doi:10.1109/tip.2024.3378466 2024

  12. [12]

    In: Avidan, S., Brostow, G., Cissé, M., Farinella, G.M., Hassner, T

    Chen, L., Chu, X., Zhang, X., Sun, J.: Simple baselines for image restoration. In: Avidan, S., Brostow, G., Cissé, M., Farinella, G.M., Hassner, T. (eds.) Computer Vision – ECCV 2022. Lecture Notes in Computer Science, vol. 13667, pp. 17–33. Springer (2022).https://doi.org/10.1007/978-3-031-20071-7_2

    work page doi:10.1007/978-3-031-20071-7_2 2022

  13. [13]

    In: IEEE Conf

    Chen, L., Li, Y., Dai, J., Gu, J., Xue, T.: A physics-informed blur learning framework for imaging systems. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 10913–10922 (2025). https://doi.org/10.1109/CVPR52734.2025.01019

    work page doi:10.1109/cvpr52734.2025.01019 2025

  14. [14]

    ACM Transactions on Graphics40(5), 192:1–192:15 (2021).https://doi.org/10.1145/3474088

    Chen, S., Feng, H., Pan, D., Xu, Z., Li, Q., Chen, Y.: Optical aberrations correction in postprocessing using imaging simulation. ACM Transactions on Graphics40(5), 192:1–192:15 (2021).https://doi.org/10.1145/3474088

    work page doi:10.1145/3474088 2021

  15. [15]

    IEEE Transactions on Image Processing 25(4), 1660–1673 (2016).https://doi.org/10.1109/TIP.2016.2526907

    D’Andrès, L., Salvador, J., Kochale, A., Süsstrunk, S.: Non-parametric blur map regression for depth of field extension. IEEE Transactions on Image Processing 25(4), 1660–1673 (2016).https://doi.org/10.1109/TIP.2016.2526907

    work page doi:10.1109/tip.2016.2526907 2016

  16. [17]

    Conte, G

    Elmacioglu, Y.G., Conlin, R., Dudt, D.W., Panici, D., Kolemen, E.: ZERNIPAX: A fast and accurate Zernike polynomial calculator in Python. Applied Mathematics and Computation505, 129534 (2025).https://doi.org/10.1016/j.amc.2025. 129534

    work page doi:10.1016/j.amc.2025 2025

  17. [18]

    Combining Appearance and Gradient Information for Image Symmetry Detection

    Foi, A., Trimeche, M., Katkovnik, V., Egiazarian, K.: Practical Poissonian- Gaussian noise modeling and fitting for single-image raw-data. IEEE Transactions on Image Processing17(10), 1737–1754 (2008).https://doi.org/10.1109/TIP. 2008.2001399

    work page doi:10.1109/tip 2008

  18. [19]

    Goodman, J.W.: Introduction to Fourier optics. W. H. Freeman, 4th edn. (2017)

    work page 2017

  19. [20]

    In: Proceedings of the IEEE International Conference on Computer Vision (ICCV)

    Hasinoff, S.W., Kutulakos, K.N.: A layer-based restoration framework for variable- aperture photography. In: Proceedings of the IEEE International Conference on Computer Vision (ICCV). pp. 1–8. IEEE (2007).https://doi.org/10.1109/ ICCV.2007.4408898

    work page arXiv 2007

  20. [21]

    In: International Conference on Learning Represen- tations (ICLR) (2019)

    Hendrycks,D.,Dietterich,T.:Benchmarkingneuralnetworkrobustnesstocommon corruptions and perturbations. In: International Conference on Learning Represen- tations (ICLR) (2019)

    work page 2019

  21. [22]

    Epona: Autoregressive Diffusion World Model for Autonomous Driving

    Ho, C.J., Belhe, Y., Rotenberg, S., Ramamoorthi, R., Li, T.M., Antipa, N.: A dif- ferentiable wave optics model for end-to-end computational imaging system opti- mization.In:ProceedingsoftheIEEE/CVFInternationalConferenceonComputer 38 Y. Lee et al. Vision (ICCV). pp. 28042–28051 (2025).https://doi.org/10.1109/ICCV51701. 2025.02603

    work page doi:10.1109/iccv51701 2025

  22. [23]

    Light: Science & Applications9(1), 119 (2020).https://doi

    Hu, Y., Wang, Z., Wang, X., Ji, S., Zhang, C., Li, J., Zhu, W., Wu, D., Chu, J.: Efficient full-path optical calculation of scalar and vector diffraction using the Bluestein method. Light: Science & Applications9(1), 119 (2020).https://doi. org/10.1038/s41377-020-00362-z

    work page doi:10.1038/s41377-020-00362-z 2020

  23. [24]

    Emerging

    Ke, J., Wang, Q., Wang, Y., Milanfar, P., Yang, F.: MUSIQ: Multi-scale image quality transformer. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 5128–5137 (2021).https://doi.org/10.1109/ ICCV48922.2021.00510

    work page arXiv 2021

  24. [25]

    IEEE Transactions on Image Processing29, 7389–7398 (2020).https://doi.org/10.1109/TIP.2020.3002345

    Kong, T., Sun, F., Liu, H., Jiang, Y., Li, L., Shi, J.: FoveaBox: Beyound anchor- based object detection. IEEE Transactions on Image Processing29, 7389–7398 (2020).https://doi.org/10.1109/TIP.2020.3002345

    work page doi:10.1109/tip.2020.3002345 2020

  25. [26]

    1365-2966.2004.08384.x

    Kraus, M., Strengert, M.: Depth-of-field rendering by pyramidal image processing. Computer Graphics Forum26(3), 645–654 (2007).https://doi.org/10.1111/j. 1467-8659.2007.01088.x

    work page doi:10.1111/j 2007

  26. [27]

    In: Advances in Neural Information Processing Systems (NeurIPS)

    Krishnan, D., Fergus, R.: Fast image deconvolution using hyper-laplacian priors. In: Advances in Neural Information Processing Systems (NeurIPS). pp. 1033–1041 (2009)

    work page 2009

  27. [28]

    Optical Engineering58(8), 082419 (2019)

    Kupinski, M.K., Bradley, C.L., Diner, D.J., Xu, F., Chipman, R.A.: Angle of linear polarization images of outdoor scenes. Optical Engineering58(8), 082419 (2019). https://doi.org/10.1117/1.OE.58.8.082419

    work page doi:10.1117/1.oe.58.8.082419 2019

  28. [29]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Lee, J., Lee, S., Cho, S., Lee, S.: Deep defocus map estimation using domain adaptation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 12222–12230 (2019).https://doi.org/10. 1109/CVPR.2019.01250

    work page arXiv 2019

  29. [30]

    P., Huang, Z., Romero, A

    Lee, J., Son, H., Rim, J., Cho, S., Lee, S.: Iterative filter adaptive network for single image defocus deblurring. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 2034–2042 (2021).https: //doi.org/10.1109/CVPR46437.2021.00207

    work page doi:10.1109/cvpr46437.2021.00207 2034

  30. [31]

    Optics Express14(23), 11277–11291 (2006).https://doi.org/10.1364/oe.14

    Leutenegger, M., Rao, R., Leitgeb, R.A., Lasser, T.: Fast focus field calculations. Optics Express14(23), 11277–11291 (2006).https://doi.org/10.1364/oe.14. 011277

    work page doi:10.1364/oe.14 2006

  31. [32]

    In: Proceedings of the AAAI Conference on Artificial Intelligence (AAAI)

    Li, Y., Ren, D., Shu, X., Zuo, W.: Learning single image defocus deblurring with misaligned training pairs. In: Proceedings of the AAAI Conference on Artificial Intelligence (AAAI). vol. 37, pp. 1495–1503 (2023).https://doi.org/10.1609/ aaai.v37i2.25235

    work page 2023

  32. [33]

    Nature Methods9(7), 637 (2012).https: //doi.org/10.1038/nmeth.2083

    Ljosa, V., Sokolnicki, K.L., Carpenter, A.E.: Annotated high-throughput mi- croscopy image sets for validation. Nature Methods9(7), 637 (2012).https: //doi.org/10.1038/nmeth.2083

    work page doi:10.1038/nmeth.2083 2012

  33. [34]

    IEEE Transactions on Pattern Analysis and Machine Intelligence46(8), 5541–5555 (2024).https://doi.org/10.1109/TPAMI

    Luo, J., Nie, Y., Ren, W., Cao, X., Yang, M.H.: Correcting optical aberration via depth-aware point spread functions. IEEE Transactions on Pattern Analysis and Machine Intelligence46(8), 5541–5555 (2024).https://doi.org/10.1109/TPAMI. 2024.3370794

    work page doi:10.1109/tpami 2024

  34. [35]

    IEEE Transactions on Image Processing31, 216–226 (2022).https://doi.org/10.1109/TIP.2021.3127850

    Ma, H., Liu, S., Liao, Q., Zhang, J., Xue, J.H.: Defocus image deblurring net- work with defocus map estimation as auxiliary task. IEEE Transactions on Image Processing31, 216–226 (2022).https://doi.org/10.1109/TIP.2021.3127850

    work page doi:10.1109/tip.2021.3127850 2022

  35. [36]

    Journal of the Optical Society of America A21(4), 510–516 (2004).https://doi.org/10.1364/josaa.21.000510 Realistic Compound-Lens Defocus Blur Synthesis 39

    Marathay, A.S., McCalmont, J.F.: On the usual approximation used in the Rayleigh–Sommerfeld diffraction theory. Journal of the Optical Society of America A21(4), 510–516 (2004).https://doi.org/10.1364/josaa.21.000510 Realistic Compound-Lens Defocus Blur Synthesis 39

    work page doi:10.1364/josaa.21.000510 2004

  36. [37]

    Optics Express32(16), 27278–27302 (2024).https://doi.org/10.1364/oe.523532

    Miora, R.H.D., Rohwer, E., Kielhorn, M., Sheppard, C., Bosman, G., Heintzmann, R.: Calculating point spread functions: methods, pitfalls, and solutions. Optics Express32(16), 27278–27302 (2024).https://doi.org/10.1364/oe.523532

    work page doi:10.1364/oe.523532 2024

  37. [38]

    completely blind

    Mittal, A., Soundararajan, R., Bovik, A.C.: Making a “completely blind” image quality analyzer. IEEE Signal Processing Letters20(3), 209–212 (2013).https: //doi.org/10.1109/LSP.2012.2227726

    work page doi:10.1109/lsp.2012.2227726 2013

  38. [39]

    Towards vision-language mechanistic interpretability: A causal tracing tool for BLIP

    Müller, P., Braun, A., Keuper, M.: Classification robustness to common optical aberrations. In: Proceedings of the IEEE/CVF International Conference on Com- puter Vision Workshops (ICCVW). pp. 3634–3645 (2023).https://doi.org/10. 1109/ICCVW60793.2023.00391

    work page arXiv 2023

  39. [40]

    IEEE Transactions on Pattern Analysis and Machine Intelligence48(3), 2139–2153 (2026).https://doi.org/10

    Müller, P., Braun, A., Keuper, M.: Examining the impact of optical aberrations to image classification and object detection models. IEEE Transactions on Pattern Analysis and Machine Intelligence48(3), 2139–2153 (2026).https://doi.org/10. 1109/TPAMI.2025.3622234

    work page arXiv 2026

  40. [41]

    IEEE Transactions on Computational Imaging9, 607–619 (2023).https://doi.org/10.1109/TCI

    Nazir, S., Vaquero, L., Mucientes, M., Brea, V.M., Coltuc, D.: Depth estimation and image restoration by deep learning from defocused images. IEEE Transactions on Computational Imaging9, 607–619 (2023).https://doi.org/10.1109/TCI. 2023.3288335

    work page doi:10.1109/tci 2023

  41. [42]

    Journal of Optics 24(12), 123001 (2022).https://doi.org/10.1088/2040-8986/ac9e08

    Niu, K., Tian, C.: Zernike polynomials and their applications. Journal of Optics 24(12), 123001 (2022).https://doi.org/10.1088/2040-8986/ac9e08

    work page doi:10.1088/2040-8986/ac9e08 2022

  42. [43]

    ACM Transactions on Graphics1(2), 85–108 (1982).https://doi

    Potmesil, M., Chakravarty, I.: Synthetic image generation with a lens and aperture camera model. ACM Transactions on Graphics1(2), 85–108 (1982).https://doi. org/10.1145/357299.357300

    work page doi:10.1145/357299.357300 1982

  43. [44]

    VideoMAE V2: Scaling Video Masked Autoencoders with Dual Masking

    Quan, Y., Wu, Z., Ji, H.: Neumann network with recursive kernels for single image defocus deblurring. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 5754–5763 (2023).https://doi. org/10.1109/CVPR52729.2023.00557

    work page doi:10.1109/cvpr52729.2023.00557 2023

  44. [45]

    Berg, Wan-Yen Lo, Piotr Doll ´ar, and Ross B

    Quan, Y., Yao, X., Ji, H.: Single image defocus deblurring via implicit neural inverse kernels. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 12566–12576 (2023).https://doi.org/10.1109/ ICCV51070.2023.01158

    work page arXiv 2023

  45. [46]

    org/10.1109/TAU.1969.1162034

    Rabiner,L.R.,Schafer,R.W.,Rader,C.M.:ThechirpZ-transformalgorithm.IEEE Transactions on Audio and Electroacoustics17(2), 86–92 (1969).https://doi. org/10.1109/TAU.1969.1162034

    work page doi:10.1109/tau.1969.1162034 1969

  46. [47]

    IEEE Transactions on Pattern Analysis and Machine Intelligence44(3), 1623–1637 (2022).https://doi.org/10.1109/TPAMI.2020.3019967

    Ranftl, R., Lasinger, K., Hafner, D., Schindler, K., Koltun, V.: Towards robust monocular depth estimation: Mixing datasets for zero-shot cross-dataset transfer. IEEE Transactions on Pattern Analysis and Machine Intelligence44(3), 1623–1637 (2022).https://doi.org/10.1109/TPAMI.2020.3019967

    work page doi:10.1109/tpami.2020.3019967 2022

  47. [48]

    In: International Conference on Learning Representations (ICLR) (2025)

    Ravi, N., Gabeur, V., Hu, Y.T., Hu, R., Ryali, C., Ma, T., Khedr, H., Rädle, R., Rolland, C., Gustafson, L., Mintun, E., Pan, J., Alwala, K.V., Carion, N., Wu, C.Y., Girshick, R., Dollár, P., Feichtenhofer, C.: SAM 2: Segment anything in images and videos. In: International Conference on Learning Representations (ICLR) (2025)

    work page 2025

  48. [49]

    Reiley, D.J.: Lens designs.https://www.lens-designs.com/, accessed: 2026-02-10

    work page 2026

  49. [50]

    Reasons to Fall (More) in Love with Combinatorial Reconfiguration

    Rim, J., Kim, G., Kim, J., Lee, J., Lee, S., Cho, S.: Realistic blur synthesis for learning image deblurring. In: Avidan, S., Brostow, G., Cissé, M., Farinella, G.M., Hassner, T. (eds.) Computer Vision – ECCV 2022. Lecture Notes in Computer Science, vol. 13667, pp. 487–503. Springer (2022).https://doi.org/10.1007/978- 3-031-20071-7_29 40 Y. Lee et al

    work page doi:10.1007/978- 2022

  50. [51]

    IEEE Transactions on Computational Imaging7, 675–688 (2021).https://doi.org/10.1109/TCI.2021.3092891

    Ruan, L., Chen, B., Li, J., Lam, M.L.: AIFNet: All-in-focus image restoration network using a light field-based dataset. IEEE Transactions on Computational Imaging7, 675–688 (2021).https://doi.org/10.1109/TCI.2021.3092891

    work page doi:10.1109/tci.2021.3092891 2021

  51. [52]

    In: 2022 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR)

    Ruan, L., Chen, B., Li, J., Lam, M.L.: Learning to deblur using light field gen- erated and real defocus images. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 16283–16292 (2022). https://doi.org/10.1109/CVPR52688.2022.01582

    work page doi:10.1109/cvpr52688.2022.01582 2022

  52. [53]

    IEEE Transactions on Image Processing25(7), 3141–3156 (2016).https://doi

    Saad, E., Hirakawa, K.: Defocus blur-invariant scale-space feature extractions. IEEE Transactions on Image Processing25(7), 3141–3156 (2016).https://doi. org/10.1109/TIP.2016.2555702

    work page doi:10.1109/tip.2016.2555702 2016

  53. [54]

    SPIE Press (2010).https://doi.org/10.1117/3.866274

    Schmidt, J.D.: Numerical simulation of optical wave propagation with examples in MATLAB. SPIE Press (2010).https://doi.org/10.1117/3.866274

    work page doi:10.1117/3.866274 2010

  54. [55]

    Berg, Wan-Yen Lo, Piotr Doll ´ar, and Ross B

    Shah, S., Kulshrestha, S., Metzler, C.A.: TiDy-PSFs: Computational imaging with time-averaged dynamic point-spread-functions. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 10623–10633 (2023). https://doi.org/10.1109/ICCV51070.2023.00978

    work page doi:10.1109/iccv51070.2023.00978 2023

  55. [56]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Shi, J., Xu, L., Jia, J.: Discriminative blur detection features. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 2965–2972 (2014).https://doi.org/10.1109/CVPR.2014.379

    work page doi:10.1109/cvpr.2014.379 2014

  56. [57]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Shi, J., Xu, L., Jia, J.: Just noticeable defocus blur detection and estimation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 657–665 (2015).https://doi.org/10.1109/CVPR.2015. 7298665

    work page doi:10.1109/cvpr.2015 2015

  57. [58]

    McGraw- Hill, 4th edn

    Smith, W.J.: Modern optical engineering: The design of optical systems. McGraw- Hill, 4th edn. (2008)

    work page 2008

  58. [59]

    Emerging

    Son, H., Lee, J., Cho, S., Lee, S.: Single image defocus deblurring using kernel- sharing parallel atrous convolutions. In: Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision (ICCV). pp. 2622–2630 (2021).https: //doi.org/10.1109/ICCV48922.2021.00264

    work page doi:10.1109/iccv48922.2021.00264 2021

  59. [60]

    In: ACM SIG- GRAPH 2024 Conference Papers

    Teh, A., Gkioulekas, I., O’Toole, M.: Aperture-aware lens design. In: ACM SIG- GRAPH 2024 Conference Papers. pp. 117:1–117:10. ACM (2024).https://doi. org/10.1145/3641519.3657398

    work page doi:10.1145/3641519.3657398 2024

  60. [61]

    ACM Transac- tions on Graphics40(2), 18:1–18:19 (2021).https://doi.org/10.1145/3446791

    Tseng, E., Mosleh, A., Mannan, F., St-Arnaud, K., Sharma, A., Peng, Y., Braun, A., Nowrouzezahrai, D., Lalonde, J.F., Heide, F.: Differentiable compound optics and processing pipeline optimization for end-to-end camera design. ACM Transac- tions on Graphics40(2), 18:1–18:19 (2021).https://doi.org/10.1145/3446791

    work page doi:10.1145/3446791 2021

  61. [62]

    IEEE Transactions on Computational Imaging 8, 905–916 (2022).https://doi.org/10.1109/TCI.2022.3212837

    Wang, C., Chen, N., Heidrich, W.: dO: A differentiable engine for deep lens design of computational imaging systems. IEEE Transactions on Computational Imaging 8, 905–916 (2022).https://doi.org/10.1109/TCI.2022.3212837

    work page doi:10.1109/tci.2022.3212837 2022

  62. [63]

    In: Proceedings of the AAAI Conference on Artificial Intelligence (AAAI)

    Wang, J., Chan, K.C.K., Loy, C.C.: Exploring CLIP for assessing the look and feel of images. In: Proceedings of the AAAI Conference on Artificial Intelligence (AAAI). vol. 37, pp. 2555–2563 (2023).https://doi.org/10.1609/aaai.v37i2. 25353

    work page doi:10.1609/aaai.v37i2 2023

  63. [64]

    Ciagan: Conditional identity anonymization generative adversarial networks,

    Wang, L., Li, D., Zhu, Y., Tian, L., Shan, Y.: Dual super-resolution learn- ing for semantic segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 3773–3782 (2020). https://doi.org/10.1109/CVPR42600.2020.00383

    work page doi:10.1109/cvpr42600.2020.00383 2020

  64. [65]

    In: Advances in Neural Information Processing Systems (NeurIPS)

    Yang,L.,Kang,B.,Huang,Z.,Zhao,Z.,Xu,X.,Feng,J.,Zhao,H.:DepthAnything V2. In: Advances in Neural Information Processing Systems (NeurIPS). vol. 37, pp. 21875–21911 (2024) Realistic Compound-Lens Defocus Blur Synthesis 41

    work page 2024

  65. [66]

    Desi: Deepfake source identifier for social media,

    Yang, S., Wu, T., Shi, S., Lao, S., Gong, Y., Cao, M., Wang, J., Yang, Y.: MANIQA: Multi-dimension attention network for no-reference image quality as- sessment. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). pp. 1190–1199 (2022).https: //doi.org/10.1109/CVPRW56347.2022.00126

    work page doi:10.1109/cvprw56347.2022.00126 2022

  66. [67]

    Nature Communications15(1), 6572 (2024).https://doi.org/ 10.1038/s41467-024-50835-7

    Yang, X., Fu, Q., Heidrich, W.: Curriculum learning for ab initio deep learned refractive optics. Nature Communications15(1), 6572 (2024).https://doi.org/ 10.1038/s41467-024-50835-7

    work page doi:10.1038/s41467-024-50835-7 2024

  67. [68]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops (ICCVW)

    Yang, X., Nguyen, C., Wang, W., Kang, K., Heidrich, W., Li, X.: Efficient depth- and spatially-varying image simulation for defocus deblur. In: Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops (ICCVW). pp. 3047–3057 (2025).https://doi.org/10.1109/ICCVW69036.2025.00318

    work page doi:10.1109/iccvw69036.2025.00318 2025

  68. [69]

    In: 2022 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR)

    Zamir, S.W., Arora, A., Khan, S., Hayat, M., Khan, F.S., Yang, M.H.: Restormer: Efficient transformer for high-resolution image restoration. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 5718–5729 (2022).https://doi.org/10.1109/CVPR52688.2022.00564

    work page doi:10.1109/cvpr52688.2022.00564 2022

  69. [70]

    In: 2018 IEEE/CVF Con- ference on Computer Vision and Pattern Recognition

    Zhang, R., Isola, P., Efros, A.A., Shechtman, E., Wang, O.: The unreasonable effec- tiveness of deep features as a perceptual metric. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 586–595 (2018).https://doi.org/10.1109/CVPR.2018.00068

    work page doi:10.1109/cvpr.2018.00068 2018

  70. [71]

    Lee et al

    Zhuo,S.,Sim,T.:Defocusmapestimationfromasingleimage.PatternRecognition 44(9), 1852–1858 (2011).https://doi.org/10.1016/j.patcog.2011.03.009 42 Y. Lee et al. (a) Dense PSF grids (b) PSFs of various lenses Fig. S11:PSFs of various lenses, depths, and fields. Each grid corresponds to a distinct lens, where the row and column correspond to the field and depth...

    work page doi:10.1016/j.patcog.2011.03.009 2011