Pith. sign in

REVIEW 4 major objections 6 minor 77 references

Optical aberrations in autonomous driving: Physics-informed parameterized temperature scaling for neural network uncertainty calibration

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read By feeding a predicted optical-aberration vector into a parameterized temperature-scaling calibrator, this paper cuts the mean expected calibration error of a semantic segmentation network under windshield-induced blur by roughly 250 ppm…

desk verdict A plausible proof-of-concept that does not yet isolate the physical prior as the cause of the calibration gain; worth reviewing but the main claim needs a control ablation. read the letter →

arxiv 2412.13695 v2 pith:Y4WXLT2W submitted 2024-12-18 cs.CV

Dominik Werner Wolf , Alexander Braun , Markus Ulrich This is my paper · ORCID
classification cs.CV
keywords neuralnetworkcalibrationpredictiveuncertaintydatasetshiftwavefrontaberrationsautonomousdrivingsemanticsegmentationparameterizedtemperaturescalingZernikecoefficients
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that the optical quality of the windshield, expressed as a vector of Zernike aberration coefficients, carries information about how badly a self-driving camera's perception network will be miscalibrated, and that a calibrator can exploit this information to keep confidence scores honest. It proposes a new post-hoc calibration method, Physics-Informed Parameterized Temperature Scaling (PIPTS), which feeds the predicted Zernike coefficient vector of the optical system into a neural-network temperature scaler alongside the segmentation logits. Tested on the A2D2 driving dataset with synthetic windshield-like blur, PIPTS reduces the mean expected calibration error (mECE) of a semantic segmentation model by roughly 250 ppm in the mean- and large-aberration regime compared with parameterized temperature scaling, with the gain declared statistically significant at the 95% level. If correct, this establishes that a physical prior about the imaging chain can improve the trustworthiness of uncertainty estimates under dataset shift, not just prediction accuracy.

What carries the argument

The central object is the Zernike coefficient vector of the effective wavefront aberration map, predicted by a residual encoder head in a multi-task UNet. Zernike coefficients are the weights of an orthogonal polynomial decomposition of the optical path difference across the aperture; in this paper they are confined to the three second-radial-order terms (oblique astigmatism, defocus, and orthogonal astigmatism). This vector serves as the physical inductive bias: it is concatenated with a CNN-encoded logit tensor and fed into a post-hoc calibrator that predicts an instance-wise temperature, so the calibrator can condition its temperature on the estimated severity and type of aberration rather than on the logits alone. The calibrator is trained end-to-end with a smoothed, modulated ECE loss, making the binning-based calibration error differentiable enough for backpropagation.

What would settle it

Take a set of real windshields, measure their actual aberration coefficients with a wavefront sensor, pass those measured coefficients through the PIPTS pipeline alongside the same segmentation logits, and check whether the roughly 250 ppm mECE improvement over PTS persists; if it vanishes or reverses, the synthetic uniform aberration distribution does not stand in for the real windshield population.

Watch

Extended reading notes

Core claim

The central discovery is that the Zernike coefficient vector of the optical system, estimated online by a multi-task network that predicts it from the input image, adds calibration-relevant signal beyond what the logit tensor alone provides. When this vector is concatenated with a CNN-encoded logit tensor and fed to a small network that predicts an instance-wise temperature, the resulting PIPTS calibrator lowers the mECE of the semantic segmentation model relative to the PTS baseline throughout the mean- and large-aberration range, with a statistically significant boost of about 250 ppm at the 95% confidence level, assessed with a deep ensemble of 11 calibrators. The improvement is not statistically significant near the diffraction limit, where the physical prior adds little information, and it diminishes near the center of the training aberration distribution, consistent with the prior being most useful in long-tail regimes.

Load-bearing premise

The predicted aberration coefficients, trained on synthetic blur with coefficients drawn uniformly from one wavelength, are accurate and representative enough of real windshield optics to add calibration information beyond the network's own confidence scores.

Editorial extensions

If this is right

  • PIPTS can keep the confidence estimates of a semantic segmentation network calibrated under optical aberrations caused by windshields, so an autonomous driving system can rely on its uncertainty estimates when the optical path degrades.
  • The multi-task network's predicted Zernike coefficients also serve end-of-line testing, absorbing the non-linear interplay of camera and windshield aberrations and enabling part-specific optical requirements.
  • Choosing the right optical metric matters: the Strehl ratio and the Optical Informative Gain (OIG) correlate with AI performance with Chatterjee rank correlation 0.75, while the MTF at half Nyquist correlates at only 0.54, making half-Nyquist MTF requirements invalid in the large-aberration regime.
  • The roughly 250 ppm calibration gain, scaled over a large fleet, corresponds to an increased safety margin that the authors quantify as 500 Gm of additional safe driving distance for 10 million cars over 200 Mm lifetimes.
  • A network trained on aberration-augmented data is best calibrated at the mean aberration level of its training distribution, implying that part tolerances should be centered on the expected optical quality.

Reading between the lines

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

  • A direct testable extension would be to replace the synthetic uniform sampling of Zernike coefficients with measured windshield aberration statistics: if the real distribution is narrower or skewed, the same architecture would show where the calibration gain concentrates, and the training augmentation could be re-centered accordingly.
  • The same conditioning idea should transfer to other dataset shifts with a measurable physical cause, such as weather blur, motion blur, defocus, or atmospheric turbulence, where a low-dimensional physical descriptor could replace the Zernike vector as the inductive bias.
  • Because the gain is largest in the low- and high-aberration tails, PIPTS-like calibrators could serve as an online monitoring signal: a large deviation between the predicted temperature and the training-mode temperature may flag an out-of-distribution optical state, potentially more cheaply than a separate OOD detector.
  • The authors implicitly treat the predicted Zernike vector as a clean channel; if that prediction is corrupted, the benefit should degrade. An ablation that adds noise to the Zernike input would quantify how much of the gain depends on the accuracy of the regression head versus the raw physics prior.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes Physics-Informed Parameterized Temperature Scaling (PIPTS), a post-hoc calibration method that extends Parameterized Temperature Scaling (PTS) by additionally feeding the predicted Zernike coefficient vector of the windshield/camera system into an instance-wise temperature predictor. The Zernike coefficients are produced by a multi-task segmentation/restoration/regression network trained on A2D2 images degraded by a Fourier-optical model with coefficients sampled uniformly in [-lambda, lambda]. The authors evaluate TS, PTS, and PIPTS in terms of mECE, report a statistically significant mECE reduction of roughly 250 ppm for PIPTS over PTS in the mean- and large-aberration regime, and analyze the rank correlation between optical merit functions and AI performance metrics, concluding that the Strehl ratio and the Optical Informative Gain are better optical proxies than the MTF at half Nyquist frequency.

Significance. If the attribution to the physical prior were properly isolated, this would be a practically relevant contribution to automotive perception verification: it would show that optical system knowledge, obtainable at low marginal cost, improves post-hoc confidence calibration under a realistic dataset shift. The paper has clear strengths: it uses a public dataset, provides a thoughtful treatment of the non-differentiability of the ECE through loss smoothing and modulation, and uses ensemble-based uncertainty estimates for the reported mECE curves. The comparison is empirical rather than forced by construction, and I see no circularity in the core experiment. However, the central attribution is not yet established, because PIPTS differs from PTS by adding an input feature, and the physical-content channel is not isolated from the mere addition of extra features. The synthetic-only evaluation also limits the applied claims, a point the authors themselves acknowledge in Section IV.A.

major comments (4)
  1. [Section IV.I and V.B (Figure 8)] The central claim that the physical prior adds calibration-relevant information is not isolated by the current experimental design. PIPTS concatenates the predicted three-dimensional Zernike vector to the encoded logits, while PTS uses only the encoded logits, so the comparison conflates the addition of any three input features with the physical content of those features. Please add at least (i) a control with three non-physical or task-agnostic features (for example, random features drawn from the same marginal distribution, or simple image statistics such as mean intensity and blur estimates) concatenated in the same way, and (ii) a PIPTS variant using ground-truth Zernike coefficients to bound the information lost by the regression head. Without these controls, the mECE gain cannot be attributed to the physics prior.
  2. [Section IV.B, IV.E, V] The accuracy of the Zernike regression head is never reported. The multi-task network is trained with an L2 Zernike loss (Section IV.E.3), but there is no held-out MAE/RMSE, R^2, or error distribution per coefficient, nor any analysis of how regression error varies with aberration magnitude. Since PIPTS consumes the predicted coefficients, a noisy or biased regression head could either dilute or distort the apparent benefit of the physical prior. Please report these metrics and, if possible, the calibration gain as a function of Zernike prediction error.
  3. [Section V.B] The statistical comparison is not fully specified. The text states that an ensemble of 11 PIPTS models is trained and that the Student t factor k10=2.23 is used, but it does not state whether the PTS baseline is trained with the same smoothed-ECE loss, the same calibrator architecture, and the same ensemble size. If PTS is trained differently, for example with the original NLL objective from the PTS paper, or with fewer ensemble members, the reported confidence intervals may reflect training-objective or ensemble-size asymmetry rather than the effect of the Zernike input. Please state the PTS training protocol and ensemble size explicitly.
  4. [Equation (3)] The definition of mECE is ambiguous and likely missing normalization. As written, mECE = sum_m (|B_m|/N_c)|acc(B_m)-conf(B_m)|, where |B_m| appears to be a bin cardinality and N_c the number of classes; this does not normalize by the total number of pixels and does not make the summation over classes explicit. Please provide the full formula, including the binning over pixels and classes, so that the absolute mECE values and the claimed 250 ppm gain are reproducible.
minor comments (6)
  1. [Abstract and Section I] The phrase 'bijective mapping' between AI performance requirements and optical metrics is not supported by the experimental evidence, which demonstrates rank correlation and a regression with unexplained variance; please replace 'bijective' with a weaker statement such as 'a consistently measurable correspondence'.
  2. [Section IV.I] The sentence 'the number seven to be the best compromise' is missing a noun; it should read 'seven encoder blocks'. In addition, the PIPTS calibrator architecture (filter counts, dense layers, dropout rates) is not described in enough detail for reproduction.
  3. [Equation (11)] The combined uncertainty formula as written equates sigma_c/k to the square-root expression; the coverage factor k should multiply the standard uncertainty on the right-hand side. Please correct the notation.
  4. [Throughout] There are minor language issues, for example 'it's simplicity' should be 'its simplicity', 'widely adapted' should probably be 'widely adopted', and 'Huellermeier' is usually written 'Hüllermeier'.
  5. [Figure 8 and Section V.B] The numerical values behind the mECE curves and confidence bands are only shown graphically; please provide a table with mean mECE, standard error, and 95% confidence intervals for TS, PTS, and PIPTS in the regimes where significance is claimed.
  6. [Section IV.A and VII] The paper's own caveat that the uniform aberration sampling does not reflect production complexity should be restated in the Abstract and Conclusion, since the fleet-level safety interpretation in Section V.B depends on it.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the PIPTS calibration gain is an empirical comparison, not a construction, and the paper's self-citations are not load-bearing.

full rationale

The central claim is that feeding a predicted Zernike coefficient vector into the PIPTS calibrator reduces mECE relative to PTS (Section IV.I, Section V.B, Figure 8). This is an empirical result: both PIPTS and PTS are trained as post-hoc calibrators with the same smoothed-ECE objective, and the comparison is evaluated on a held-out test set. No equation in the paper reduces the claimed gain to an input by construction; the Zernike vector is a predicted feature, not a fitted parameter renamed as a prediction. The paper does have limitations that the reader should weigh: the accuracy of the Zernike regression head (Section IV.B) is never reported, and the PIPTS-versus-PTS comparison conflates 'extra input features' with 'physical information' because no ablation removes the Zernike channel while keeping the same network capacity. These are validity and attribution concerns, not circularity. The paper also cites the authors' prior work on OIG and on decoupling of calibration measures (refs [5], [10], [37]), but those citations supply background metrics and measure-selection arguments; the main calibration comparison also relies on the standard Strehl ratio and the standard mECE of Naeini et al., so the self-citations are not load-bearing. The abstract's 'bijective mapping' is asserted as motivation rather than derived, but the experimental conclusion does not depend on that premise. Accordingly, the derivation chain is self-contained against external benchmarks and no circular step is present.

Assumptions & free parameters 4 free parameters · 7 assumptions · 0 invented entities

The central claim rests on several domain assumptions about the synthetic degradation model, the accuracy of the predicted Zernike coefficients, and the transferability of tuned hyperparameters. No new physical entities are introduced. The free parameters include augmentation choices and hyperparameters that are tuned but not fully reported.

free parameters (4)
  • Synthetic aberration sampling range = alpha_n in [-lambda, lambda], lambda unspecified
    Augmentation distribution for training and test; chosen as a physically sound baseline, explicitly not matched to production tolerances (Section IV.A).
  • ECE loss modulation hyperparameters = beta_s=1000, eta=50, kappa=8
    Set by hyperparameter tuning; shape differentiability of the training loss and temperature regularizer (Section IV.I).
  • Multi-task network hyperparameters = not reported numerically
    Loss weighting factors, learning rate, focal exponent gamma, batchsize, and PIPTS encoder block count (7) are tuned but specific values are not all given (Sections IV.D-F, IV.I).
  • Regression function coefficients beta_1..beta_5 = fitted per metric, values not tabulated
    Exponential-plus-linear model in Equation (10) fitted to mIoU/mECE versus optical metrics; used for sensitivity analysis and confidence bands (Section V.A).
assumptions (7)
  • standard math Zernike decomposition and Fourier optics (OTF/PSF, Plancherel theorem) correctly describe the imaging chain.
    Invoked in Section III.A and used to generate synthetic degradations and define OIG.
  • domain assumption Windshield aberrations are dominated by second radial order coefficients (oblique astigmatism, defocus, orthogonal astigmatism), with tilt excluded.
    Section IV.A restricts the Zernike vector to radial order two; if higher-order terms dominate, the physical prior is incomplete.
  • domain assumption Synthetic Fourier degradation with uniformly sampled Zernike coefficients models the real dataset shift caused by windshields.
    Section IV.A states this does not guarantee production process complexity, so the assumption is acknowledged but untested against real part tolerances.
  • domain assumption The Zernike regression head provides sufficiently accurate coefficients for calibration.
    PIPTS consumes the predicted Zernike vector (Sections IV.B, IV.I), yet the paper never reports regression accuracy or uncertainty.
  • domain assumption The smoothed ECE with softmax (beta_s=1000) is a valid differentiable training objective, and eta=50, kappa=8 transfer across datasets and architectures.
    Section IV.I asserts transferability as long as normalization and ECE bin characteristics are unchanged, but no cross-dataset evidence is provided.
  • ad hoc to paper A bijective mapping exists between AI performance requirements and optical metrics.
    The abstract and introduction assume this mapping, but the paper only demonstrates correlation, not bijectivity or a constructed mapping.
  • standard math Chatterjee rank correlation and the law of variance decomposition are valid tools for the analysis.
    Used in Section III.C to quantify non-linear correlation between optical quality and mIoU/mECE.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Optical aberrations in autonomous driving: Physics-informed parameterized temperature scaling for neural network uncertainty calibration." pith.science (2026). https://pith.science/paper/Y4WXLT2W

@misc{pith2026241213695,
  author       = {Pith},
  title        = {Pith review of: Optical aberrations in autonomous driving: Physics-informed parameterized temperature scaling for neural network uncertainty calibration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Y4WXLT2W}},
  note         = {Machine review of arXiv:2412.13695}
}
read the original abstract

'A trustworthy representation of uncertainty is desirable and should be considered as a key feature of any machine learning method' (Huellermeier and Waegeman, 2021). This conclusion of Huellermeier et al. underpins the importance of calibrated uncertainties. Since AI-based algorithms are heavily impacted by dataset shifts, the automotive industry needs to safeguard its system against all possible contingencies. One important but often neglected dataset shift is caused by optical aberrations induced by the windshield. For the verification of the perception system performance, requirements on the AI performance need to be translated into optical metrics by a bijective mapping. Given this bijective mapping it is evident that the optical system characteristics add additional information about the magnitude of the dataset shift. As a consequence, we propose to incorporate a physical inductive bias into the neural network calibration architecture to enhance the robustness and the trustworthiness of the AI target application, which we demonstrate by using a semantic segmentation task as an example. By utilizing the Zernike coefficient vector of the optical system as a physical prior we can significantly reduce the mean expected calibration error in case of optical aberrations. As a result, we pave the way for a trustworthy uncertainty representation and for a holistic verification strategy of the perception chain.

Figures

Figures reproduced from arXiv: 2412.13695 by the authors.

Figure 2
Figure 2. The Chatterjee’s rank correlation measure [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 1
Figure 1. The Chatterjee’s rank correlation measure [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 3
Figure 3. The layout of the multi-task network for semantic segmentation and for predicting the effective Zernike coefficients [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Loss function study for the PIPTS calibration network. The loss is indicated for a random instance as a function of [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: The dependency of the mIoU (upper row) and the mECE (lower row) on the MTF at half Nyquist frequency (left [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Segmentation map predicted by the multi-task net [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: The reliability diagrams for the multi-task network are presented if calibrated by: (left) TS, (middle) PTS, and [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: On the left-hand side, the mECE curves for the PTS and PIPTS calibrator are plotted versus the TS calibration [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: The temperature deviation ∆T between the predicted temperature of the (top) PTS, (bottom) PIPTS calibrator and the optimal temperature is plotted as a histogram. Both distri￾butions indicate a significant bias, which reflects the influence of the dataset shift on the p…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

77 extracted references · 62 canonical work pages

  1. [1]

    Aleatoric and epistemic uncer- tainty in machine learning: an introduction to concepts and meth- ods,

    E. Huellermeier and W. Waegeman, “Aleatoric and epistemic uncer- tainty in machine learning: an introduction to concepts and meth- ods,” Machine Learning, vol. 110, 2021, URL: https://doi.org/10.1007/ s10994-021-05946-3

    work page 2021

  2. [2]

    Systems and methods for detecting traffic lights,

    I. B. Shalom, “Systems and methods for detecting traffic lights,” U.S. Patent US9 272 709B2, March 1, 2016, URL: https://patents.google.com/ patent/US9272709B2/en

    work page 2016

  3. [3]

    Vision-based machine learning model for autonomous driving with adjustable virtual camera,

    J. Emmons, D. Hung, E. Knight, and L. McIntosh, “Vision-based machine learning model for autonomous driving with adjustable virtual camera,” U.S. Patent US20 230 057 509A1, February 23, 2023, URL: https://patents.google.com/patent/US20230057509A1/en

    work page 2023

  4. [4]

    Neural network trajectory prediction,

    Y . Chen, B. Ivanovic, and M. Pavone, “Neural network trajectory prediction,” U.S. Patent US20 230 391 374A1, December 7, 2023, URL: https://patents.google.com/patent/US20230391374A1/en

    work page 2023

  5. [5]

    Sensitivity analysis of AI-based algorithms for autonomous driving on optical wavefront aberrations induced by the windshield,

    D. W. Wolf, M. Ulrich, and N. Kapoor, “Sensitivity analysis of AI-based algorithms for autonomous driving on optical wavefront aberrations induced by the windshield,” 2023 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW) , pp. 4102–4111, 10 2023, URL: https://ieeexplore.ieee.org/document/10350923/

    work page arXiv 2023

  6. [6]

    Obtaining Well Calibrated Probabilities Using Bayesian Binning,

    M. Pakdaman Naeini, G. Cooper, and M. Hauskrecht, “Obtaining Well Calibrated Probabilities Using Bayesian Binning,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 29, no. 1, 2 2015, URL: https://ojs.aaai.org/index.php/AAAI/article/view/9602

    work page 2015

  7. [7]

    On the Calibration of Un- certainty Estimation in LiDAR-based Semantic Segmentation,

    M. Dreissig, F. Piewak, and J. Boedecker, “On the Calibration of Un- certainty Estimation in LiDAR-based Semantic Segmentation,” in IEEE 26th International Conference on Intelligent Transportation Systems (ITSC). IEEE, 2023, pp. 4798–4805

    work page 2023

  8. [8]

    A mathematical theory of communication,

    C. E. Shannon, “A mathematical theory of communication,” The Bell System Technical Journal, vol. 27, no. 3, pp. 379–423, 1948

    1948

Show all 77 references
  1. [9]

    Time-dynamic estimates of the reliability of deep semantic segmentation networks,

    K. Maag, M. Rottmann, and H. Gottschalk, “Time-dynamic estimates of the reliability of deep semantic segmentation networks,” IEEE 32nd International Conference on Tools with Artificial Intelligence (ICTAI) , pp. 502–509, 2020

    2020

  2. [10]

    Decoupling of neural network calibration measures,

    D. W. Wolf, P. Balaji, A. Braun, and M. Ulrich, “Decoupling of neural network calibration measures,” German Conference on Pattern Recognition (GCPR), 2024

    2024

  3. [11]

    Parameterized temperature scaling for boosting the expressive power in post-hoc uncertainty cali- bration,

    C. Tomani, D. Cremers, and F. Buettner, “Parameterized temperature scaling for boosting the expressive power in post-hoc uncertainty cali- bration,” European Conference on Computer Vision (ECCV) , pp. 555– 569, 2022

    2022

  4. [12]

    Physics-Informed Computer Vision: A Review and Perspectives,

    C. Banerjee, K. Nguyen, C. Fookes, and K. George, “Physics-Informed Computer Vision: A Review and Perspectives,” ACM Comput. Surv. , vol. 57, no. 1, Oct. 2024, URL: https://doi.org/10.1145/3689037

    2024 doi

  5. [13]

    On calibration of modern neural networks,

    C. Guo, G. Pleiss, Y . Sun, and K. Q. Weinberger, “On calibration of modern neural networks,” International conference on machine learning (ICML), pp. 1321–1330, 2017

    2017

  6. [14]

    D. L. Goodstein, States of matter , ser. Prentice-Hall physics series. Englewood Cliffs, NJ: Prentice-Hall, 1975

    1975

  7. [15]

    Calibrating language models with adaptive temperature scaling,

    J. Xie, A. S. Chen, Y . Lee, E. Mitchell, and C. Finn, “Calibrating language models with adaptive temperature scaling,” in Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, Y . Al-Onaizan, M. Bansal, and Y .-N. Chen, Eds. Miami, Florida, U...

    2024

  8. [16]

    Mix-n-Match: ensemble and compositional methods for uncertainty calibration in deep learning,

    J. Zhang, B. Kailkhura, and T. Y .-J. Han, “Mix-n-Match: ensemble and compositional methods for uncertainty calibration in deep learning,” in Proceedings of the 37th International Conference on Machine Learning , ser. ICML’20. JMLR.org, 2020

    2020

  9. [17]

    Sample- dependent adaptive temperature scaling for improved calibration,

    T. Joy, F. Pinto, S.-N. Lim, P. H. S. Torr, and P. K. Dokania, “Sample- dependent adaptive temperature scaling for improved calibration,” in Proceedings of the Thirty-Seventh AAAI Conference on Artificial Intelli- gence and Thirty-Fifth Conference on Innovative Applications of...

    2023 doi

  10. [18]

    Auto-Encoding Variational Bayes,

    D. P. Kingma and M. Welling, “Auto-Encoding Variational Bayes,” CoRR, vol. abs/1312.6114, 2013, URL: https://api.semanticscholar.org/ CorpusID:216078090

    2013 arXiv

  11. [19]

    ATS: Adap- tive Temperature Scaling for Enhancing Out-of-Distribution Detection Methods,

    G. Krumpl, H. Avenhaus, H. Possegger, and H. Bischof, “ATS: Adap- tive Temperature Scaling for Enhancing Out-of-Distribution Detection Methods,” in 2024 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2024, pp. 3852–3861

    2024

  12. [20]

    Rudolph, J

    A. Rudolph, J. Krois, and K. Hartmann. (2023) Statistics and Geodata Analysis using Python (SOGA-Py), Department of Earth Sciences, Freie Universitaet Berlin. [Online]. Available: https://www.geo.fu-berlin.de/en/v/soga-py/Basics-of-statistics/ Hypothesis-Tests/Introduction-to-...

    2023

  13. [21]

    Upton and I

    G. Upton and I. Cook, A Dictionary of Statistics , ser. Oxford Paperback Reference. OUP Oxford, 2008, URL: https://books.google.de/books? id=u97pzxRjaCQC

    2008

  14. [22]

    A survey of uncertainty in deep neural networks,

    J. Gawlikowski, C. R. N. Tassi, M. Ali, J. Lee, M. Humt, J. Feng, A. Kruspe, R. Triebel, P. Jung, R. Roscher, M. Shahzad, W. Yang, R. Bamler, and X. X. Zhu, “A survey of uncertainty in deep neural networks,” Artificial Intelligence Review, vol. 56, no. 1, pp. 1513–1589, 2023, ...

    2023 doi

  15. [23]

    Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles,

    B. Lakshminarayanan, A. Pritzel, and C. Blundell, “Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles,” Neural In- formation Processing Systems , 2016, URL: https://api.semanticscholar. org/CorpusID:6294674

    2016

  16. [24]

    DUDES: Deep Uncertainty Distillation using Ensembles for Semantic Seg- mentation,

    S. Landgraf, K. Wursthorn, M. Hillemann, and M. Ulrich, “DUDES: Deep Uncertainty Distillation using Ensembles for Semantic Seg- mentation,” PFG –Journal of Photogrammetry, Remote Sensing and Geoinformation Science , vol. 92, no. 2, pp. 101–114, 2024, URL: https://doi.org/10.10...

    2024 doi

  17. [25]

    Efficient multi- task uncertainties for joint semantic segmentation and monocular depth estimation,

    S. Landgraf, M. Hillemann, T. Kapler, and M. Ulrich, “Efficient multi- task uncertainties for joint semantic segmentation and monocular depth estimation,” German Conference on Pattern Recognition (GCPR), 2024

    2024

  18. [26]

    Diffraction Theory of the Knife-Edge Test and its Improved Form, The Phase-Contrast Method,

    F. Zernike and F. Stratton, “Diffraction Theory of the Knife-Edge Test and its Improved Form, The Phase-Contrast Method,” Monthly Notices of the Royal Astronomical Society, vol. 94, no. 5, pp. 377–384, 03 1934, URL: https://doi.org/10.1093/mnras/94.5.377

    1934 doi

  19. [27]

    Beugungstheorie des schneidenver-fahrens und seiner verbesserten form, der phasenkontrastmethode,

    F. Zernike, “Beugungstheorie des schneidenver-fahrens und seiner verbesserten form, der phasenkontrastmethode,” Physica, vol. 1, no. 7, pp. 689–704, 1934, URL: https://www.sciencedirect.com/science/article/ pii/S0031891434802595

    1934

  20. [28]

    On the circle polynomials of Zernike and related orthogonal sets,

    A. B. Bhatia and E. Wolf, “On the circle polynomials of Zernike and related orthogonal sets,” Mathematical Proceedings of the Cambridge Philosophical Society, vol. 50, no. 1, pp. 40–48, 1954

    1954

  21. [29]

    Adaptive optics: Neural network wavefront sensing, reconstruction, and prediction,

    P. C. McGuire, D. G. Sandler, M. Lloyd-Hart, and T. A. Rhoadarmer, “Adaptive optics: Neural network wavefront sensing, reconstruction, and prediction,” in Scientific Applications of Neural Nets , J. W. Clark, T. Lindenau, and M. L. Ristig, Eds. Berlin, Heidelberg: Springer Ber...

    1999

  22. [30]

    Image-based wavefront sensing for astronomy using neural networks,

    T. E. Andersen, M. Owner-Petersen, and A. Enmark, “Image-based wavefront sensing for astronomy using neural networks,” Journal of Astronomical Telescopes, Instruments, and Systems , vol. 6, no. 3, p. 034002, 2020, URL: https://doi.org/10.1117/1.JATIS.6.3.034002

    2020 doi

  23. [31]

    Adaptive Optics for the ESO Very Large Telescope,

    F. Merkle and N. Hubin, “Adaptive Optics for the ESO Very Large Telescope,” in Adaptive Optics for Large Telescopes . Optica Pub- lishing Group, 1992, URL: https://opg.optica.org/abstract.cfm?URI= AOLT-1992-ATuA4

    1992

  24. [32]

    Adaptive optics for high-resolution imaging,

    K. M. Hampson, R. Turcotte, D. T. Miller, K. Kurokawa, J. R. Males, N. Ji, and M. J. Booth, “Adaptive optics for high-resolution imaging,” Nature Reviews Methods Primers, vol. 1, no. 1, p. 68, 2021, URL: https: //doi.org/10.1038/s43586-021-00066-7

    2021 doi

  25. [33]

    Physics-Driven Turbulence Image Restoration with Stochastic Refinement,

    A. Jaiswal, X. Zhang, S. H. Chan, and Z. Wang, “Physics-Driven Turbulence Image Restoration with Stochastic Refinement,” in 2023 IEEE/CVF International Conference on Computer Vision (ICCV) . Los Alamitos, CA, USA: IEEE Computer Society, Oct 2023, pp. 12 136– 12 147, URL: https...

    2023

  26. [34]

    An Image is Worth 16x16 Words: IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. XX, NO. X, APRIL 2025 15 Transformers for Image Recognition at Scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, “An Image is Worth 16x16 Words: IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. XX, NO. X, APRIL 2025 1...

    2025

  27. [35]

    A New Coefficient of Correlation,

    S. Chatterjee, “A New Coefficient of Correlation,” Journal of the American Statistical Association , vol. 116, no. 536, pp. 2009–2022, 2021

    2009

  28. [36]

    On the power of Chatterjees rank correlation,

    H. Shi, M. Drton, and F. Han, “On the power of Chatterjees rank correlation,” Biometrika, vol. 109, no. 2, pp. 317–333, 10 2021

    2021

  29. [37]

    Windscreen Optical Quality for AI Algorithms: Refractive Power and MTF not Sufficient,

    D. W. Wolf, M. Ulrich, and A. Braun, “Windscreen Optical Quality for AI Algorithms: Refractive Power and MTF not Sufficient,” 2023 IEEE 26th International Conference on Intelligent Transportation Sys- tems (ITSC) , pp. 5190–5197, 9 2023, URL: https://ieeexplore.ieee.org/ docum...

    2023

  30. [38]

    J. W. Goodman, Introduction to Fourier optics . McGraw-Hill, 1968

    1968

  31. [39]

    Physiological Control Systems : Analysis, Simulation, and Estimation,

    M. C. Khoo, “Physiological Control Systems : Analysis, Simulation, and Estimation,” 2018, iEEE Press series in biomedical engineering, ISBN: 978-1-119-05879-3

    2018

  32. [40]

    Calibration of transient groundwater models using time series analysis and moment matching,

    M. Bakker, K. Maas, and J. R. V on Asmuth, “Calibration of transient groundwater models using time series analysis and moment matching,” 2008, water Resources Research, V ol.44. URL: https://doi.org/10.1029/ 2007WR006239

    2008

  33. [41]

    MTF as a performance indicator for AI algorithms?

    P. Mueller and A. Braun, “MTF as a performance indicator for AI algorithms?” Electronic Imaging , vol. 35, no. 16, pp. 125–1–125–1, 2023, URL: https://library.imaging.org/ei/articles/35/16/A VM-125

    2023

  34. [42]

    Principles of Harmonic Analysis,

    A. Deitmar and S. Echterhoff, “Principles of Harmonic Analysis,” 2008, springer New York. ISBN: 9780387854687

    2008

  35. [43]

    Image Segmentation Using Deep Learning: A Survey,

    S. Minaee, Y . Boykov, F. Porikli, A. Plaza, N. Kehtarnavaz, and D. Terzopoulos, “Image Segmentation Using Deep Learning: A Survey,” 2020, URL: https://arxiv.org/abs/2001.05566

    2020 arXiv

  36. [44]

    Uncertainty Quantification with Deep Ensembles for 6D Object Pose Estimation,

    K. Wursthorn, M. Hillemann, and M. Ulrich, “Uncertainty Quantification with Deep Ensembles for 6D Object Pose Estimation,” ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences , vol. X-2-2024, pp. 223–230, 2024, URL: https://isprs-annals.copernic...

    2024

  37. [45]

    Weiss, P

    N. Weiss, P. Holmes, and M. Hardy, A Course in Probability . Pearson Addison Wesley, 2005

    2005

  38. [46]

    Tilt-Then-Blur or Blur-Then-Tilt? Clarifying the Atmo- spheric Turbulence Model,

    S. H. Chan, “Tilt-Then-Blur or Blur-Then-Tilt? Clarifying the Atmo- spheric Turbulence Model,” IEEE Signal Processing Letters , vol. 29, pp. 1833–1837, 2022

    2022

  39. [47]

    Novel developments of refractive power measurement techniques in the automotive world,

    D. W. Wolf, M. Ulrich, and A. Braun, “Novel developments of refractive power measurement techniques in the automotive world,” Metrolo- gia, vol. 60, 9 2023, URL: https://iopscience.iop.org/article/10.1088/ 1681-7575/acf1a4

    2023

  40. [48]

    Deep Residual Learning for Image Recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778, 2016

    2016

  41. [49]

    Residual networks behave like ensembles of relatively shallow networks,

    A. Veit, M. Wilber, and S. Belongie, “Residual networks behave like ensembles of relatively shallow networks,” Proceedings of the 30th International Conference on Neural Information Processing Systems, pp. 550–558, 2016

    2016

  42. [50]

    Norm-Preservation: Why Residual Networks Can Become Extremely Deep?

    A. Zaeemzadeh, N. Rahnavard, and M. Shah, “Norm-Preservation: Why Residual Networks Can Become Extremely Deep?” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 43, pp. 3980–3990, 2018, URL: https://api.semanticscholar.org/CorpusID:29155803

    2018

  43. [51]

    A2D2: Audi Autonomous Driving Dataset,

    J. Geyer, Y . Kassahun, M. Mahmudi, X. Ricou, R. Durgesh, A. S. Chung, L. Hauswald, V . H. Pham, M. Mühlegg, S. Dorn, T. Fer- nandez, M. Jänicke, S. Mirashi, C. Savani, M. Sturm, O. V orobiov, M. Oelker, S. Garreis, and P. Schuberth, “A2D2: Audi Autonomous Driving Dataset,” ar...

    2020

  44. [52]

    The Cityscapes Dataset for Semantic Urban Scene Understanding,

    M. Cordts, M. Omran, S. Ramos, T. Rehfeld, M. Enzweiler, R. Benen- son, U. Franke, S. Roth, and B. Schiele, “The Cityscapes Dataset for Semantic Urban Scene Understanding,” in Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 2016

    2016

  45. [53]

    Focal Loss for Dense Object Detection,

    T.-Y . Lin, P. Goyal, R. Girshick, K. He, and P. Dollar, “Focal Loss for Dense Object Detection,” IEEE International Conference on Computer Vision (ICCV) , pp. 2999–3007, 2017, URL: https://doi. ieeecomputersociety.org/10.1109/ICCV .2017.324

    2017 doi

  46. [54]

    Loss Functions for Image Restoration With Neural Networks,

    H. Zhao, O. Gallo, I. Frosio, and J. Kautz, “Loss Functions for Image Restoration With Neural Networks,” IEEE Transactions on Computa- tional Imaging, vol. 3, no. 1, pp. 47–57, 2017

    2017

  47. [55]

    On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima,

    N. S. Keskar, D. Mudigere, J. Nocedal, M. Smelyanskiy, and P. T. P. Tang, “On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima,” 5th International Conference on Learning Repre- sentations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Trac...

    2017

  48. [56]

    Gaussian Error Linear Units (GELUs),

    D. Hendrycks and K. Gimpel, “Gaussian Error Linear Units (GELUs),” arXiv, 2023, URL: https://arxiv.org/abs/1606.08415

    2023 arXiv

  49. [57]

    Rectified linear units improve restricted Boltzmann machines,

    V . Nair and G. E. Hinton, “Rectified linear units improve restricted Boltzmann machines,” Proceedings of the 27th International Conference on International Conference on Machine Learning (ICML), pp. 807–814, 2010

    2010

  50. [58]

    Lecture Notes on Machine Learning: Neurons with Non-Monotonic Activation Functions,

    C. Bauckhage and D. Speicher, “Lecture Notes on Machine Learning: Neurons with Non-Monotonic Activation Functions,” University of Bonn, Tech. Rep., 07 2019

    2019

  51. [59]

    Fast and accurate deep network learning by exponential linear units (elus),

    D. Clevert, T. Unterthiner, and S. Hochreiter, “Fast and accurate deep network learning by exponential linear units (elus),” in 4th International Conference on Learning Representations, ICLR 2016, San Juan, Puerto Rico, May 2-4, 2016, Conference Track Proceedings , 2016, URL: ...

    2016 arXiv

  52. [60]

    Revisiting Orthogonality Regularization: A Study for Convolutional Neural Networks in Image Classification,

    T. Kim and S.-Y . Yun, “Revisiting Orthogonality Regularization: A Study for Convolutional Neural Networks in Image Classification,”IEEE Access, vol. 10, pp. 69 741–69 749, 2022

    2022

  53. [61]

    Deep Optics for Monocular Depth Es- timation and 3D Object Detection,

    J. Chang and G. Wetzstein, “Deep Optics for Monocular Depth Es- timation and 3D Object Detection,” in 2019 IEEE/CVF International Conference on Computer Vision (ICCV) , 2019, pp. 10 192–10 201

    2019

  54. [62]

    Differen- tiable Compound Optics and Processing Pipeline Optimization for End- to-end Camera Design,

    E. Tseng, A. Mosleh, F. Mannan, K. St-Arnaud, A. Sharma, Y . Peng, A. Braun, D. Nowrouzezahrai, J.-F. Lalonde, and F. Heide, “Differen- tiable Compound Optics and Processing Pipeline Optimization for End- to-end Camera Design,” ACM Transactions on Graphics, vol. 40, no. 2, 6 2...

    2021 doi

  55. [63]

    Image Quality Is Not All You Want: Task-Driven Lens Design for Image Classification,

    X. Yang, Q. Fu, Y . Nie, and W. Heidrich, “Image Quality Is Not All You Want: Task-Driven Lens Design for Image Classification,” 05 2023, URL: https://doi.org/10.48550/arXiv.2305.17185

    2023 doi

  56. [64]

    GUM-compliant neural network robustness verification,

    B. Ludwig, “GUM-compliant neural network robustness verification,” Master thesis at the Technical University of Berlin, Zuse Institute Berlin and Physikalisch-Technische Bundesanstalt, 2023

    2023

  57. [65]

    Introduction to Monte Carlo simulation,

    S. Raychaudhuri, “Introduction to Monte Carlo simulation,” Winter Simulation Conference, pp. 91–100, 2008

    2008

  58. [66]

    Evaluation of measure- ment data, Guide to the expression of uncertainty in measurement (GUM),

    J. C. for Guides in Metrology (JCGM), “Evaluation of measure- ment data, Guide to the expression of uncertainty in measurement (GUM),” International Bureau of Weights and Measures (BIPM) , 2008, URL: https://www.bipm.org/documents/20126/2071204/JCGM_ 100_2008_E.pdf/cb0ef43f-ba...

    2008

  59. [67]

    Pesch, Bestimmung der Messunsicherheit nach GUM, Grundlagen der Metrologie

    B. Pesch, Bestimmung der Messunsicherheit nach GUM, Grundlagen der Metrologie. Books on Demand (BoD), 2003

    2003

  60. [68]

    M. Taboga. (2021) Ridge regression, Lectures on probability theory and mathematical statistics. Kindle Direct Publishing. [Online]. Available: https://www.statlect.com/fundamentals-of-statistics/ridge-regression

    2021

  61. [69]

    Generalizations of Mean Square Error Applied to Ridge Regression,

    C. M. Theobald, “Generalizations of Mean Square Error Applied to Ridge Regression,” Journal of the Royal Statistical Society. Series B (Methodological), vol. 36, no. 1, pp. 103–106, 1974, URL: http://www. jstor.org/stable/2984775

    1974

  62. [70]

    Further results on the mean square error of ridge regression,

    R. W. Farebrother, “Further results on the mean square error of ridge regression,” Journal of the Royal Statistical Society. Series B (Method- ological), vol. 38, no. 3, pp. 248–250, 1976, URL: http://www.jstor.org/ stable/2984971

    1976

  63. [71]

    Estimating the mean and variance of the target probability distribution,

    D. Nix and A. Weigend, “Estimating the mean and variance of the target probability distribution,” in Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN’94) , vol. 1, no. 1, 1994, pp. 55–60

    1994

  64. [72]

    Cramér, Mathematical Methods of Statistics

    H. Cramér, Mathematical Methods of Statistics . Princeton: Princeton University Press, 1946, URL: https://doi.org/10.1515/9781400883868

    1946 doi

  65. [73]

    Information and the Accuracy Attainable in the Estimation of Statistical Parameters,

    C. Rao, “Information and the Accuracy Attainable in the Estimation of Statistical Parameters,” Bulletin of Calcutta Mathematical Society , vol. 37, pp. 81–91, 1945

    1945

  66. [74]

    Sur l’extension de certaines evaluations statistiques au cas de petits echantillons,

    M. Fréchet, “Sur l’extension de certaines evaluations statistiques au cas de petits echantillons,” Revue de l’Institut International de Statistique , vol. 11, no. 3, pp. 182–205, 1943, URL: http://www.jstor.org/stable/ 1401114

    1943

  67. [75]

    Automotive mass production of camera systems: Linking im- age quality to AI performance,

    A. Braun, “Automotive mass production of camera systems: Linking im- age quality to AI performance,” tm - Technisches Messen, vol. 90, no. 3, pp. 205–218, 2023, URL: https://doi.org/10.1515/teme-2022-0029

    2023 doi

  68. [76]

    D. D. Walden, T. M. Shortell, G. J. Roedler, B. A. Delicado, O. Mornas, Y . Yew-Seng, and D. Endler, INCOSE Systems Engineering Handbook , 5th ed. Wiley, 6 2023, URL: https://www.wiley.com/en-ie/INCOSE+ Systems+Engineering+Handbook%2C+5th+Edition-p-9781119814313

    2023

  69. [77]

    Wavefront aberration measurements based on the Background Oriented Schlieren method,

    D. W. Wolf, B. Thielbeer, M. Ulrich, and A. Braun, “Wavefront aberration measurements based on the Background Oriented Schlieren method,” Measurement: Sensors, p. 101509, 2024, URL: https://doi.org/ 10.1016/j.measen.2024.101509

    2024

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.