Pith. sign in

REVIEW 4 major objections 5 minor 46 references

DASC: Depth-of-Field Aware Scene Complexity Metric for 3D Visualization on Light Field Display

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A single scene-complexity score predicts viewers' preferred blur level on light field displays.

desk verdict Genuine empirical effort, but the DASC metric is ill-defined and the validation is too thin to support the predictive claim. read the letter →

arxiv 2508.08928 v1 pith:EKMI4DJQ submitted 2025-08-12 cs.MM cs.GR

Kamran Akbar , Robert Bregovic , Federica Battisti This is my paper · ORCID
classification cs.MMcs.GR
keywords lightfielddisplaydepthof3DvisualizationscenecomplexitymetricaliasingartifactsDoFrenderingsubjectivequalityassessmentperceptualpreference
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

This paper proposes a scene-complexity metric, DASC, that turns a 3D scene's geometry, texture, and position relative to a light field display's depth-of-field (DoF) into a single number. It then fits a sigmoid curve to subjective preference data so that the DASC value directly gives the blurring level—the circular-filter aperture radius—that viewers prefer when the scene is shown on such a display. If the model holds, content creators could skip subjective testing and set the DoF-rendering blur automatically. The claim is supported by paired-comparison experiments on nine synthetic light fields and a validation study with new scenes and participants.

What carries the argument

The load-bearing machinery is the DASC aggregation f=(1/m)Σ_i(1−ω(Ω_i))ν(Ω_i)ψ(Ω_i). Here ν is the mean of normalized curvature standard deviation, edge density, and entropy; ω is the fraction of the object's pixels inside the display's DoF; and ψ is a piecewise weight depending on where the object's depth range sits relative to the DoF. This condenses each object's geometric complexity and positional risk into one score. The preference mapping is then a sigmoid, r(f)=κ/(1+$e^{{β(f−γ)}}$), which is the device that converts the metric into a concrete rendering parameter.

What would settle it

Take a scene containing one object entirely inside the display's DoF, so ω(Ω)=1, and compute Eq. (18) using the fourth case of Eq. (20): the weight becomes −l(Ω)/(1−ω(Ω)), a division by zero, so f is undefined. A reader could also check whether the first case of Eq. (20) is ever satisfiable for a real object; if not, the aggregation rule needs correction before the sigmoid model can be evaluated.

Watch

Extended reading notes

Core claim

The central claim is that a single scalar, f, computed from per-object entropy, edge density, curvature variation, size, and placement relative to the display's DoF, predicts the preferred aperture radius r for DoF rendering on a light field display. The paper models the mapping as a sigmoid r(f)=κ/(1+$e^{{β(f−γ)}}$) with fitted coefficients κ=21.9, β=4.5, and γ=9.0, calibrated against Bradley-Terry preference scores from the correlation study. Scenes that sit essentially inside the display's DoF receive a mapping that avoids harmful blurring, while more complex scenes with content outside the DoF map to stronger filtering. A validation study using different scenes and a fresh set of participants reproduced the predicted preferred radii, so the paper claims the metric transfers across content.

Load-bearing premise

The load-bearing premise is that the piecewise aggregation in Eq. (20) is well-defined for real scenes: in particular, the case that weights an object fully inside the DoF by −l/(1−ω) is only meaningful when ω≠1, and the listed first case (Zmin>dΦ/2 with Zmax≤dΦ/2) cannot actually occur. If those clauses are typos, the intended correction must still make f computable for every scene the model is applied to; if they are not, DASC is undefined exactly for the scenes used to fit and validate the curve.

Editorial extensions

If this is right

  • For a new scene, a content creator can render segmentation and depth maps, compute f, and read the preferred aperture radius directly from the sigmoid, removing the need for per-scene subjective tests.
  • DoF rendering can be automated: the same pipeline that detects aliasing-prone regions can set the circular-filter radius from f, keeping in-DoF content sharp and blurring only where needed.
  • The model gives a concrete target for content preparation: alias-free scenes should be shown without additional blur, while complex scenes with substantial out-of-DoF content should receive stronger filtering.
  • Subjective preference appears content-dependent rather than purely observer-dependent, so a metric based on scene properties is the right type of predictor; only about a fifth of observers in the study had preferences stable across scenes.

Reading between the lines

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

  • The fitted coefficients κ, β, and γ are tied to the specific display's angular sampling and to the filter-radius units; recalibrating them for other light field displays is a direct test of the model's generality, which the paper itself lists as future work.
  • Because f depends only on geometry and texture, the same DASC computation could be applied to real captured light fields, not just synthetic renders, as a test of whether the preference curve survives realistic content; the paper's dataset is entirely synthetic.
  • The two-peak preference structure could also be modelled per observer rather than per scene: if a minority of viewers do have consistent preferences across scenes, a personalized version of the sigmoid might improve predictions for those viewers.
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 / 5 minor

Summary. The paper proposes DASC, a depth-of-field-aware scene complexity metric that combines per-object geometric factors (entropy, edge density, curvature standard deviation) and positional factors (DoF overlap, depth position, depth span) into a single score f. The score is then mapped through a fitted sigmoid to predict the aperture radius that viewers prefer on a light field display. The authors describe a paired-comparison subjective study on a Holovizio 722 RC display with six correlation-study scenes and three validation-study scenes, use the Bradley-Terry model to derive opinion scores, fit the sigmoid, and report that the validation scenes align with the model predictions.

Significance. If the metric were well-defined and the model reproducible, the work would be practically useful for content creators and would address a real gap: prior light field quality assessment is mostly done on 2D displays, whereas this study evaluates DoF-guided rendering on an actual light field display. The paper also contributes a custom synthetic light field dataset and a subjective testing protocol, and states that code and data are public. However, the central mathematical definition of DASC is internally inconsistent in ways that affect every computed scene score, and the reported sigmoid coefficients contradict the accompanying textual claim. These issues must be corrected before the empirical results can be trusted.

major comments (4)
  1. [Eq. (20)] The first branch of the piecewise definition of ψ is impossible: it requires Zmin(Ωi) > dΦ/2 and Zmax(Ωi) ≤ dΦ/2, which cannot hold because Zmin ≤ Zmax. Consequently, the piecewise function does not partition the depth axis and no case is defined for objects entirely in front of the DoF. Any object with Zmax ≤ dΦ/2, including a front-outside object at z ∈ [−2dΦ, −dΦ], falls into the fourth branch and receives ψ = −l/(1−ω); after multiplication by (1−ω) in Eq. (18), this contributes −lν to f, decreasing the complexity score for an object that is outside the DoF and prone to aliasing. This contradicts the stated modeling assumption and affects all subsequently reported DASC values.
  2. [Section IV-C, bullet list and Eq. (18)] The bullet defining the weight for objects fully inside the DoF as −l(Ωi)/(1−ω(Ωi)) is not a well-defined function because for a fully inside-DoF object ω(Ωi) = 1, making the denominator zero. Even if one silently cancels the (1−ω) factor against the same factor in Eq. (18), the resulting contribution is −l(Ωi)ν(Ωi), which can push f below the claimed lower bound of −1 whenever an object's depth span exceeds dΦ. The definition must be rewritten as a total function, either with explicit cancellation or with a separate branch for ω = 1.
  3. [Section VI-B, Eq. (26)] The reported sigmoid coefficients κ = 21.9, β = 4.5, γ = 9.0 are inconsistent with the accompanying statement that scenes with f between −1 and 0 receive r = 9. With these coefficients, r(−1) = 21.9/(1+e^{4.5(−10)}) ≈ 21.9 and r(0) ≈ 21.9, not 9; the value r = 9 occurs near f ≈ 9.1. As written, Eq. (26) and the stated coefficients cannot reproduce the mapping that the paper claims, so the preference model is not reproducible and the plotted correlation in Figure 8 cannot be checked.
  4. [Section VI-B and VI-C] The preference model is fitted using only six scenes that exhibit essentially two distinct preferred radii (r = 0 and r = 9). A three-parameter sigmoid fitted to two output levels is not identifiable, and the reported κ, β, γ cannot be meaningfully estimated from such data. The three-scene validation study is too small to support the strong conclusion that the model 'efficiently predicts the preferred level of blurring for any given 3D scene'; the authors partly acknowledge this in Section VII, but the generalization claim should be substantially tempered.
minor comments (5)
  1. [Eq. (12)] There is a typo: 'max s)' should be 'max(s)'.
  2. [Eqs. (15)-(17) and (20)] The depth notation is inconsistent: Eqs. (15)-(16) use Z(Ωi), Eq. (17) writes 'max(ZΩi)' with a missing parenthesis, and Eq. (20) uses Zmin/Zmax without mentioning Ωi. Please unify the notation.
  3. [Section VI-B] The text refers to 'the optimal radius r̂ mentioned in Table I', but Table I contains only likelihood-ratio values and no r̂ value; please clarify where r̂ is defined.
  4. [Figure 8 and Section VI] The DASC values f for the individual scenes are not reported in a table, so readers cannot verify the sigmoid fit or the claimed correlations. A table with f, preferred radius, and opinion-score peaks for each scene would greatly improve reproducibility.
  5. [Section IV-C] The sentence 'the score value f is scaled by the number of object m' should read 'number of objects m'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DASC is computed from scene geometry independently of subjective scores, the sigmoid mapping is an explicitly labeled regression, and the validation uses separate scenes and participants.

full rationale

The derivation chain is not circular in the prohibited sense. The DASC value f (Eq. 18) is computed from segmentation and depth maps, geometric factors (Eqs. 6-13), and position factors (Eqs. 14-17); none of these inputs are the subjective preference scores or the fitted aperture radii. The preference model is a three-parameter sigmoid (Eq. 26) whose coefficients κ=21.9, β=4.5, γ=9.0 are obtained by nonlinear least-squares fitting to the six correlation-scene (f, preferred r) pairs. This is explicitly presented as a regression ('The sigmoid function utilized for fitting is...'), not as a first-principles derivation, and predictions on the fitting scenes would indeed be reproductions of the fit. However, the paper does not present those six scenes as predictions; it tests the fitted curve on three new scenes (Flower, Camper, Garden) with new participants and reports that the estimated preferred radii (9, 0, 9) match. That validation is independent of the parameter fitting. Self-citations to [3] and [7] supply the DoF rendering filter and display-resolution model used to generate stimuli; they are experimental tools, and the claimed DASC-to-preference link does not reduce to those citations. The serious internal inconsistencies in the branch definition of ψ (Eq. 20), including an impossible first case and an undefined fourth case at ω=1, are correctness threats to the DASC computation rather than circularities: the metric is not defined in terms of the target preference, and correcting the branches would not change the input-output structure. Therefore, under the hard rule requiring a specific reduction to inputs, the circularity score is 0.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim depends on three fitted sigmoid parameters and an ad hoc equal weighting of scene features. The display model and optimal filter radius are borrowed from prior work, including two self-citations. No new physical entities are introduced.

free parameters (4)
  • sigmoid κ = 21.9
    Scale constant in Eq. (26), fitted to the six correlation-scene preferred radii using MATLAB fitnlm.
  • sigmoid β = 4.5
    Steepness constant in Eq. (26), fitted to the same data.
  • sigmoid γ = 9.0
    Midpoint constant in Eq. (26), fitted to the same data.
  • Equal geometric weights = 1/3 each
    Weights in Eq. (19) for entropy, edge density, and curvature; chosen by hand with no tuning or justification for equality.
assumptions (5)
  • domain assumption The display resolution model p(z) = p0 + |z| tan(αs) from [4] accurately describes the light field display's spatial resolution as a function of depth.
    Adopted from prior literature in Section II-B to define the depth of field; no validation is provided for the Holovizio display used.
  • domain assumption The optimal antialiasing filter radius r̂ = 0.5(αs/αc) from the authors' prior work [3] is correct.
    Used in Section VI-B to justify choosing blur radii greater than r̂; relies on a self-cited model.
  • standard math The Bradley-Terry model correctly estimates preference scores from paired comparisons.
    Standard model used in Section VI-A; assumed appropriate for the collected paired-comparison data.
  • ad hoc to paper The chosen geometric and position features (entropy, edge density, curvature, overlap, distance, size) are sufficient to characterize perceived aliasing and preferred blur.
    This is the core modeling assumption behind DASC; no evidence or comparison is given to show these features capture the relevant perceptual dimensions.
  • standard math The Likelihood Ratio Test with the χ² distribution at df=1 and α=0.05 is appropriate for comparing preferred aperture radii.
    Used in Section VI-A; standard statistical test, though the degrees of freedom and independence assumptions are not discussed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DASC: Depth-of-Field Aware Scene Complexity Metric for 3D Visualization on Light Field Display." pith.science (2026). https://pith.science/paper/EKMI4DJQ

@misc{pith2026250808928,
  author       = {Pith},
  title        = {Pith review of: DASC: Depth-of-Field Aware Scene Complexity Metric for 3D Visualization on Light Field Display},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EKMI4DJQ}},
  note         = {Machine review of arXiv:2508.08928}
}
read the original abstract

Light field display is one of the technologies providing 3D immersive visualization. However, a light field display generates only a limited number of light rays which results in finite angular and spatial resolutions. Therefore, 3D content can be shown with high quality only within a narrow depth range notated as Depth of Field (DoF) around the display screen. Outside this range, due to the appearance of aliasing artifacts, the quality degrades proportionally to the distance from the screen. One solution to mitigate the artifacts is depth of field rendering which blurs the content in the distorted regions, but can result in the removal of scene details. This research focuses on proposing a DoF Aware Scene Complexity (DASC) metric that characterizes 3D content based on geometrical and positional factors considering the light field display's DoF. In this research, we also evaluate the observers' preference across different level of blurriness caused by DoF rendering ranging from sharp, aliased scenes to overly smoothed alias-free scenes. We have conducted this study over multiple scenes that we created to account for different types of content. Based on the outcome of subjective studies, we propose a model that takes the value of DASC metric as input and predicts the preferred level of blurring for the given scene as output.

Figures

Figures reproduced from arXiv: 2508.08928 by the authors.

Figure 1
Figure 1. The smallest reproducible feature size at a specific distance [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the light field rendering and visualization setup. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Example of (a) a segmentation map and (b) depth map when [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Visualization of different blurring levels [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: The result of Bradley-Terry model which demonstrates the preference of different filtering radii for (a) [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: (a) The preferred aperture radii r for each participant and each scene. (b) The normal distribution of participants preference for aperture radii across all scenes. As shown in [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: The result of Bradley-Terry model which demonstrates the preference of different filtering radii for validation scenes, (a) [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 9
Figure 9. Figure 9: The comparison of the proposed model’s value between preferred [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 8
Figure 8. Figure 8: The correlation between the chosen circular filter radii and the [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 3 linked inside Pith

  1. [1]

    Levoy and P

    M. Levoy and P. Hanrahan, Light Field Rendering , 1st ed. New York, NY , USA: Association for Computing Machinery, 2023. [Online]. Available: https://doi.org/10.1145/3596711.3596759 JOURNAL OF IEEE TRANSACTION ON MULTIMEDIA 10

    arXiv 2023

  2. [2]

    Bregovic, E

    R. Bregovic, E. Sahin, S. Vagharshakyan, and A. Gotchev, Signal Processing Methods for Light Field Displays . Springer

  3. [3]

    Antialiasing filtering for projection-based light field displays,

    K. Akbar and R. Bregovic, “Antialiasing filtering for projection-based light field displays,” in 2023 International Symposium on Image and Signal Processing and Analysis (ISPA) , 2023, pp. 1–6

    2023

  4. [4]

    The holovizio system - new opportunity offered by 3d displays,

    T. Balogh, P. T. Kov ´acs, Z. Dobranyi, A. Barsi, Z. Megyesi, Z. Ga ´al, and G. Balogh, “The holovizio system - new opportunity offered by 3d displays,” 2008. [Online]. Available: https://api.semanticscholar.org/ CorpusID:110209951

    2008

  5. [5]

    Antialiasing for automultiscopic 3d displays,

    M. Zwicker, W. Matusik, F. Durand, and H. Pfister, “Antialiasing for automultiscopic 3d displays,” in Proceedings of the 17th Eurographics Conference on Rendering Techniques , ser. EGSR ’06. Goslar, DEU: Eurographics Association, 2006, p. 73–82

    2006

  6. [6]

    Dynamically reparameterized light fields,

    A. Isaksen, L. McMillan, and S. J. Gortler, “Dynamically reparameterized light fields,” ser. SIGGRAPH ’00. USA: ACM Press/Addison-Wesley Publishing Co., 2000, p. 297–306. [Online]. Available: https://doi.org/10.1145/344779.344929

    arXiv

  7. [7]

    Depth-of-field guided rendering for light field displays,

    K. Akbar and R. Bregovic, “Depth-of-field guided rendering for light field displays,” Electronic Imaging, vol. 36, no. 10, pp. 367–1–367–1,

  8. [8]

    Tensor displays: compressive light field synthesis using multilayer displays with directional backlighting,

    G. Wetzstein, D. Lanman, M. Hirsch, and R. Raskar, “Tensor displays: compressive light field synthesis using multilayer displays with directional backlighting,” vol. 31, no. 4, Jul. 2012. [Online]. Available: https://doi.org/10.1145/2185520.2185576

    arXiv 2012

Show all 46 references
  1. [9]

    Quan- tifying spatial and angular resolution of light-field 3-d displays,

    P. T. Kov ´acs, R. Bregovi ´c, A. Boev, A. Barsi, and A. Gotchev, “Quan- tifying spatial and angular resolution of light-field 3-d displays,” IEEE Journal of Selected Topics in Signal Processing , vol. 11, no. 7, pp. 1213–1222, 2017

    2017

  2. [10]

    Single-image evaluation of angular and spatial resolution for projection-based light field displays,

    K. Akbar and R. Bregovic, “Single-image evaluation of angular and spatial resolution for projection-based light field displays,” in 2023 IEEE 25th International Workshop on Multimedia Signal Processing (MMSP), 2023, pp. 1–6

    2023

  3. [11]

    Beyond perceptual thresholds and personal preference: Towards novel research questions and methodologies of quality of experience studies on light field visualization,

    P. A. Kara, R. R. Tamboli, E. Shafiee, M. G. Martini, A. Simon, and M. Guindy, “Beyond perceptual thresholds and personal preference: Towards novel research questions and methodologies of quality of experience studies on light field visualization,” Electronics, vol. 11, no. 6,...

    2022

  4. [12]

    A study on visual perception of light field content,

    A. Gill, E. Zerman, C. Ozcinar, and A. Smolic, “A study on visual perception of light field content,” CoRR, vol. abs/2008.03195, 2020. [Online]. Available: https://arxiv.org/abs/2008.03195

    2008 arXiv

  5. [13]

    A study on the impact of visu- alization techniques on light field perception,

    F. Battisti, M. Carli, and P. L. Callet, “A study on the impact of visu- alization techniques on light field perception,” in 2018 26th European Signal Processing Conference (EUSIPCO) , 2018, pp. 2155–2159

    2018

  6. [14]

    Charac- terization and selection of light field content for perceptual assessment,

    P. Paudyal, J. Guti ´errez, P. Le Callet, M. Carli, and F. Battisti, “Charac- terization and selection of light field content for perceptual assessment,” in 2017 Ninth International Conference on Quality of Multimedia Experience (QoMEX), 2017, pp. 1–6

    2017

  7. [15]

    Percep- tual quality of light field images and impact of visualization techniques,

    P. Paudyal, F. Battisti, P. Le Callet, J. Guti ´errez, and M. Carli, “Percep- tual quality of light field images and impact of visualization techniques,” IEEE Transactions on Broadcasting, vol. 67, no. 2, pp. 395–408, 2021

    2021

  8. [16]

    Reduced reference quality assess- ment of light field images,

    P. Paudyal, F. Battisti, and M. Carli, “Reduced reference quality assess- ment of light field images,” IEEE Transactions on Broadcasting, vol. 65, no. 1, pp. 152–165, 2019

    2019

  9. [17]

    Super- multiview content with high angular resolution: 3d quality assessment on horizontal-parallax lightfield display,

    R. R. Tamboli, B. Appina, S. Channappayya, and S. Jana, “Super- multiview content with high angular resolution: 3d quality assessment on horizontal-parallax lightfield display,” Signal Processing: Image Communication, vol. 47, pp. 42–55, 2016. [Online]. Available: https://www....

    2016

  10. [18]

    Towards a quality metric for dense light fields,

    V . K. Adhikarla, M. Vinkler, D. Sumin, R. K. Mantiuk, K. Myszkowski, H.-P. Seidel, and P. Didyk, “Towards a quality metric for dense light fields,” in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017, pp. 3720–3729

    2017

  11. [19]

    A metric for light field reconstruction, compression, and display quality evaluation,

    X. Min, J. Zhou, G. Zhai, P. Le Callet, X. Yang, and X. Guan, “A metric for light field reconstruction, compression, and display quality evaluation,” IEEE Transactions on Image Processing, vol. 29, pp. 3790– 3804, 2020

    2020

  12. [20]

    Investigating epipolar plane image representa- tions for objective quality evaluation of light field images,

    A. Ak and P. Le-Callet, “Investigating epipolar plane image representa- tions for objective quality evaluation of light field images,” in 2019 8th European Workshop on Visual Information Processing (EUVIP) , 2019, pp. 135–139

    2019

  13. [21]

    Natural image utility assessment using image contours,

    D. M. Rouse and S. S. Hemami, “Natural image utility assessment using image contours,” in Proceedings of the 16th IEEE International Conference on Image Processing , ser. ICIP’09. IEEE Press, 2009, p. 2193–2196

    2009

  14. [22]

    Gradient magnitude similarity deviation: A highly efficient perceptual image quality index,

    W. Xue, L. Zhang, X. Mou, and A. C. Bovik, “Gradient magnitude similarity deviation: A highly efficient perceptual image quality index,” IEEE Transactions on Image Processing , vol. 23, no. 2, pp. 684–695, 2014

    2014

  15. [23]

    Dibr-synthesized image quality assessment based on morphological multi-scale approach,

    D. Sandi ´c-Stankovi´c, D. Kukolj, and P. Le Callet, “Dibr-synthesized image quality assessment based on morphological multi-scale approach,” EURASIP Journal on Image and Video Processing, vol. 2017, no. 1, p. 4, 2016

    2017

  16. [24]

    A light field image quality assessment model based on symmetry and depth features,

    Y . Tian, H. Zeng, J. Hou, J. Chen, J. Zhu, and K.-K. Ma, “A light field image quality assessment model based on symmetry and depth features,” IEEE Transactions on Circuits and Systems for Video Technology , vol. 31, no. 5, pp. 2046–2050, 2021

    2021

  17. [25]

    Segnet: A deep con- volutional encoder-decoder architecture for image segmentation,

    V . Badrinarayanan, A. Kendall, and R. Cipolla, “Segnet: A deep con- volutional encoder-decoder architecture for image segmentation,” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 39, no. 12, pp. 2481–2495, 2017

    2017

  18. [26]

    Semantic image segmentation: Two decades of research,

    G. Csurka, R. V olpi, and B. Chidlovskii, “Semantic image segmentation: Two decades of research,” 2023. [Online]. Available: https://arxiv.org/ abs/2302.06378

    2023 arXiv

  19. [27]

    Deep robust single image depth estimation neural network using scene understanding,

    H. Ren, M. El-khamy, and J. Lee, “Deep robust single image depth estimation neural network using scene understanding,” 2019. [Online]. Available: https://arxiv.org/abs/1906.03279

    2019 arXiv

  20. [28]

    Edge-aware bidirectional dif- fusion for dense depth estimation from light fields,

    N. Khan, M. H. Kim, and J. Tompkin, “Edge-aware bidirectional dif- fusion for dense depth estimation from light fields,” in British Machine Vision Conference (BMVC), 2021

    2021

  21. [29]

    B. O. Community, Blender - a 3D modelling and rendering package , Blender Foundation, Stichting Blender Foundation, Amsterdam, 2018. [Online]. Available: http://www.blender.org

    2018

  22. [30]

    Entropy in image analysis,

    A. C. Sparavigna, “Entropy in image analysis,” Entropy, vol. 21, no. 5,

  23. [31]

    R. C. Gonzalez and R. E. Woods, Digital image processing . Upper Saddle River, N.J.: Prentice Hall, 2008. [Online]. Available: http://www. amazon.com/Digital-Image-Processing-3rd-Edition/dp/013168728X

    2008

  24. [32]

    C. I. Gonzalez, P. Melin, J. R. Castro, and O. Castillo, Edge Detection Methods and Filters Used on Digital Image Processing . Cham: Springer International Publishing, 2017, pp. 11–16. [Online]. Available: https://doi.org/10.1007/978-3-319-53994-2 3

    2017 doi

  25. [33]

    Hessian estimates for the lagrangian mean curvature flow,

    A. Bhattacharya and J. Wall, “Hessian estimates for the lagrangian mean curvature flow,” Calculus of Variations and Partial Differential Equations, vol. 63, 08 2024

    2024

  26. [34]

    CIVIT dataset: Horizontal-parallax-only densely-sampled light-fields,

    S. Moreschini, F. Gama, R. Bregovic, and A. Gotchev, “CIVIT dataset: Horizontal-parallax-only densely-sampled light-fields,” in Euro- pean Light Field Imaging (ELFI) Workshop, 2019, european Light Field Imaging Workshop ; Conference date: 04-06-2019 Through 06-06-2019

    2019

  27. [35]

    Saliency detection on light field,

    N. Li, J. Ye, Y . Ji, H. Ling, and J. Yu, “Saliency detection on light field,” in The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2014

    2014

  28. [36]

    New light field image dataset,

    M. ˇReˇr´abek and T. Ebrahimi, “New light field image dataset,” in 8th International Workshop on Quality of Multimedia Experience (QoMEX), Lisbon, Portugal, 2016

    2016

  29. [37]

    LiFF: Light field features in scale and depth,

    D. G. Dansereau, B. Girod, and G. Wetzstein, “LiFF: Light field features in scale and depth,” in Computer Vision and Pattern Recognition (CVPR) . IEEE, Jun. 2019. [Online]. Available: http://dgd.vision/Papers/dansereau2019liff.pdf

    2019

  30. [38]

    Recommendation 500-10: Methodology for the subjective assessment of the quality of television pictures,

    “Recommendation 500-10: Methodology for the subjective assessment of the quality of television pictures,” ITU-R Rec. BT.500, 2000

    2000

  31. [39]

    Analysis and improvement of a paired comparison method in the application of 3dtv subjective experiment,

    J. Li, M. Barkowsky, and P. Le Callet, “Analysis and improvement of a paired comparison method in the application of 3dtv subjective experiment,” in 2012 19th IEEE International Conference on Image Processing, 2012, pp. 629–632

    2012

  32. [40]

    Die berechnung der turnier-ergebnisse als ein maximumproblem der wahrscheinlichkeitsrechnung,

    E. Zermelo, “Die berechnung der turnier-ergebnisse als ein maximumproblem der wahrscheinlichkeitsrechnung,” Mathematische Zeitschrift, vol. 29, 1929

    1929

  33. [41]

    New York, NY: Springer New York, 2008, pp

    Likelihood Ratio Test . New York, NY: Springer New York, 2008, pp. 309–316. [Online]. Available: https://doi.org/10.1007/ 978-0-387-32833-1 233

    2008

  34. [42]

    The large-sample distribution of the likelihood ratio for testing composite hypotheses,

    S. S. Wilks, “The large-sample distribution of the likelihood ratio for testing composite hypotheses,” The Annals of Mathematical Statistics, vol. 9, no. 1, pp. 60–62, 1938. [Online]. Available: http://www.jstor.org/stable/2957648

    1938

  35. [43]

    Goodfellow, Y

    I. Goodfellow, Y . Bengio, and A. Courville, Deep Learning . MIT Press, 2016, book in preparation for MIT Press. [Online]. Available: http://www.deeplearningbook.org

    2016

  36. [44]

    Statistics and machine learning toolbox,

    T. M. Inc., “Statistics and machine learning toolbox,” Natick, Massachusetts, United States, 2023. [Online]. Available: https: //www.mathworks.com/help/stats/index.html

    2023

  37. [2019]

    Available: https://www.mdpi.com/1099-4300/21/5/502

    [Online]. Available: https://www.mdpi.com/1099-4300/21/5/502

  38. [2024]

    Available: https://library.imaging.org/ei/articles/36/10/ IPAS-367

    [Online]. Available: https://library.imaging.org/ei/articles/36/10/ IPAS-367

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.