Pith. sign in

REVIEW 4 major objections 5 minor 61 references

FNIN: A Fourier Neural Operator-based Numerical Integration Network for Surface-form-gradients

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Learned Fourier operator reconstructs surfaces with sub-0.1 mm error

desk verdict Solid new application of FNO to normal integration with honest limitations, but the abstract and dataset-level SOTA claims are oversold. read the letter →

arxiv 2501.11876 v1 pith:3BVXP5NG submitted 2025-01-21 cs.CV

Jiaqi Leng , Yakun Ju , Yuanxu Duan , Jiangnan Zhang , Qingxuan Lv , Zuxuan Wu , Hao Fan This is my paper · ORCID
classification cs.CV
keywords surface-from-gradientsnormalintegrationFourierneuraloperatordiscontinuitydetectionweightedleastsquaresphotometricstereo3Dreconstruction
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 proposes FNIN, a two-stage learned framework for surface-from-gradients: recovering a 3D depth map from a given normal map. It claims that treating integration as learning a translation-invariant convolution kernel in Fourier space, via a Fourier neural operator, lets a network approximate the global solution operator of the underlying PDE. Coupled with an attention-detected discontinuity weight map and a one-step weighted least squares refinement, the method reports sub-millimeter accuracy (mean absolute error below 0.1 mm on some LUCES objects) with runtimes of one to two seconds. If correct, this shows a learning-based integrator can generalize to unseen real objects without fine-tuning and replace classical PDE solvers inside photometric-stereo pipelines.

What carries the argument

The central object is the translation-invariant kernel assumption in Eq. 6: the depth correction at a point depends only on the relative position to its neighbors and the difference in slopes, which turns the kernel integration into a convolution. Combined with the convolution theorem, the Fourier neural operator parameterizes the kernel directly in Fourier space as $R_\theta$, truncated at a maximum mode count $k_{\max}$, and learns it data-driven. A multi-resolution iterative network (initial net plus iterative net) applies this operator to the lifted residual $g - \nabla\hat{z}$, while an attention network (three-layer CNNs) produces a normalized weight map $\omega$ from one-sided difference features; a detail-weighted loss with $\gamma = 0.25$ suppresses discontinuities during training. Stage II then minimizes a weighted least squares objective over four difference directions with weights from $\omega$ and a small regularization $\lambda = 10^{-3}$ to the network's log-depth output, solved via conjugate gradient on the normal equation $A^T W A z = A^T W b$.

What would settle it

Probe the trained Stage I network with single-pixel gradient impulses placed at different image locations and compare the output depth patterns: translation invariance predicts identical shifted patterns, so any systematic position-dependent variation in the effective kernel would falsify the convolution assumption.

Watch

Extended reading notes

Core claim

The central claim is that surface-from-gradients can be posed as an operator-learning problem: after subtracting an approximate low-resolution depth, the remaining depth correction is computed by a Fourier neural operator whose kernel is assumed translation-invariant, making the integration a convolution in Fourier space. Discontinuities, which break the PDE, are detected by an attention network trained with a detail-weighted loss, and their jumps are resolved by solving a weighted least squares system once rather than by repeated optimization. The paper reports state-of-the-art mean absolute errors on LUCES (e.g., CUP below 0.01 mm, BELL 0.12 mm, BEAR 0.15 mm) and on DiLiGenT (BEAR 0.11 mm with the sigmoid-regularized variant FNIN-S), outperforming both classical solvers and the previous learned method Rec-Net. The key novelty is that the integration operator is learned in Fourier space from data, so integrability enforcement is not hand-crafted but adapted to the statistics of real normal maps.

Load-bearing premise

The entire Stage I rests on the assumption that the depth correction at a point depends only on the relative distance to its neighbors and the difference in slopes, not on the absolute position in the image; the paper justifies this with a one-dimensional two-point example and does not prove it for general 2D surfaces.

Editorial extensions

If this is right

  • Learned normal integration can act as a fast, differentiable post-processor for photometric stereo networks, replacing classical solvers in end-to-end pipelines.
  • The one-to-two-second runtime at 2048×1536 resolution makes high-resolution surface reconstruction practical where large linear solvers become infeasible or out-of-memory.
  • Because discontinuities are resolved in a single weighted least squares solve rather than repeated optimization, the decoupling reduces both computation time and numerical instability.
  • Generalization to unseen DiLiGenT objects without fine-tuning indicates the learned operator captures transferable structure of normal integration across objects and resolutions.

Reading between the lines

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

  • If the translation-invariance assumption holds only approximately, accuracy should degrade on surfaces with strong boundary influence or non-stationary curvature; a boundary-probe experiment or training on crops with varied boundary conditions would test this.
  • The two-stage decoupling suggests a template for other global-information inverse problems in vision where the solution operator is a smooth pseudo-differential map, such as depth from defocus or optical flow integration.
  • Replacing the supervised detail-weighted loss with a self-supervised integrability consistency loss could reduce reliance on synthetic training data, which the paper itself flags as a limitation for large discontinuities.
  • The same Fourier operator formulation might be extended to solve photometric stereo directly, treating the normal map as an intermediate output rather than the operator's input.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes FNIN, a two-stage learning framework for surface-from-gradients (SfG). Stage I uses a Fourier neural operator (FNO) inside a multi-resolution iterative ResNet-like architecture to approximate the integration operator in Fourier space, with a self-learning attention module and a detail-weighted loss to handle discontinuities. Stage II refines the reconstruction by solving a one-step weighted least squares problem with the Stage I depth as a prior and the attention map as weights. The method is trained on synthetic normal/depth pairs and evaluated without fine-tuning on real datasets: LUCES (14 objects, 2048x1536) and DiLiGenT (9 objects, 612x512). The paper reports lower MAE than many classical solvers and Rec-Net on several objects, shorter runtimes, and includes ablations on the FNO layers, truncated modes, attention, and discontinuity optimization, with code released on GitHub.

Significance. If the claims hold, FNIN would be a practically useful contribution: a learned normal integrator that is fast on high-resolution inputs, generalizes from synthetic training data to real objects without fine-tuning, and preserves discontinuities better than pure optimization baselines. The paper also brings neural-operator methods to a new inverse problem, which is a reasonable extension of FNO beyond PDE regression. The supplementary material strengthens the work by reporting five-trial means and standard deviations for the ablations, giving training details, and documenting limitations (surface islands, insufficient attention weights, nonlinear near-field bias). These elements make the empirical study substantially more reproducible than many vision papers. However, the central accuracy claims are currently overstated in the abstract and in the dataset-level SOTA statement, and the theoretical justification of the convolution kernel in Eq. 6 is not established for the general 2D bounded-domain problem.

major comments (4)
  1. [Abstract and Table 1] The abstract claims 'achieving errors of fewer than 0.1 mm on tested objects', but Table 1 shows that only two of the fourteen LUCES objects meet this threshold (CUP <0.01 mm and BOWL 0.08 mm for FNIN; BOWL 0.05 mm for FNIN-S). Most objects have MAEs between 0.12 mm and 10.50 mm. Please either restrict the claim to specific objects or replace it with a summary statistic that honestly reflects the full dataset.
  2. [Table 2 and Section 'Comparison with Previous Methods'] The claim of state-of-the-art accuracy on DiLiGenT is not supported by Table 2. Averaging the nine per-object MAEs gives 4.23 mm for FNIN-S and 4.62 mm for FNIN, whereas IPF achieves 4.20 mm; FNIN-S is worse than IPF on Buddha, Goblet, Harvest, and Reading. Since the supplementary Table 6 reports a standard deviation of only 0.0083 mm for FNIN on DiLiGenT, the 0.03 mm deficit is not explained by run-to-run variability. The text should either remove the dataset-level SOTA claim or support it with a paired statistical test across the nine objects.
  3. [Section 'Integration operator in spatial domain', Eq. (6)] The derivation of the translation-invariant convolution kernel from the one-dimensional two-point example in Fig. 4 does not extend to the general two-dimensional boundary-value problem in Eq. (5). For a bounded domain with Dirichlet boundary conditions, the Green's function depends on both x and y separately, not on the difference x-y, so Eq. 6 is not a 'perfect convolution' for the full problem. Please either prove the translation-invariance under the stated conditions, or revise the text to describe Eq. 6 as a modeling assumption whose validity is to be judged empirically. This is load-bearing because the entire FNO architecture in Eq. 7 is justified by this convolution.
  4. [Tables 1 and 2] The main empirical comparisons report single-run MAE values without error bars or repeated-trial statistics. Some of the claimed improvements over baselines are fractions of a millimeter (e.g., IPF 0.64 vs FNIN-S 0.49 on Pot 2), and the table does not indicate whether these differences are stable across training runs. The paper should report at least the mean and standard deviation over multiple seeds for the learned methods in the main tables, or present a statistical significance test, as already partially done in the supplementary Table 6 for ablations.
minor comments (5)
  1. [Section 'Stage II of FNIN', Eq. (13)] The displayed formula for W has unbalanced parentheses and a stray punctuation mark; it should be written as W = diag([wr/2, wl/2, wt/2, wb/2]^T) with a single closing bracket.
  2. [Section 'Operator approximation in Fourier space'] The text refers to 'solving Eq. 14' but Eq. 14 is not defined in the main paper; it appears only in the supplementary material. Please define the gradient equation in the main text or refer to the supplementary equation explicitly.
  3. [References] The citation of Rec-Net is inconsistent: the main text refers to 'Rec-Net (Lichy, Sengupta, and Jacobs 2022)' in Fig. 1 and elsewhere, while the reference list contains 'Lichy, D.; Wu, J.; Sengupta, S.; and Jacobs, D. W. 2021. Shape and Material Capture at Home' as the Rec-Net source, and a separate 2022 paper 'Fast Light-Weight Near-Field Photometric Stereo'. Please harmonize the citations.
  4. [Fig. 5 caption] The caption spells the dataset object as 'BRAR'; this should be 'BEAR'.
  5. [Section 'Stage II of FNIN'] The text says 'We directly solve the normal equation of Eq. 22', but Eq. 22 is in the supplementary material; the intended reference in the main text is Eq. 11 or Eq. 12.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: FNIN's two-stage pipeline is trained on synthetic data and tested on unseen real datasets; its core convolution/Fourier step is an explicit modeling assumption, not a restatement of the target.

full rationale

I traced the derivation chain. In Stage I, the paper transforms the SfG equation into Eq. 5 with a Dirichlet condition on z - z_hat, then introduces an explicit assumption about the integration kernel: 'our assumption is that Δy depends only on the distance x1 - x2 between the red point A and the blue point B, and the difference in slope k1 - k2.' This leads to the convolution form in Eq. 6 and the Fourier-space approximation in Eq. 7. That is a stated modeling ansatz, not a definitional identification of the target solution operator with the network output, so it is not circular. Stage II solves the weighted least squares system in Eq. 11/12 with the Stage-I depth zR appearing only as a soft prior (λ = 1e-3) and with attention-derived weights; the reported MAEs are computed on the final WLS solution against ground-truth meshes, so the prediction is not equal to a fitted input by construction. Training is on a synthetic dataset rendered by Lichy et al. 2021, while the headline results are evaluated on unseen LUCES and DiLiGenT objects without fine-tuning, making the central empirical claim externally falsifiable rather than circular. The paper does cite the authors' own prior work (Ju et al. 2020, 2022) for the attention mechanism and detail-weighted loss, but those citations support an architectural component, not the load-bearing accuracy claim, and the component is evaluated through ablations. The limitations in Section 9 (surface-island ambiguity, insufficient attention weights for large discontinuities, and the open problem of nonlinear equations) are honest caveats; they do not reveal a circular step. I found no place where an equation reduces to its own input by definition or where a fitted parameter is renamed as a prediction. The main weaknesses are empirical (the DiLiGenT dataset-level SOTA is arguable, and the '<0.1 mm' abstract claim applies to only a subset of LUCES objects), but those are correctness concerns, not circularity.

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

The central method rests on the translation-invariant kernel assumption (Eq. 6), the use of an upsampled lower-resolution depth as a boundary anchor, and the transferability of a synthetic training set to real datasets. No new physical entities are introduced. Several hyperparameters (γ, λ, number of Fourier layers/modes) are hand-chosen or ablated.

free parameters (6)
  • Fourier kernel parameters R_theta = learned on synthetic training set
    Stage I approximates the integration operator in Fourier space via FNO; the kernel is the main trainable component and is fit to 1.75k synthetic 512x512 samples.
  • Attention network parameters (f_theta_ae, f_theta_ar) = learned on synthetic training set
    These three-layer CNNs produce the discontinuity weight map omega used in Eq. 9 and in Stage II WLS; their outputs depend entirely on training data.
  • detail-weighted loss weight gamma = 0.25
    Hand-set hyperparameter in Eq. 9 that balances absolute depth error and normal error; the paper uses it to suppress discontinuities.
  • WLS prior weight lambda = 1e-3
    Fixed constant in Eq. 11/22 that controls how strongly the final surface follows the Stage I depth prior zR; chosen by hand.
  • Fourier modes truncated (kmax) = 16 (final)
    Truncation of the Fourier series in Eq. 7; Table 5 shows accuracy degrades with fewer modes, so this choice affects the central result.
  • Number of Fourier layers = 4 (final)
    Ablation Table 4 shows trade-off between accuracy and computation; 4 layers was selected as final.
assumptions (5)
  • domain assumption The solution operator of Eq. 5 can be approximated by a translation-invariant convolution kernel as in Eq. 6.
    Stated in 'Integration operator in spatial domain' via the 1D two-point example in Fig. 4. It is the basis for using FNO, but no derivation for general 2D bounded domains or non-smooth fields is given.
  • domain assumption The upsampled lower-resolution depth z_hat provides a valid anchor such that solving the residual with zero boundary in Eq. 5 gives the correct surface.
    The paper uses z_hat from the previous resolution and imposes z - z_hat = 0 on the boundary; if z_hat has a global scale or offset error, Eq. 5 cannot recover the absolute depth. The authors acknowledge this with the 'surface islands' limitation.
  • domain assumption The synthetic training set (14 statue objects) is representative enough for the learned operator to generalize to the real LUCES and DiLiGenT objects without fine-tuning.
    The paper trains only on Lichy et al.'s synthetic data and evaluates directly on real datasets; no domain adaptation or test-time training is used.
  • domain assumption Discontinuities can be detected from a point-to-plane distance map computed from the current depth estimate, via the attention network.
    The attention map is trained with the detail-weighted loss using synthetic ground truth; the same map is then trusted to weight the WLS solve on real data.
  • standard math Convolution theorem and FNO spectral convolution (Eq. 7) are valid for approximating the kernel integral.
    Standard Fourier analysis used in Li et al. 2020a; the paper truncates the series at a finite number of modes.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FNIN: A Fourier Neural Operator-based Numerical Integration Network for Surface-form-gradients." pith.science (2026). https://pith.science/paper/3BVXP5NG

@misc{pith2026250111876,
  author       = {Pith},
  title        = {Pith review of: FNIN: A Fourier Neural Operator-based Numerical Integration Network for Surface-form-gradients},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3BVXP5NG}},
  note         = {Machine review of arXiv:2501.11876}
}
read the original abstract

Surface-from-gradients (SfG) aims to recover a three-dimensional (3D) surface from its gradients. Traditional methods encounter significant challenges in achieving high accuracy and handling high-resolution inputs, particularly facing the complex nature of discontinuities and the inefficiencies associated with large-scale linear solvers. Although recent advances in deep learning, such as photometric stereo, have enhanced normal estimation accuracy, they do not fully address the intricacies of gradient-based surface reconstruction. To overcome these limitations, we propose a Fourier neural operator-based Numerical Integration Network (FNIN) within a two-stage optimization framework. In the first stage, our approach employs an iterative architecture for numerical integration, harnessing an advanced Fourier neural operator to approximate the solution operator in Fourier space. Additionally, a self-learning attention mechanism is incorporated to effectively detect and handle discontinuities. In the second stage, we refine the surface reconstruction by formulating a weighted least squares problem, addressing the identified discontinuities rationally. Extensive experiments demonstrate that our method achieves significant improvements in both accuracy and efficiency compared to current state-of-the-art solvers. This is particularly evident in handling high-resolution images with complex data, achieving errors of fewer than 0.1 mm on tested objects.

Figures

Figures reproduced from arXiv: 2501.11876 by the authors.

Figure 1
Figure 1. Reconstruction of the “HOUSE” in the LUCES [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The Fourier neural operator-based numerical integration network operates within a two-stage framework. In stage [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Integration is performed at an example resolution. Gradient [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: A simple example of (a) our assumption and [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Qualitative results on “BRAR”, “COW” and [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Qualitative results on “BELL”, “BUDDHA” and “HIPPO” objects from LUCES (Mecca et al. 2021) Dataset. [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Ablation study on the “CUP” object from the [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Visual results on our dataset. Input normal estimated by classic PS (Woodham 1980), mesh estimated by BiNI (Cao [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Image capture device and one pair of calibration [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]
Figure 12
Figure 12. Figure 12: Near-field light bias [PITH_FULL_IMAGE:figures/full_fig_p014_12.png]
Figure 11
Figure 11. Figure 11: Insufficient weight [PITH_FULL_IMAGE:figures/full_fig_p014_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

61 extracted references · 53 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Badri, H.; Yahia, H.; and Aboutajdine, D. 2014. Robust Surface Reconstruction via Triple Sparsity. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2291--2298

    work page 2014

  4. [4]

    P.; Taubman, D

    Bradshaw, A. P.; Taubman, D. S.; Todd, M. J.; Magnussen, J. S.; and Halmagyi, G. M. 2013. Augmented Active Surface Model for the Recovery of Small Structures in CT. IEEE Transactions on Image Processing, 22(11): 4394--4406

    work page 2013

  5. [5]

    Cao, S. 2021. Choose a Transformer: Fourier or Galerkin. In Proceedings of the Advances in Neural Information Processing Systems, volume 34, 24924--24940. Curran Associates, Inc

    work page 2021

  6. [6]

    Cao, X.; Santo, H.; Shi, B.; Okura, F.; and Matsushita, Y. 2022. Bilateral Normal Integration. In Proceedings of the European Conference on Computer Vision

    work page 2022

  7. [7]

    Cao, X.; Shi, B.; Okura, F.; and Matsushita, Y. 2021. Normal Integration via Inverse Plane Fitting with Minimum Point-to-Plane Distance. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2382--2391

    work page 2021

  8. [8]

    K.; and Zhao, X.-L

    Chen, C.; Ng, M. K.; and Zhao, X.-L. 2015. Alternating Direction Method of Multipliers for Nonlinear Image Restoration Problems. IEEE Transactions on Image Processing, 24(1): 33--43

    work page 2015

Show all 61 references
  1. [9]

    Chen, G.; Han, K.; and Wong, K.-Y. K. 2018. PS-FCN: A Flexible Learning Framework for Photometric Stereo. In Proceedings of the European Conference on Computer Vision, 3--19

    2018

  2. [10]

    Fan, L.; Huang, W.; Gan, C.; Ermon, S.; Gong, B.; and Huang, J. 2018. End-to-End Learning of Motion Representation for Video Understanding. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 6016--6025

    2018

  3. [11]

    Fang, X.; Easwaran, A.; Genest, B.; and Suganthan, P. N. 2024. Your Data Is Not Perfect: Towards Cross-Domain Out-of-Distribution Detection in Class-Imbalanced Data. ESWA

    2024

  4. [12]

    Frankot, R.; and Chellappa, R. 1988. A Method for Enforcing Integrability in Shape from Shading Algorithms. IEEE Transactions on Pattern Analysis and Machine Intelligence, 10(4): 439--451

    1988

  5. [13]

    Harker, M.; and O’Leary, P. 2014. Regularized Reconstruction of a Surface from its Measured Gradient Field . Journal of Mathematical Imaging and Vision, 51(1): 46--70

    2014

  6. [14]

    R.; and Stiefel, E

    Hestenes, M. R.; and Stiefel, E. 1952. Methods of Conjugate Gradients for Solving Linear Systems. Journal of research of the National Bureau of Standards, 49: 409--435

    1952

  7. [15]

    K.; and Brooks, M

    Horn, B. K.; and Brooks, M. J. 1986. The Variational Approach to Shape from Shading. Computer Vision, Graphics, and Image Processing, 33(2): 174--208

    1986

  8. [16]

    Hsieh, J.-W.; Liao, H.-Y.; Ko, M.-T.; and Fan, K.-C. 1994. Wavelet-based Shape from Shading. In Proceedings of International Conference on Image Processing, volume 2, 125--129

    1994

  9. [17]

    Ikehata, S. 2018. CNN-PS: CNN-Based Photometric Stereo for General Non-convex Surfaces . In Proceedings of the European Conference on Computer Vision, 3--19

    2018

  10. [18]

    H.; McCann, M

    Jin, K. H.; McCann, M. T.; Froustey, E.; and Unser, M. 2017. Deep Convolutional Neural Network for Inverse Problems in Imaging. IEEE Transactions on Image Processing, 26(9): 4509--4522

    2017

  11. [19]

    Ju, Y.; Lam, K.-M.; Chen, Y.; Qi, L.; and Dong, J. 2020. Pay Attention to Devils: A Photometric Stereo Network for Better Details. In Proceedings of the International Joint Conference on Artificial Intelligence, 694--700

    2020

  12. [20]

    Ju, Y.; Lam, K.-M.; Xiao, J.; Zhang, C.; Yang, C.; and Dong, J. 2023. Efficient feature fusion for learning-based photometric stereo. In ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 1--5. IEEE

    2023

  13. [21]

    Ju, Y.; Lam, K.-M.; Xie, W.; Zhou, H.; Dong, J.; and Shi, B. 2024. Deep Learning Methods for Calibrated Photometric Stereo and Beyond. IEEE Transactions on Pattern Analysis and Machine Intelligence

    2024

  14. [22]

    Ju, Y.; Shi, B.; Jian, M.; Qi, L.; Dong, J.; and Lam, K.-M. 2022. Normattention-psn: A High-frequency Region Enhanced Photometric Stereo Network with Normalized Attention. International Journal of Computer Vision, 130(12): 3014--3034

    2022

  15. [23]

    Karaçalı, B.; and Snyder, W. 2003. Reconstructing discontinuous surfaces from a given gradient field using partial integrability. Computer Vision and Image Understanding, 92(1): 78--111

    2003

  16. [24]

    Kim, H.; Jung, Y.; and Lee, S. 2024. Discontinuity-preserving Normal Integration with Auxiliary Edges. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 11915--11923

    2024

  17. [25]

    Kossaifi, J.; Kovachki, N.; Azizzadenesheli, K.; and Anandkumar, A. 2023. Multi-grid Tensorized Fourier Neural Operator for High-resolution PDEs. arXiv:2310.00120

    2023 arXiv

  18. [26]

    Kovesi, P. 2005. Shapelets Correlated with Surface Normals Produce Surfaces. In Proceedings of the IEEE International Conference on Computer Vision, volume 2, 994--1001

    2005

  19. [27]

    Li, Z.; Kovachki, N.; Azizzadenesheli, K.; Liu, B.; Bhattacharya, K.; Stuart, A.; and Anandkumar, A. 2020 a . Fourier Neural Operator for Parametric Partial Differential Equations. arXiv:2010.08895

    2020 arXiv

  20. [28]

    Li, Z.; Kovachki, N.; Azizzadenesheli, K.; Liu, B.; Bhattacharya, K.; Stuart, A.; and Anandkumar, A. 2020 b . Neural Operator: Graph Kernel Network for Partial Differential Equations. arXiv:2003.03485

    2020 arXiv

  21. [29]

    Lichy, D.; Sengupta, S.; and Jacobs, D. W. 2022. Fast Light-Weight Near-Field Photometric Stereo. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 12602--12611

    2022

  22. [30]

    Lichy, D.; Wu, J.; Sengupta, S.; and Jacobs, D. W. 2021. Shape and Material Capture at Home. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 6119--6129

    2021

  23. [31]

    Liu, N.; Jafarzadeh, S.; and Yu, Y. 2023. Domain Agnostic Fourier Neural Operators. In Proceedings of the Advances in Neural Information Processing Systems, volume 36, 47438--47450. Curran Associates, Inc

    2023

  24. [32]

    Logothetis, F.; Mecca, R.; and Cipolla, R. 2017. Semi-calibrated near field photometric stereo. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 941--950

    2017

  25. [33]

    Loshchilov, I.; and Hutter, F. 2017. Decoupled Weight Decay Regularization. arXiv:1711.05101

    2017 arXiv

  26. [34]

    Lu, L.; Jin, P.; Pang, G.; Zhang, Z.; and Karniadakis, G. E. 2021. Learning Nonlinear Operators via DeepONet Based on the Universal Approximation Theorem of Operators . Nature Machine Intelligence, 3(3): 218--229

    2021

  27. [35]

    Luo, W.; Li, Y.; Urtasun, R.; and Zemel, R. 2016. Understanding the Effective Receptive Field in Deep Convolutional Neural Networks. In Proceedings of the Advances in Neural Information Processing Systems, 4905–4913

    2016

  28. [36]

    Mecca, R.; Logothetis, F.; Budvytis, I.; and Cipolla, R. 2021. Luces: A dataset for Near-field Point Light Source Photometric Stereo. arXiv:2104.13135

    2021 arXiv

  29. [37]

    Ming, Y.; Meng, X.; Fan, C.; and Yu, H. 2021. Deep Learning for Monocular Depth Estimation: A Review. Neurocomputing, 438: 14--33

    2021

  30. [38]

    W.; and Mayers, D

    Morton, K. W.; and Mayers, D. F. 2005. Numerical Solution of Partial Differential Equations: An Introduction. Cambridge university press

    2005

  31. [39]

    Ouyang, H.; Wang, T.; and Chen, Q. 2022. Restorable Image Operators with Quasi-Invertible Networks. Proceedings of the AAAI Conference on Artificial Intelligence, 36(2): 2008--2016

    2022

  32. [40]

    Paszke, A.; Gross, S.; Chintala, S.; Chanan, G.; Yang, E.; DeVito, Z.; Lin, Z.; Desmaison, A.; Antiga, L.; and Lerer, A. 2017. Automatic differentiation in pytorch

    2017

  33. [41]

    Patil, V.; Sakaridis, C.; Liniger, A.; and Van Gool, L. 2022. P3Depth: Monocular Depth Estimation with a Piecewise Planarity Prior. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 1600--1611

    2022

  34. [42]

    K.; and Chellappa, R

    Peng, C.; Liao, H.; Wong, G.; Luo, J.; Zhou, S. K.; and Chellappa, R. 2021. XraySyn: Realistic View Synthesis From a Single Radiograph Through CT Priors. Proceedings of the AAAI Conference on Artificial Intelligence, 35(1): 436--444

    2021

  35. [43]

    Qu \'e au, Y.; and Durou, J.-D. 2015. Edge-Preserving Integration of a Normal Field: Weighted Least-Squares, TV and L^1 Approaches. In Proceedings of Scale Space and Variational Methods in Computer Vision, 576--588. Springer

    2015

  36. [44]

    Quéau, Y.; Durou, J.-D.; and Aujol, J.-F. 2017 a . Normal Integration: A survey . Journal of Mathematical Imaging and Vision, 60(4): 576--593

    2017

  37. [45]

    Quéau, Y.; Durou, J.-D.; and Aujol, J.-F. 2017 b . Variational Methods for Normal Integration . Journal of Mathematical Imaging and Vision, 60(4): 609--632

    2017

  38. [46]

    Raonic, B.; Molinaro, R.; De Ryck, T.; Rohner, T.; Bartolucci, F.; Alaifari, R.; Mishra, S.; and de B\' e zenac, E. 2023. Convolutional Neural Operators for Robust and Accurate Learning of PDEs. In Proceedings of the Advances in Neural Information Processing Systems, volume 36...

    2023

  39. [47]

    Shi, B.; Mo, Z.; Wu, Z.; Duan, D.; Yeung, S.-K.; and Tan, P. 2019 a . A Benchmark Dataset and Evaluation for Non-Lambertian and Uncalibrated Photometric Stereo. IEEE Transactions on Pattern Analysis and Machine Intelligence, 41(02): 271--284

    2019

  40. [48]

    Shi, B.; Mo, Z.; Wu, Z.; Duan, D.; Yeung, S.-K.; and Tan, P. 2019 b . A Benchmark Dataset and Evaluation for Non-Lambertian and Uncalibrated Photometric Stereo. IEEE Transactions on Pattern Analysis and Machine Intelligence, 41(2): 271--284

    2019

  41. [49]

    Simchony, T.; Chellappa, R.; and Shao, M. 1990. Direct Analytical Methods for Solving Poisson Equations in Computer Vision Problems. IEEE Transactions on Pattern Analysis and Machine Intelligence, 12(5): 435--446

    1990

  42. [50]

    Sofiiuk, K.; Petrov, I.; Barinova, O.; and Konushin, A. 2020. F-BRS: Rethinking Backpropagating Refinement for Interactive Segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 8620--8629

    2020

  43. [51]

    Wang, M.; Xie, W.; and Cui, M. 2020. Surface Reconstruction with Unconnected Normal Maps: An Efficient Mesh-based Approach. In Proceedings of the ACM International Conference on Multimedia, 2617–2625. Association for Computing Machinery

    2020

  44. [52]

    Woodham, R. J. 1980. Photometric Method for Determining Surface Orientation from Multiple Images . Optical Engineering, 19(1): 191139

    1980

  45. [53]

    Wu, Z.; and Li, L. 1988. A Line Integration Based Method for Depth Recovery from Surface Normals. In Proceedings of International Conference on Pattern Recognition, 591--595

    1988

  46. [54]

    Xie, W.; Wang, M.; Wei, M.; Jiang, J.; and Qin, J. 2019. Surface Reconstruction From Normals: A Robust DGP-Based Discontinuity Preservation Approach. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 5323--5331

    2019

  47. [55]

    C.; and Chung, R

    Xie, W.; Zhang, Y.; Wang, C. C.; and Chung, R. C.-K. 2014. Surface-from-Gradients: An Approach Based on Discrete Geometry Processing. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2203--2210

    2014

  48. [56]

    Xu, G.; Wang, X.; Ding, X.; and Yang, X. 2023. Iterative Geometry Encoding Volume for Stereo Matching. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 21919--21928

    2023

  49. [57]

    Yang, Z.; Yao, M.; Huang, J.; Zhou, M.; and Zhao, F. 2022. SIR-Former: Stereo Image Restoration Using Transformer. In Proceedings of the 30th ACM International Conference on Multimedia, 6377--6385

    2022

  50. [58]

    Zhang, W.; Zhang, W.; and Zhang, Y. 2020. GeoLayout: Geometry Driven Room Layout Estimation Based on Depth Maps of Planes. In Proceedings of the European Conference on Computer Vision, 632--648

    2020

  51. [59]

    Zhang, Z. 2000. A Flexible New Technique for Camera Calibration. IEEE Transactions on Pattern Analysis and Machine Intelligence, 22(11): 1330--1334

    2000

  52. [60]

    Zhu, D.; and Smith, W. A. P. 2020. Least Squares Surface Reconstruction on Arbitrary Domains. In Proceedings of the European Conference on Computer Vision, 530--545. Springer

    2020

  53. [61]

    Zisserman, A.; and Wiles, O. 2017. SilNet: Single-and Multi-View Reconstruction by Learning from Silhouettes. In Proceedings of the British Machine Vision Conference. British Machine Vision Association and Society for Pattern Recognition

    2017

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.