Pith. sign in

REVIEW 5 major objections 4 minor 34 references

Cracks in concrete

T0 review · 5 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read The paper argues a Riesz-transform network can segment cracks in 3D concrete CT images as well as a fine-tuned 3D U-Net, with 7,153 trainable parameters.

desk verdict A promising 3D RieszNet for crack segmentation, but the central comparison lacks a training protocol and is not reproducible as written. read the letter →

arxiv 2501.18376 v1 pith:JFNPGMZT submitted 2025-01-30 cs.CV eess.IVstat.AP

Tin Barisin , Christian Jung , Anna Nowacka , Claudia Redenbach , Katja Schladitz This is my paper · ORCID
classification cs.CVeess.IVstat.AP
keywords cracksegmentationconcreteRiesztransformscaleequivariance3DU-Netcomputedtomographysynthetictrainingdatadeeplearning
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 argues that a neural network whose encoder uses Riesz transforms instead of learned convolutions can segment cracks in 3D computed-tomography images of concrete as reliably as a fine-tuned 3D U-Net, while needing only about 7,153 trainable parameters. The authors show that cracks have strongly varying thickness, so the segmentation method should be invariant to scale changes, and they build that invariance into the RieszNet architecture directly. On 15 synthetic images with multiscale Voronoi-model cracks, RieszNet reaches a mean F1 of 0.872 at zero tolerance, close to the fine-tuned U-Net's 0.897, and it keeps a recall of 0.805 on the hardest image where the U-Net drops to 0.046. The paper also details how to generate semi-synthetic training data by embedding simulated crack surfaces into real CT images of crack-free concrete.

What carries the argument

The load-bearing object is the Riesz transform, defined in d dimensions by a singular integral, which commutes with scaling: $R_j(L_a f)=L_a(R_j f)$. A Riesz layer forms each output channel as a learned linear combination of first- and second-order Riesz transforms of the input channels. Stacked with batch normalization and ReLU activation, these layers replace the convolutional encoder, so the network inherits approximate scale equivariance without image pyramids or fine-tuning on multiscale data. The other machinery is the synthetic-data pipeline: cracks modeled as minimum-weight surfaces in 3D Voronoi diagrams, with thickness varied by adaptive dilation, embedded into real CT images of crack-free concrete.

What would settle it

Segment a set of real CT concrete volumes with manually annotated crack ground truth using the same trained RieszNet and fine-tuned U-Net, and compare their F1 at zero tolerance: if RieszNet's recall or F1 falls below the U-Net baseline on real cracks, the claim that it competes with the fine-tuned U-Net on realistic data is refuted.

Watch

Extended reading notes

Core claim

The central claim is that a network whose layers combine first- and second-order Riesz transforms instead of learned convolutions is naturally scale-equivariant and therefore suited to segmenting cracks whose thickness varies from 1 to 20 voxels. On a test set of 15 synthetic multiscale-crack images, the three-layer RieszNet achieves a mean F1 of 0.872 at zero tolerance, close to the fine-tuned 3D U-Net's 0.897 and slightly above the multiscale U-Net's 0.869, while using only 7,153 trainable parameters. On the hardest image, its recall remains 0.805 where the fine-tuned U-Net drops to 0.046. The paper argues this makes RieszNet a promising alternative to CNN-based segmentation whenever scale invariance is a natural requirement and annotated training data are scarce.

Load-bearing premise

The quantitative comparison assumes that cracks generated by the authors' Voronoi surface model, embedded in real CT backgrounds, behave like real cracks; if these synthetic images do not predict real-crack performance, the comparison only holds for synthetic data.

Editorial extensions

If this is right

  • RieszNet can segment cracks of widely varying thickness without dedicated multiscale processing or fine-tuning on multiscale data.
  • A 7,153-parameter network can reach U-Net-level performance, which is valuable when annotated training data are scarce.
  • The synthetic Voronoi-based crack model with adaptive dilation produces training data on which deep networks generalize to multiscale cracks, at least within the synthetic domain.
  • The class-imbalance weighting scheme used for the U-Net training transfers to the RieszNet training setup.

Reading between the lines

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

  • If RieszNet's scale equivariance holds as strongly as the synthetic results suggest, the same architecture could segment other thin, dark lower-dimensional structures in 3D images, such as fibers, platelets, or biological sheets, without retuning the scale range.
  • A direct test of the paper's central claim would be to train RieszNet on synthetic cracks and evaluate on a set of real CT cracks with voxel-level annotations; the paper only shows real-crack images without quantitative ground truth, so this transfer remains the open question it flags.
  • The tiny parameter count suggests RieszNet may be suitable for 4D in-situ crack-evolution analysis, where many time steps must be processed and manual annotation is infeasible.
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

5 major / 4 minor

Summary. The paper addresses 3D segmentation of cracks in concrete CT images. It describes a semi-synthetic data-generation pipeline based on fractional Brownian surfaces and Voronoi minimum-weight surfaces with adaptive dilation, trains 3D U-Net baselines, and introduces RieszNet, a scale-equivariant network built from Riesz transforms with 7,153 trainable parameters. The central empirical claim, based on 15 synthetic multiscale-crack test images in Table 1, is that RieszNet competes with a fine-tuned 3D U-Net (mean F1 0.872 vs 0.897 at tolerance 0) and is substantially more robust on the hardest image (recall 0.805 vs 0.046). The paper also shows qualitative segmentation results on real concrete and on fiber-reinforced concrete.

Significance. If the RieszNet result is reproducible, it is significant: the parameter count is three orders of magnitude smaller than the 3D U-Net, and the better worst-case recall on the difficult synthetic image is a practically relevant advantage in a setting where annotated training data are scarce. The Voronoi-based crack generator with locally varying thickness is a useful contribution that goes beyond fixed-width fractional Brownian surfaces. The paper is also commendably honest about limits: Section 7 explicitly states that generalization to unseen concrete formulations is an open challenge, and the real-crack results in Table 3 are only qualitative. However, the empirical support for the central competitiveness claim is thin and incompletely documented, so the significance currently rests on a promising but not yet fully verified result.

major comments (5)
  1. [Section 5.4 / Table 1] The RieszNet evaluation is not reproducible because the training protocol is never given. Section 5.3 describes the architecture and parameter count, but Section 5.4 only states that three approaches were tested on 15 images. The paper does not specify the training set size, the train/validation/test split, the patch size, the loss function, the optimizer, the learning rate schedule, the number of epochs, or the threshold used to binarize the sigmoid output. This contrasts with the detailed protocols provided for the 3D U-Net baselines in Sections 4.3 and 5.2. Without this information, the Table 1 comparison could be confounded by unstated training choices or by threshold selection on the test images, and the central claim that RieszNet 'can compete with 3D U-Net' cannot be verified.
  2. [Section 5.4 / Table 1] The evidence for the central claim consists of only 15 synthetic test images, with no error bars, confidence intervals, or significance tests. At tolerance 0 the mean F1 difference between RieszNet (0.872) and the fine-tuned 3D U-Net (0.897) is small relative to the per-image variation (RieszNet's minimum F1 is 0.773 on image 10, while the fine-tuned U-Net's minimum is 0.088 on image 7). The claim of competitiveness would be much better supported by reporting per-image results, bootstrap intervals, or a paired test, or by explicitly labeling the result as a case study on a limited benchmark.
  3. [Section 5.3] The paper states that the Riesz network imposes scale invariance, but the actual layer composition is batch normalization, a Riesz layer, and ReLU. Batch normalization and ReLU do not commute with scaling, so the network is at best approximately scale-invariant. The paper should either soften the claim to approximate scale invariance or provide empirical evidence (e.g., segmentation performance across crack widths from 1 to 20 voxels) showing that the intended invariance is preserved in practice. This is load-bearing because the motivation for RieszNet, and the expected improvement over U-Net on multiscale cracks, rests on this property.
  4. [Section 7 / Table 3] The paper explicitly acknowledges that generalization to unseen concrete formulations is an open challenge, and the real-crack results in Table 3 are shown only as images with no quantitative ground truth. The synthetic test data come from the authors' own Voronoi-based generator, and the RieszNet architecture was also developed by the same group; this creates a validity burden for the claim that the synthetic benchmark predicts real performance. The paper should either add quantitative real-image evaluation with ground truth, or clearly state that the real-crack demonstrations are qualitative and that the competitiveness claim is restricted to the synthetic benchmark.
  5. [Section 4.2 / Section 5.4] The number of Voronoi-based multiscale crack images is not stated explicitly. Section 5.2 says five such images are used for fine-tuning and Section 5.4 refers to 'the remaining 15 images', which implies a total of 20, but Section 4.2 only gives the count of 60 for the fixed-width crack images. This omission is part of the missing evaluation protocol and should be corrected so that the reader can understand the full data split.
minor comments (4)
  1. [Section 4.1] The sentence defining p0 and p1 says 'the proportion of background (class 0) and crack pixels (class 0)' — the crack class should be class 1, not class 0.
  2. [Figure 3 caption] The caption refers to a 'Bernoulli random walk with parameter' without specifying the parameter; it should say 'with parameter p' and give its value or range.
  3. [Section 5.2] The fine-tuning description says the U-Net is 'further train[ed] on five of these multiscale crack images', but it is not stated how many epochs, what learning rate, or whether the same augmentation as in Section 4.3 was used; this makes the fine-tuning baseline less reproducible than the original U-Net training.
  4. [General] There is no statement about code or data availability; given that the paper introduces a new network and a new synthetic data generator, a statement about releasing code and the generated test images would substantially improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the RieszNet comparison is an empirical benchmark on held-out generated test images, not a derivation that folds its inputs into its outputs.

full rationale

No step in the paper's argument reduces by construction to its own inputs. The central quantitative claim in Table 1 is an empirical comparison on 15 held-out multiscale Voronoi-crack images; the ground-truth labels are produced by the Section 4.2 construction (Dijkstra paths on weighted Voronoi facets, integer-program minimum-weight surfaces, and adaptive dilation), and there is no indication that RieszNet parameters were fitted to those 15 test images. The Riesz layer is defined directly in Eqs. (1)-(2), and the scale-equivariance identity R_j(L_a(f))(x)=L_a(R_j(f))(x) is quoted from the classical Riesz-transform literature (Unser and Van De Ville [32]), so the architecture does not rest solely on the self-citation to Barisin et al. [4]. The fine-tuned 3D U-Net is trained on five multiscale images from the same generator and tested on the remaining 15; that is a normal train/test split rather than a fitted-input-called-prediction pattern. The paper's own stated limitations, such as in Section 7 that 'generalization to unseen concrete formulations' remains an open challenge and in Section 5.4 the omission of the RieszNet training protocol, are reproducibility and external-validity concerns, not circularity. No equation is equivalent to another by definition, and no load-bearing conclusion is forced by a self-citation chain.

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

The central claim depends on the synthetic data generator (both the Voronoi model and the fractional Brownian model), the Riesz transform scale equivariance, and the unverified assumption that batch norm and ReLU keep the network effectively scale invariant. No new physical entities are introduced.

free parameters (4)
  • Hurst index H for fractional Brownian crack surfaces = 0.97 (in Figure 3 example)
    Controls the roughness of synthetic cracks; value chosen by hand, not fitted to outcome, but shapes the training distribution.
  • Bernoulli parameter p for adaptive dilation = not specified
    Controls the local thickness variation of Voronoi-based synthetic cracks in Section 4.2; value not reported, affecting the realism and difficulty of the test data.
  • Evaluation tolerance = 0 and 1 voxel
    Results are reported with tolerance 0 and 1; this is a hand-chosen evaluation parameter that changes the scores.
  • RieszNet architecture channel widths = (1,16,16,32,1)
    Network depth and width chosen by the authors; 7,153 trainable parameters are learned, but the architecture itself is a free design choice.
assumptions (4)
  • standard math Scale equivariance of the Riesz transform
    The Riesz transform commutes with dilation, which is the mathematical basis for the RieszNet's scale invariance (Section 5.3).
  • ad hoc to paper Batch normalization and ReLU preserve scale invariance sufficiently for the network to generalize across crack widths
    The paper assumes that adding batch norm and ReLU to Riesz layers does not destroy the scale-equivariance property in a way that matters for segmentation; no formal analysis is given.
  • domain assumption Voronoi minimum-weight surfaces adequately model real crack geometry
    The quantitative test set uses synthetic cracks from the Voronoi model (Sections 4.2 and 5.4); the paper's own discussion in Section 7 flags generalization to real cracks as open.
  • domain assumption Cracks are air-filled and have gray values similar to pores
    The synthetic training data assigns crack gray values based on pore gray values (Section 4.2); this assumes real cracks have the same contrast behavior in CT.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Cracks in concrete." pith.science (2026). https://pith.science/paper/JFNPGMZT

@misc{pith2026250118376,
  author       = {Pith},
  title        = {Pith review of: Cracks in concrete},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JFNPGMZT}},
  note         = {Machine review of arXiv:2501.18376}
}
read the original abstract

Finding and properly segmenting cracks in images of concrete is a challenging task. Cracks are thin and rough and being air filled do yield a very weak contrast in 3D images obtained by computed tomography. Enhancing and segmenting dark lower-dimensional structures is already demanding. The heterogeneous concrete matrix and the size of the images further increase the complexity. ML methods have proven to solve difficult segmentation problems when trained on enough and well annotated data. However, so far, there is not much 3D image data of cracks available at all, let alone annotated. Interactive annotation is error-prone as humans can easily tell cats from dogs or roads without from roads with cars but have a hard time deciding whether a thin and dark structure seen in a 2D slice continues in the next one. Training networks by synthetic, simulated images is an elegant way out, bears however its own challenges. In this contribution, we describe how to generate semi-synthetic image data to train CNN like the well known 3D U-Net or random forests for segmenting cracks in 3D images of concrete. The thickness of real cracks varies widely, both, within one crack as well as from crack to crack in the same sample. The segmentation method should therefore be invariant with respect to scale changes. We introduce the so-called RieszNet, designed for exactly this purpose. Finally, we discuss how to generalize the ML crack segmentation methods to other concrete types.

Figures

Figures reproduced from arXiv: 2501.18376 by the authors.

Figure 1
Figure 1. Example of a crack in concrete as they appear in optical images. Small cutout [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Examples of cracks in concrete as they appear in CT images. 2D slices from [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Examples of simulated cracks in 256 × 256 × 256 voxel images. Left: two fractional Brownian surfaces, both widened to constant thickness 3 pixels and with Hurst index 0.97. Right: a minimal surface from a spatial Voronoi tessellation, width varying according to a Bernoulli random walk with parameter. Top: volume render￾ings. Bottom: 2D slices of the 3D images of the cracks superimposed on CT images of high performan… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Examples of segmentation results for fiber reinforced concrete. The 3D U [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 26 canonical work pages

  1. [4]

    Journal of Mathematical Imaging and Vision 66, 246–270 (2024)

    Barisin, T., Redenbach, C., Schladitz, K.: Riesz networks: scale invariant neural networks in single forward pass. Journal of Mathematical Imaging and Vision 66, 246–270 (2024). DOI 10.1007/s10851-024-01171-4

    work page doi:10.1007/s10851-024-01171-4 2024

  2. [1]

    In: Eighth International Symposium on SPRAYED CONCRETE - Modern Use of Wet Mix Sprayed Concrete for Underground Support, pp

    Ansell, A., Nordstr ¨om, E., Guarin, A.: Laboratory investigation of steel fibre reinforced sprayed concrete using a computed tomography method. In: Eighth International Symposium on SPRAYED CONCRETE - Modern Use of Wet Mix Sprayed Concrete for Underground Support, pp. 24–38 (2018)

    work page 2018

  3. [2]

    Badrinarayanan, V ., Kendall, A., Cipolla, R.: SegNet: A deep convolutional encoder-decoder architecture for image segmentation. IEEE T. Pattern Anal. 39(12), 2481–2495 (2017). https://doi.org/10.1109/TPAMI.2016.2644615

    arXiv 2017

  4. [3]

    Pattern Recognition 129, 108747 (2022)

    Barisin, T., Jung, C., M ¨usebeck, F., Redenbach, C., Schladitz, K.: Methods for seg- menting cracks in 3d images of concrete: A comparison based on semi-synthetic im- ages. Pattern Recognition 129, 108747 (2022). DOI https://doi.org/10.1016/j.patcog.2022. 108747. URL https://www.sciencedirect.com/science/article/pii/ S003132032200228X

    doi:10.1016/j.patcog.2022 2022

  5. [5]

    In: Proceedings of the 9th Conference on Industrial Computed Tomography

    Bellon, C., Burger, K., Gollwitzer, C.: Virtual ct acquisition and reconstruction of complex and noisy scanning trajectories in aRTist. In: Proceedings of the 9th Conference on Industrial Computed Tomography. Padova, Italy (2019)

    work page 2019

  6. [6]

    In: DGZfP Jahrestagung 2021 (2021)

    Binder, F., Bellon, C., Wohlgemuth, F., Hausotte, T.: Ein praxisnaher Leitfaden f ¨ur com- putertomografische Untersuchungen mit der radiografischen Simulationsumgebung aRTist. In: DGZfP Jahrestagung 2021 (2021). URL https://jahrestagung.dgzfp.de/ portals/jt2021/bb176/inhalt/p4.pdf

    work page 2021

  7. [7]

    https: //de.mathworks.com/matlabcentral/fileexchange/ 38945-fractional-brownian-field-or-surface-generator

    Botev, Z.: Fractional Brownian field or surface generator. https: //de.mathworks.com/matlabcentral/fileexchange/ 38945-fractional-brownian-field-or-surface-generator . (accessed 02 February 2022)

    work page 2022

  8. [8]

    In: Lect

    C ¸ ic ¸ek,¨O., Abdulkadir, A., Lienkamp, S.S., Brox, T., Ronneberger, O.: 3D U-Net: Learning dense volumetric segmentation from sparse annotation. In: Lect. Notes Comput. Sc., pp. 424– 432 (2016). https://doi.org/10.1007/978-3-319-46723-8_49

    doi:10.1007/978-3-319-46723-8_49 2016

Show all 34 references
  1. [9]

    Chambon, S., Moliard, J.M.: Automatic Road Pavement Assessment with Image Processing: Review and Comparison. Int. J. Geoph. 2011, 1–20 (2011). https://doi.org/10. 1155/2011/989354

    2011

  2. [10]

    Comput.-Aided Civ

    Chun, P., Izumi, S., Yamane, T.: Automatic detection method of cracks from concrete surface imagery using two-step light gradient boosting machine. Comput.-Aided Civ. Inf. 36, 1–12 (2020). https://doi.org/10.1111/mice.12564 12 T. Barisin, C. Jung, A. Nowacka, C. Redenbach, K. ...

    2020 doi

  3. [11]

    Depeursinge, A., Foncubierta-Rodriguez, A., Van de Ville, D., M ¨uller, H.: Multiscale lung texture signature learning using the Riesz transform. In: N. Ayache, H. Delingette, P. Golland, K. Mori (eds.) Medical Image Computing and Computer-Assisted Intervention – MICCAI 2012, ...

    2012 doi

  4. [12]

    Image Analysis & Stereology 38(1), 107–119 (2019)

    Dobrovolskij, D., Persch, J., Schladitz, K., Steidl, G.: Structure detection with second order Riesz transform. Image Analysis & Stereology 38(1), 107–119 (2019). DOI 10.5566/ias.1964

    2019 doi

  5. [13]

    Ehrig, K., Goebbels, J., Meinel, D., Paetsch, O., Prohaska, S., Zobel, V .: Comparison of Crack Detection Methods for Analyzing Damage Processes in Concrete with Computed Tomogra- phy. In: Int. Symp. Dig. Ind. Radiol. Comp. Tomogr. (2011)

    2011

  6. [14]

    Materials and Structures 56(163) (2023)

    Grzesiak, S., Barisin, T., Schladitz, K., Pahn, M.: Analysis of the bond behaviour of a GFRP rebar in concrete by in-situ 3d imaging test. Materials and Structures 56(163) (2023). DOI 10.1617/s11527-023-02247-0

    2023 doi

  7. [15]

    International Journal of Wavelets, Multiresolution and Information Processing 12(03), 1450027 (2014)

    H ¨auser, S., Heise, B., Steidl, G.: Linearized Riesz transform and quasi-monogenic shear- lets. International Journal of Wavelets, Multiresolution and Information Processing 12(03), 1450027 (2014). DOI 10.1142/S0219691314500271

    2014 doi

  8. [16]

    Journal of Computing in Civil Engineering 34(5), 04020038 (2020)

    Hsieh, Y .A., Tsai, Y .J.: Machine learning for crack detection: Review and model performance comparison. Journal of Computing in Civil Engineering 34(5), 04020038 (2020). DOI 10. 1061/(ASCE)CP.1943-5487.0000918

    2020

  9. [17]

    In: 2018 26th European Signal Processing Conference (EUSIPCO), pp

    Jenkins, M.D., Carr, T.A., Iglesias, M.I., Buggy, T.W., Morison, G.: A deep convolutional neural network for semantic pixel-wise segmentation of road and pavement surface cracks. In: 2018 26th European Signal Processing Conference (EUSIPCO), pp. 2120–2124 (2018)

    2018

  10. [18]

    Medical Image Analysis 56 (2019)

    Joyseeree, R., Ot ´alora Montenegro, J., M¨uller, H., Depeursinge, A.: Fusing learned represen- tations from Riesz filters and deep CNN for lung tissue classification. Medical Image Analysis 56 (2019). DOI 10.1016/j.media.2019.06.006

    2019 doi

  11. [19]

    In: 11th Conference on Industrial Computed Tomography, Wels, Austria (iCT 2022) (2022)

    Jung, C., M ¨usebeck, F., Barisin, T., Schladitz, K., Redenbach, C., Kiesche, M., Pahn, M.: To- wards automatic crack segmentation in 3d concrete images. In: 11th Conference on Industrial Computed Tomography, Wels, Austria (iCT 2022) (2022)

    2022

  12. [20]

    Journal of Mathematics in Industry 13(10) (2023)

    Jung, C., Redenbach, C.: Crack modeling via minimum-weight surfaces in 3d Voronoi dia- grams. Journal of Mathematics in Industry 13(10) (2023). DOI 10.1186/s13362-023-00138-1

    2023 doi

  13. [21]

    K ¨onig, J., Jenkins, M., Mannion, M., Barrie, P., Morison, G.: Weakly-Supervised Surface Crack Segmentation by Generating Pseudo-Labels using Localization with a Classifier and Thresholding. (2021). https://arxiv.org/pdf/2109.00456.pdf (accessed 02 February 2022)

    2021 arXiv

  14. [22]

    IEEE Access 8, 51446–51459 (2020)

    Li, G., Wan, J., He, S., Liu, Q., Ma, B.: Semi-supervised semantic segmentation using ad- versarial learning for pavement crack detection. IEEE Access 8, 51446–51459 (2020). https://doi.org/10.1109/ACCESS.2020.2980086

    2020

  15. [23]

    Forum Bildverarbeitung 2018 (2018)

    M ¨uller, O., Moghiseh, A., Stephani, H., Rottmayer, N., Huang, F.: Application of deep learning for crack segmentation on concrete surface. Forum Bildverarbeitung 2018 (2018). https://doi.org/10.5445/KSP/1000085290

    2018

  16. [24]

    Strain 58(3), e12412 (2022)

    Nogatz, T., Redenbach, C., Schladitz, K.: 3D optical flow for large CT data of materials mi- crostructures. Strain 58(3), e12412 (2022). DOI https://doi.org/10.1111/str.12412. URL https://onlinelibrary.wiley.com/doi/abs/10.1111/str.12412

    2022 doi

  17. [25]

    Journal of Soft Computing in Civil Engineering (2025)

    Nowacka, A., Schladitz, K., Grzesiak, S., Pahn, M.: Segmentation of spatial crack structures in concrete by deep learning enabling image based characterization. Journal of Soft Computing in Civil Engineering (2025). To appear

    2025

  18. [26]

    In: Proc

    Paetsch, O., Baum, D., Ehrig, K., Meinel, D., Prohaska, S.: Automated 3D Crack Detection for Analyzing Damage Processes in Concrete with Computed Tomography. In: Proc. Conf. Ind. Comp. Tomogr., pp. 321–330 (2012)

    2012

  19. [27]

    International Journal of Wavelets, Multiresolution and Information Process- ing 20(03), 2040007 (2022)

    Reinhardt, M., Bernstein, S., Heise, B.: Multi-scale orientation estimation using higher order Riesz transforms. International Journal of Wavelets, Multiresolution and Information Process- ing 20(03), 2040007 (2022). DOI 10.1142/S021969132040007X

    2022 doi

  20. [28]

    In: Lect

    Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomedical image segmentation. In: Lect. Notes Comput. Sc., pp. 234–241 (2015). https://doi.org/10. 1007/978-3-319-24574-4_28 Cracks in concrete 13

    2015

  21. [29]

    Salamon, M., Reims, N., Hanke, R.: High energy X-ray tomography for large concrete struc- tures. In: M. Pahn, C. Thiele, C. Glock, J. Schnell (eds.) Vielfalt im Massivbau: Festschrift zum 65. Geburtstag von Prof. Dr.-Ing. J¨urgen Schnell, pp. 595–609. Ernst & Sohn (2018)

    2018

  22. [30]

    Shi, Y ., Cui, L., Qi, Z., Meng, F., Chen, Z.: Automatic Road Crack Detection Using Random Structured Forests. IEEE T. Intell. Transp. 17(12), 3434–3445 (2016). https://doi. org/10.1109/TITS.2016.2552248

    2016

  23. [31]

    IEEE Access 8, 170939–170950 (2020)

    Shim, S., Kim, J., Cho, G.C., Lee, S.W.: Multiscale and adversarial learning-based semi- supervised semantic segmentation approach for crack detection in concrete structures. IEEE Access 8, 170939–170950 (2020). https://doi.org/10.1109/ACCESS.2020. 3022786

    2020 doi

  24. [32]

    IEEE Transactions on Image Processing 19(3), 636–652 (2010)

    Unser, M., Van De Ville, D.: Wavelet steerability and the higher-order Riesz transform. IEEE Transactions on Image Processing 19(3), 636–652 (2010). DOI 10.1109/TIP.2009.2038832

    2010

  25. [33]

    Ultramicroscopy 157, 35–47 (2015)

    van Aarle, W., Palenstijn, W.J., De Beenhouwer, J., Altantzis, T., Bals, S., Batenburg, K.J., Sijbers, J.: The ASTRA toolbox: A platform for advanced algorithm development in electron tomography. Ultramicroscopy 157, 35–47 (2015). DOI https://doi.org/10.1016/j.ultramic. 2015.05.002

    2015 doi

  26. [34]

    Automat Constr

    Wang, W., Su, C.: Semi-supervised semantic segmentation network for surface crack detec- tion. Automat Constr. 128, 103786 (2021). https://doi.org/10.1016/j.autcon. 2021.103786 14 T. Barisin, C. Jung, A. Nowacka, C. Redenbach, K. Schladitz Fig. 2: Examples of cracks in concret...

    2021

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.