Pith. sign in

REVIEW 4 major objections 5 minor 61 references

A 3D Multimodal Feature for Infrastructure Anomaly Detection

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

Pith's one-line read A handcrafted intensity histogram fused with FPFH lets PatchCore detect cracks and water ingress in masonry bridge and tunnel point clouds.

desk verdict A useful proof-of-concept for fusing FPFH with a 3D intensity histogram, but the paper's own Limitations section undercuts its headline claim of improved small-crack detection. read the letter →

arxiv 2502.05779 v1 pith:HPGNSIRL submitted 2025-02-09 cs.CV

classification cs.CV
keywords anomalydetectionpointcloud3DMulti-FPFHIFPFHintensityfeaturecrackwateringressinfrastructureinspection
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 tries to establish that infrastructure anomaly detection on point clouds improves when a handcrafted 3D intensity histogram is concatenated with a geometric descriptor (FPFH). If it is right, an inspection pipeline that needs only two point clouds—one normal and one test—can catch small cracks that geometry alone misses and can also flag water ingress, which has no geometric signature at all. On synthetic masonry arches the combined feature localizes intrados cracks more sharply than FPFH and beats a multimodal baseline; on real bridge and tunnel scans it finds defects in cases where pure geometry scores zero. The paper is explicit that the gain depends on local intensity contrast, so uniform-intensity patches and cracks thinner than the scan resolution remain hard cases.

What carries the argument

The central object is 3DMulti-FPFHI. For each point, a ball query finds neighbours; FPFH encodes local shape through histograms of Darboux-frame angles, and the new 3D intensity feature encodes local radiometric contrast through a histogram of absolute relative intensity values between the centre point and its neighbours. Concatenating the two histograms gives PatchCore a single feature vector carrying both geometric and radiometric anomaly signals, and the absolute-relative encoding is meant to damp intensity drift between scans taken under different environmental conditions.

What would settle it

A controlled experiment would register two scans of the same defect-free wall taken under different lighting, moisture, or scanner settings and run the 3D intensity feature alone through PatchCore: if non-defect points produce anomaly scores as high as those from a real crack or water patch, the absolute-relative encoding has not cancelled environmental intensity drift. A second check is to apply the full method to a large uniform water patch and observe whether only its boundary is flagged, which the paper predicts.

Watch

Extended reading notes

Core claim

The paper's central claim is that 3DMulti-FPFHI—the vector formed by concatenating Fast Point Feature Histograms (FPFH) with a per-point 3D intensity histogram built from absolute relative intensity differences between each point and its neighbours—outperforms FPFH alone and the CPMF multimodal baseline for infrastructure anomaly detection inside the PatchCore framework. The intensity channel increases anomaly separation for intrados cracks and for water patches, while the FPFH channel preserves detection of extrados cracks that appear only as geometric distortions. The method requires no training, only a normal point cloud and a test point cloud, which the paper argues makes it practical where learning-based methods lack data.

Load-bearing premise

The method assumes that for undamaged surfaces, local brightness contrast between a point and its neighbours stays nearly unchanged between the normal scan and the test scan, even when lighting, moisture, or scanner settings change; the paper itself notes this fails where brightness is nearly uniform, which is why the interiors of large water patches are missed.

Editorial extensions

If this is right

  • On the synthetic arch, 3DMulti-FPFHI achieves higher F1 scores than FPFH in most support-movement cases and localizes intrados cracks more sharply, while also outperforming CPMF.
  • On the real masonry bridge, FPFH alone scores F1 = 0, whereas 3DMulti-FPFHI locates the intrados crack and several water patches.
  • On the concrete tunnel, 3DMulti-FPFHI keeps intrados-crack detection close to FPFH despite intensity noise, with higher F1 scores at later loading steps.
  • The pipeline needs only two registered point clouds and no training, making it usable for anomaly detection when labelled defect data are scarce.
  • Intensity anomalies enable detection of non-geometric defects such as water patches, which no purely geometric feature can reveal.

Reading between the lines

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

  • Beyond the paper: because the 3D intensity histogram works through local contrast, a pre-processing step that normalizes intensity across scans (or over surface albedo) could extend detection into the uniform interiors of large water patches.
  • Beyond the paper: the handcrafted intensity histogram could be combined with learned geometric descriptors, not just FPFH, to push crack detection below the current resolution limits once denser scans are available.
  • Beyond the paper: the claimed superiority over image-based multimodal features should be re-tested on cracks wider than a few millimetres, where the paper's limitation section says both geometry and intensity channels start to fail.
  • Beyond the paper: the F1-based comparisons depend on user-chosen thresholds; a threshold-free metric such as area under the ROC curve on the same datasets would give a more portable comparison across bridges and tunnels.
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 a 3D multimodal feature, 3DMulti-FPFHI, that concatenates a modified FPFH descriptor with a handcrafted 3D intensity histogram and integrates it into the PatchCore anomaly detection framework. The method is evaluated on synthetic masonry arch point clouds, a real masonry arch bridge, and a full-scale concrete tunnel experiment. The authors claim that the intensity feature improves crack detection and enables detection of water ingress, and that 3DMulti-FPFHI outperforms FPFH and the CPMF baseline. The paper also documents updates to a synthetic dataset, including crack-surface geometry and intensity simulation.

Significance. If the claims held, the method would offer a training-free anomaly detection approach for infrastructure point clouds that leverages both geometric and radiometric information, with the code and dataset made publicly available. The empirical evidence, however, does not fully support the central claims: the method fails on small cracks (the motivating regime), CPMF is not evaluated on real data, and the F1 results are low and threshold-dependent. The water-patch detection results and the release of code/data are positive contributions, as is the effort to create a more realistic synthetic dataset with crack surfaces.

major comments (4)
  1. [§1 and Limitations] The central claim of the paper is that 3DMulti-FPFHI improves crack detection, with the introduction explicitly motivated by FPFH's insufficient sensitivity to small cracks. However, the Limitations section states verbatim that the proposed 3D multimodal feature failed to capture small cracks in the synthetic dataset (intrados and extrados cracks with maximum values of 2 mm and 6 mm, respectively) and in the real tunnel (crack widths from 0.02 mm to several millimetres). This is a direct contradiction with the abstract's 'improving crack detection' claim. The stated reasons (subtle geometry and point-cloud resolution below crack width) are structural rather than incidental, so the claim must be qualified to the detectable crack-width regime or supported with evidence of small-crack detection.
  2. [§4.2, Table 2 and §4.4, Table 4] The claim that 3DMulti-FPFHI 'outperforms FPFH and a state-of-the-art multimodal anomaly detection method' is only partially supported. On the synthetic dataset, CPMF achieves a higher F1 score than 3DMulti-FPFHI for the z-displacement case at threshold 0.5*s (0.531 vs 0.102) and for the x-rotation/z-translation case at 0.5*s (0.427 vs 0.318); on the real tunnel, FPFH alone has a higher F1 at loading step 103 for threshold 0.5*s (0.043 vs 0.026), and Section 4.4 notes that the intensity feature compromises extrados crack detection. The superiority claim should be restricted to the datasets, thresholds, and crack types for which it actually holds, with exceptions acknowledged.
  3. [§4.1 and §4.2] The CPMF baseline is evaluated only on synthetic data; it is not applied to the real arch or tunnel datasets, yet the abstract and conclusions claim outperformance against this state-of-the-art method without this restriction. In addition, no confidence intervals, standard deviations, or repeated-run statistics are provided for any F1 score or distribution metric, and the 'statistical analysis' in Section 4.2.2 is purely descriptive (KDE and sample means) rather than inferential. The paper should either provide statistical evidence for the claimed differences or explicitly frame the comparisons as indicative rather than conclusive.
  4. [§2.1 and §4.3] The 3D intensity feature is built on the assumption that absolute relative intensity differences between a point and its neighbours remain stable for non-defect points between the normal and test scans. The paper acknowledges in Section 4.3 that this assumption fails when intensity is nearly uniform (water patch interiors) and in Section 4.4 that scan resolution below crack width makes the intensity feature useless. These are not edge cases but central operating conditions for the proposed feature; the paper should state these limitations upfront and temper the conclusions about generalizability accordingly, since the claimed advantages of the feature disappear in precisely those regimes.
minor comments (5)
  1. [Abstract] The abstract's statement that the method 'outperforms FPFH and a state-of-the-art multimodal anomaly detection method' should be qualified, since the CPMF comparison is conducted only on synthetic data and the F1 results are mixed across thresholds and datasets.
  2. [§4.1] The choice of anomaly thresholds 0.3*s and 0.5*s is arbitrary and not justified; the paper would benefit from reporting a threshold-independent metric such as area under the ROC curve or precision-recall curve.
  3. [§4.3] The statement that 'the metric alone cannot fairly evaluate the performance' is not accompanied by an alternative evaluation metric; providing a quantitative localization metric would strengthen the assessment.
  4. [References] There is a typographical error in the reference list: the Dais et al. (2021) entry contains the stray string 'claslogoglu2016cospairs' in the title, and several author names appear with unintended spaces (e.g., 'Sta lowska').
  5. [Figure 9] In the discussion of Figure 9, the text states that '(c) is retrieved by overlapping (b) and (c)', which appears to be a typo; likely one of these should be (a).

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation found: the intensity feature is a handcrafted input to an empirical benchmark, not a quantity defined by the target it predicts.

full rationale

The paper's derivation chain is empirical rather than analytical. 3DMulti-FPFHI is built by concatenating a standard FPFH descriptor with a handcrafted histogram of absolute relative intensity differences; neither component is fitted to crack labels or to the anomaly scores it later predicts. PatchCore builds a memory bank from normal point clouds only and scores test points by nearest-neighbour distance, so no fitted parameter is renamed as a prediction. The thresholds and radii are dataset-specific hyperparameters reported for reproducibility, and F1-scores are shown at multiple thresholds rather than optimised on the test labels. The synthetic dataset intentionally couples crack geometry to simulated intensity reduction, so the synthetic ablation partly tests the feature against a signal that was inserted by construction; however, the real masonry arch and tunnel experiments provide external data where the intensity contrast arises from actual mortar reflectivity or is absent, so the central claim is not forced by the simulation. Self-citations (Jing et al. 2024b for the PatchCore implementation and synthetic arch framework) provide the baseline and infrastructure but are not invoked as a uniqueness theorem or as the justification for the intensity feature's benefit. The Limitations section concedes that small cracks are missed, which weakens the scope of the claim but is not circularity. No step equates the output to the input by definition.

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

The central claim rests on standard algorithms (FPFH, PatchCore) and on assumptions about intensity invariance and synthetic reflectivity. The fused feature itself has no derived guarantees, and the evaluation depends on several hand-set hyperparameters. No new physical entities are postulated.

free parameters (6)
  • FPFH bin count = 30
    Increased from the standard 11 to 30 in Section 2.1 to improve sensitivity; all FPFH and 3DMulti-FPFHI results depend on this choice.
  • 3D intensity histogram bin count = 30
    Set equal to the FPFH bin count in Section 2.1; no sensitivity analysis across bin counts is provided.
  • Ball query and FPFH radii = 0.12 m normals, 1 m arch FPFH, 0.6 m tunnel FPFH
    Set per dataset in Section 4.1; anomaly scores and the diffusion effect depend on these radii.
  • PatchCore memory bank size = 4000
    Set in Section 4.1; the coreset size affects nearest-neighbour distances and anomaly scores.
  • Anomaly threshold and F1 thresholds = 0.5 visual; 0.3*s and 0.5*s for F1
    The 0.5 threshold was chosen in Section 4.2.1 to highlight extrados cracks, and F1 comparisons across all datasets use 0.3*s and 0.5*s, making rankings threshold-dependent.
  • CPMF camera settings = focal 700/450 mm, 512x512, 27 orientations
    Defined in Section 2.2 for the CPMF baseline; these settings are not tuned to infrastructure data and likely affect CPMF's poor performance.
assumptions (5)
  • standard math FPFH and PatchCore methods work as published
    The paper imports FPFH (Rusu et al., 2009) and PatchCore (Roth et al., 2022) without re-deriving them; Sections 2.1 and 2.3 rely on their correctness.
  • domain assumption A scalar intensity value represents meaningful material response
    Intensity amplitude is assumed to reflect crack surfaces and water patches; introduced in Section 2.1 and used throughout the method.
  • domain assumption Uniform reflectivity in synthetic models
    Section 3.1 assumes uniform reflectivity across masonry, mortar, and inner crack surfaces, so intensity differences are attributed to crack geometry rather than material changes.
  • domain assumption Absolute-relative intensity encoding makes non-defect points comparable across scans
    Section 4.3 states that PatchCore's pairing assumption requires non-defect points to match across dates and that this fails when intensity is nearly uniform or when environmental conditions vary strongly.
  • domain assumption Crack labels extend uniformly along the tunnel ring
    Section 3.3 assumes cracks observed within a 20 cm grid on the concrete surface extend uniformly along the full ring length, which directly shapes the ground truth used for F1 scores.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A 3D Multimodal Feature for Infrastructure Anomaly Detection." pith.science (2026). https://pith.science/paper/HPGNSIRL

@misc{pith2026250205779,
  author       = {Pith},
  title        = {Pith review of: A 3D Multimodal Feature for Infrastructure Anomaly Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HPGNSIRL}},
  note         = {Machine review of arXiv:2502.05779}
}
read the original abstract

Ageing structures require periodic inspections to identify structural defects. Previous work has used geometric distortions to locate cracks in synthetic masonry bridge point clouds but has struggled to detect small cracks. To address this limitation, this study proposes a novel 3D multimodal feature, 3DMulti-FPFHI, that combines a customized Fast Point Feature Histogram (FPFH) with an intensity feature. This feature is integrated into the PatchCore anomaly detection algorithm and evaluated through statistical and parametric analyses. The method is further evaluated using point clouds of a real masonry arch bridge and a full-scale experimental model of a concrete tunnel. Results show that the 3D intensity feature enhances inspection quality by improving crack detection; it also enables the identification of water ingress which introduces intensity anomalies. The 3DMulti-FPFHI outperforms FPFH and a state-of-the-art multimodal anomaly detection method. The potential of the method to address diverse infrastructure anomaly detection scenarios is highlighted by the minimal requirements for data compared to learning-based methods. The code and related point cloud dataset are available at https://github.com/Jingyixiong/3D-Multi-FPFHI.

Figures

Figures reproduced from arXiv: 2502.05779 by the authors.

Figure 1
Figure 1. Overview of anomaly detection using PatchCore. The input point cloud data consist of intensity values along with x, y, and z coordinates. FPFH are extracted from both normal and test data (test data includes both normal and anomalous points) using point coordinates. The 3D intensity feature is derived from the intensity information of the point clouds. By concatenating FPFH and the 3D intensity feature, the proposed… view at source ↗
Figure 2
Figure 2. Computation of intensity histogram using FPFH. Intensity values of neighbours [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 1
Figure 1. Inspired by Logoglu et al. (2016), this embedding computes the [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figures from the paper (17 more)
Figure 3
Figure 3. Figure 3: Computation of the pseudo-2D image feature on normal data. The point cloud [PITH_FULL_IMAGE:figures/full_fig_p009_3.png]
Figure 4
Figure 4. Figure 4: Modifications to the synthetic datasets, including updates to both the FEM [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Integration of the intrados crack surfaces in the updated synthetic models. [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Visualization of the London Bridge Station point cloud for arch E57. (a) shows [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: (a) Section view of Arch E57, highlighting settlement records indicated by red [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Top view of the segmented point clouds of arches 1 and 2, with intensity values [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: Visualization of the full-scale physical experiment and the resulting point clouds. [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]
Figure 10
Figure 10. Figure 10: Visualization of the geometric distortion of the real tunnel in bird view and [PITH_FULL_IMAGE:figures/full_fig_p018_10.png]
Figure 11
Figure 11. Figure 11: Anomaly detection results on the masonry arch for four different support move [PITH_FULL_IMAGE:figures/full_fig_p020_11.png]
Figure 12
Figure 12. Figure 12: Distributions of min dists for the x displacement case. (a)-(e) represents distributions of ’non-crack’ (blue) and ’crack’ (red) points for five different feature types. (f)-(j) show distributions of ’non-crack’ points together with their corresponding mean and varian…
Figure 13
Figure 13. Figure 13: Anomaly detection results for Arch 1 and 2 of the real masonry arch point [PITH_FULL_IMAGE:figures/full_fig_p026_13.png]
Figure 14
Figure 14. Figure 14: Visualization of 3DMulti-FPFHI in detecting intrados cracks and water patches [PITH_FULL_IMAGE:figures/full_fig_p027_14.png]
Figure 15
Figure 15. Figure 15: Visualization of intensity map discrepancies in Arch 1. (a) and (b) show [PITH_FULL_IMAGE:figures/full_fig_p029_15.png]
Figure 16
Figure 16. Figure 16: Anomaly detection results corresponding to four different loading steps on the [PITH_FULL_IMAGE:figures/full_fig_p030_16.png]
Figure 17
Figure 17. Figure 17: Visualization of zoomed crack detection results using FPFH. (a)-(c), (d)-(f), [PITH_FULL_IMAGE:figures/full_fig_p032_17.png]
Figure 18
Figure 18. Figure 18: Limitations of intensity information in detecting cracks in the real tunnel point [PITH_FULL_IMAGE:figures/full_fig_p033_18.png]
Figure 19
Figure 19. Figure 19: Comparison of detection results using FPFH and 3DMulti-FPFHI at loading [PITH_FULL_IMAGE:figures/full_fig_p034_19.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

61 extracted references · 35 canonical work pages

  1. [1]

    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 ":" * " " * FUNCTION f...

  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 ":" * " " * FUNCTION f...

  3. [3]

    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 ":" * " " * FUNCTION f...

  4. [4]

    , author DeJong, M.J

    author Acikgoz, S. , author DeJong, M.J. , author Soga, K. , year 2018 . title Sensing dynamic displacements in masonry rail bridges using 2d digital image correlation . journal Structural Control and Health Monitoring volume 25 , pages e2187 . https://doi.org/10.1002/stc.2187

  5. [5]

    , author Soga, K

    author Acikgoz, S. , author Soga, K. , author Woodhams, J. , year 2017 . title Evaluation of the response of a vaulted masonry structure to differential settlements using point cloud data and limit analyses . journal Construction and Building Materials volume 150 , pages 916--931 . https://doi.org/10.1016/j.conbuildmat.2017.05.075

  6. [7]

    , author Debono, C.J

    author Attard, L. , author Debono, C.J. , author Valentino, G. , author Di Castro, M. , author Masi, A. , author Scibile, L. , year 2019 . title Automatic crack detection using mask r-cnn , in: booktitle 2019 11th international symposium on image and signal processing and analysis (ISPA) , organization IEEE . pp. pages 152--157 . 10.1109/ISPA.2019.8868619

  7. [8]

    , author Pfeiffer, M

    author Beggel, L. , author Pfeiffer, M. , author Bischl, B. , year 2020 . title Robust anomaly detection in images using adversarial autoencoders , in: booktitle Machine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2019, W \"u rzburg, Germany, September 16--20, 2019, Proceedings, Part I , organization Springer . pp. pages ...

  8. [9]

    , author Jin, X

    author Bergmann, P. , author Jin, X. , author Sattlegger, D. , author Steger, C. , year 2021 . title The mvtec 3d-ad dataset for unsupervised 3d anomaly detection and localization . journal arXiv preprint arXiv:2112.09045 https://doi.org/10.5220/0010865000003124

Show all 61 references
  1. [10]

    , year 2022

    author Brackenbury, D. , year 2022 . title Automated image-based inspection of masonry arch bridges

  2. [11]

    , author Wan, Q

    author Cao, Y. , author Wan, Q. , author Shen, W. , author Gao, L. , year 2022 . title Informative knowledge distillation for image anomaly segmentation . journal Knowledge-Based Systems volume 248 , pages 108846 . https://doi.org/10.1016/j.knosys.2022.108846

  3. [12]

    , author Xu, X

    author Cao, Y. , author Xu, X. , author Shen, W. , year 2023 . title Complementary pseudo multimodal feature for point cloud anomaly detection . journal arXiv preprint arXiv:2303.13194 https://doi.org/10.48550/arXiv.2303.13194

  4. [13]

    u y \"u k \

    author Cha, Y.J. , author Choi, W. , author B \"u y \"u k \"o zt \"u rk, O. , year 2017 . title Deep learning-based crack damage detection using convolutional neural networks . journal Computer-Aided Civil and Infrastructure Engineering volume 32 , pages 361--378 . https://doi...

  5. [14]

    u y \"u k \

    author Cha, Y.J. , author Choi, W. , author Suh, G. , author Mahmoudkhani, S. , author B \"u y \"u k \"o zt \"u rk, O. , year 2018 . title Autonomous structural visual inspection using region-based deep learning for detecting multiple damage types . journal Computer-Aided Civi...

  6. [15]

    , author Jahanshahi, M.R

    author Chen, F.C. , author Jahanshahi, M.R. , year 2017 . title Nb-cnn: Deep learning-based crack detection using convolutional neural network and na \" ve bayes data fusion . journal IEEE Transactions on Industrial Electronics volume 65 , pages 4392--4400 . 10.1109/TIE.2017.2764844

  7. [16]

    , author Ramirez, P.Z

    author Costanzino, A. , author Ramirez, P.Z. , author Lisanti, G. , author Di Stefano, L. , year 2024 . title Multimodal industrial anomaly detection by crossmodal feature mapping , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition...

  8. [17]

    , author Bal, I.E

    author Dais, D. , author Bal, I.E. , author Smyrou, E. , author Sarhosis, V. , year 2021 . title Automatic crack claslogoglu2016cospairsification and segmentation on masonry surfaces using convolutional neural networks and transfer learning . journal Automation in Construction...

  9. [18]

    , author Birdal, T

    author Deng, H. , author Birdal, T. , author Ilic, S. , year 2018 . title Ppfnet: Global context aware local features for robust 3d point matching , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , pp. pages 195--205 . https://doi.o...

  10. [19]

    , author Dong, W

    author Deng, J. , author Dong, W. , author Socher, R. , author Li, L.J. , author Li, K. , author Fei-Fei, L. , year 2009 . title Imagenet: A large-scale hierarchical image database , in: booktitle 2009 IEEE Conference on Computer Vision and Pattern Recognition , pp. pages 248-...

  11. [20]

    , author Wang, S

    author Dong, Q. , author Wang, S. , author Chen, X. , author Jiang, W. , author Li, R. , author Gu, X. , year 2023 . title Pavement crack detection based on point cloud data and data fusion . journal Philosophical Transactions of the Royal Society A volume 381 , pages 20220165...

  12. [21]

    , author Li, C

    author Fan, Z. , author Li, C. , author Chen, Y. , author Wei, J. , author Loprencipe, G. , author Chen, X. , author Di Mascio, P. , year 2020 . title Automatic crack detection on road pavements using encoder-decoder architecture . journal Materials volume 13 , pages 2960 . ht...

  13. [22]

    , author Dixit, M

    author Gupta, P. , author Dixit, M. , year 2022 . title Image-based crack detection approaches: a comprehensive survey . journal Multimedia Tools and Applications volume 81 , pages 40181--40229 . https://doi.org/10.1007/s11042-022-13152-z

  14. [23]

    , author Napolitano, R.K

    author Hallee, M.J. , author Napolitano, R.K. , author Reinhart, W.F. , author Glisic, B. , year 2021 . title Crack detection in images of masonry using cnns . journal Sensors volume 21 , pages 4929 . https://doi.org/10.3390/s21144929

  15. [24]

    , author Zhang, X

    author He, K. , author Zhang, X. , author Ren, S. , author Sun, J. , year 2016 . title Deep residual learning for image recognition , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , pp. pages 770--778

  16. [25]

    , author Han, S

    author Jiang, Y. , author Han, S. , author Bai, Y. , year 2021 . title Building and infrastructure defect detection and visualization using drone and deep learning technologies . journal Journal of Performance of Constructed Facilities volume 35 , pages 04021092 . https://doi....

  17. [26]

    , author Sheil, B

    author Jing, Y. , author Sheil, B. , author Acikgoz, S. , year 2022 . title Segmentation of large-scale masonry arch bridge point clouds with a synthetic simulator and the bridgenet neural network . journal Automation in Construction volume 142 , pages 104459 . https://doi.org...

  18. [27]

    , author Sheil, B

    author Jing, Y. , author Sheil, B. , author Acikgoz, S. , year 2023 . title A method to generate realistic synthetic point clouds of damaged single-span masonry arch bridges , in: booktitle International Conference on Structural Analysis of Historical Constructions , organizat...

  19. [28]

    , author Sheil, B

    author Jing, Y. , author Sheil, B. , author Acikgoz, S. , year 2024 a. title A lightweight transformer-based neural network for large-scale masonry arch bridge point cloud segmentation . journal Computer-Aided Civil and Infrastructure Engineering https://doi.org/10.1111/mice.13201

  20. [29]

    , author Zhong, J.X

    author Jing, Y. , author Zhong, J.X. , author Sheil, B. , author Acikgoz, S. , year 2024 b. title Anomaly detection of cracks in synthetic masonry arch bridge point clouds using fast point feature histograms and patchcore . journal Automation in Construction volume 168 , pages 105766

  21. [30]

    , author Fan, Z

    author Lee, E.T. , author Fan, Z. , author Sencer, B. , year 2023 . title A new approach to detect surface defects from 3d point cloud data with surface normal gabor filter (sngf) . journal Journal of Manufacturing Processes volume 92 , pages 196--205 . https://doi.org/10.1016...

  22. [31]

    , author Ma, B

    author Li, G. , author Ma, B. , author He, S. , author Ren, X. , author Liu, Q. , year 2020 . title Automatic tunnel crack detection based on u-net and a convolutional neural network with alternately updated clique . journal Sensors volume 20 , pages 717 . https://doi.org/10.3...

  23. [32]

    , author Rui, Y

    author Li, T. , author Rui, Y. , author Zhu, H. , author Lu, L. , author Li, X. , year 2024 a. title Comprehensive digital twin for infrastructure: A novel ontology and graph-based modelling paradigm . journal Advanced Engineering Informatics volume 62 , pages 102747 . https:/...

  24. [33]

    , author Yao, J

    author Li, Y. , author Yao, J. , author Chen, K. , author Zhang, H. , author Sun, X. , author Qian, Q. , author Wu, X. , year 2024 b. title A collaborative anomaly localization method based on multi-modal images , in: booktitle 2024 27th International Conference on Computer Su...

  25. [34]

    , author Yue, Y

    author Liao, J. , author Yue, Y. , author Zhang, D. , author Tu, W. , author Cao, R. , author Zou, Q. , author Li, Q. , year 2022 a. title Automatic tunnel crack inspection using an efficient mobile imaging module and a lightweight cnn . journal IEEE Transactions on Intelligen...

  26. [35]

    , author Yue, Y

    author Liao, J. , author Yue, Y. , author Zhang, D. , author Tu, W. , author Cao, R. , author Zou, Q. , author Li, Q. , year 2022 b. title Automatic tunnel crack inspection using an efficient mobile imaging module and a lightweight cnn . journal IEEE Transactions on Intelligen...

  27. [36]

    , author Goyal, P

    author Lin, T.Y. , author Goyal, P. , author Girshick, R. , author He, K. , author Dollar, P. , year 2017 . title Focal loss for dense object detection , in: booktitle Proceedings of the IEEE International Conference on Computer Vision (ICCV) . https://doi.org/10.48550/arXiv.1...

  28. [37]

    , author Li, P

    author Lin, W. , author Li, P. , author Xie, X. , author Cao, Y. , author Zhang, Y. , year 2023 . title A novel back-analysis approach for the external loads on shield tunnel lining in service based on monitored deformation . journal Structural Control and Health Monitoring vo...

  29. [38]

    , author Sheil, B

    author Lin, W. , author Sheil, B. , author Zhang, P. , author Zhou, B. , author Wang, C. , author Xie, X. , year 2024 . title Seg2tunnel: A hierarchical point cloud dataset and benchmarks for segmentation of segmental tunnel linings . journal Tunnelling and Underground Space T...

  30. [39]

    , author Xie, G

    author Liu, J. , author Xie, G. , author Chen, R. , author Li, X. , author Wang, J. , author Liu, Y. , author Wang, C. , author Zheng, F. , year 2024 . title Real3d-ad: A dataset of point cloud anomaly detection . journal Advances in Neural Information Processing Systems volum...

  31. [40]

    , author Chen, S

    author Liu, W. , author Chen, S. , author Hauser, E. , year 2011 . title Lidar-based bridge structure defect detection . journal Experimental Techniques volume 35 , pages 27--34 . https://doi.org/10.1111/j.1747-1567.2010.00644.x

  32. [41]

    , author Cao, Y

    author Liu, Z. , author Cao, Y. , author Wang, Y. , author Wang, W. , year 2019 . title Computer vision-based concrete crack detection using u-net fully convolutional networks . journal Automation in Construction volume 104 , pages 129--139 . https://doi.org/10.1016/j.autcon.2...

  33. [42]

    , author Kalkan, S

    author Logoglu, K.B. , author Kalkan, S. , author Temizel, A. , year 2016 . title Cospair: colored histograms of spatial concentric surflet-pairs for 3d object recognition . journal Robotics and Autonomous Systems volume 75 , pages 558--570 . https://doi.org/10.1016/j.robot.20...

  34. [43]

    , author Sarhosis, V

    author Loverdos, D. , author Sarhosis, V. , year 2022 . title Automatic image-based brick segmentation and crack detection of masonry walls using machine learning . journal Automation in Construction volume 140 , pages 104389 . https://doi.org/10.1016/j.autcon.2022.104389

  35. [44]

    , author Wood, R.L

    author Mohammadi, M.E. , author Wood, R.L. , author Wittich, C.E. , year 2019 . title Non-temporal point cloud analysis for surface damage in civil structures . journal ISPRS International Journal of Geo-Information volume 8 , pages 527 . https://doi.org/10.3390/ijgi8120527

  36. [45]

    , author Hoskere, V

    author Narazaki, Y. , author Hoskere, V. , author Yoshida, K. , author Spencer, B.F. , author Fujino, Y. , year 2021 . title Synthetic environments for vision-based structural condition assessment of japanese high-speed railway viaducts . journal Mechanical Systems and Signal ...

  37. [46]

    , author Lou, Z

    author Nguyen, D.T. , author Lou, Z. , author Klar, M. , author Brox, T. , year 2019 . title Anomaly detection with multiple-hypotheses predictions , in: booktitle International Conference on Machine Learning , organization PMLR . pp. pages 4800--4809 . https://doi.org/10.4855...

  38. [47]

    , author Tah, J.H

    author Perez, H. , author Tah, J.H. , author Mosavi, A. , year 2019 . title Deep learning for detecting building defects using convolutional neural networks . journal Sensors volume 19 , pages 3556 . https://doi.org/10.3390/s19163556

  39. [48]

    , author Casula, G

    author Pesci, A. , author Casula, G. , author Boschi, E. , year 2011 . title Laser scanning the garisenda and asinelli towers in bologna (italy): Detailed deformation patterns of two ancient leaning buildings . journal Journal of cultural heritage volume 12 , pages 117--127 . ...

  40. [49]

    , author Voulodimos, A

    author Protopapadakis, E. , author Voulodimos, A. , author Doulamis, A. , author Doulamis, N. , author Stathaki, T. , year 2019 . title Automatic crack detection for tunnel inspection using deep learning and heuristic image post-processing . journal Applied intelligence volume...

  41. [50]

    , author Neumann, L

    author Reitmann, S. , author Neumann, L. , author Jung, B. , year 2021 . title Blainder—a blender ai add-on for generation of semantically labeled depth-sensing data . journal Sensors volume 21 , pages 2144 . https://doi.org/10.3390/s21062144

  42. [51]

    , author Soil \'a n, M

    author del R \' o-Barral, P. , author Soil \'a n, M. , author Gonz \'a lez-Collazo, S.M. , author Arias, P. , year 2022 . title Pavement crack detection and clustering via region-growing algorithm from 3d mls point clouds . journal Remote Sensing volume 14 , pages 5866 . https...

  43. [52]

    , author Pemula, L

    author Roth, K. , author Pemula, L. , author Zepeda, J. , author Sch \"o lkopf, B. , author Brox, T. , author Gehler, P. , year 2022 . title Towards total recall in industrial anomaly detection , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Patte...

  44. [53]

    , author Blodow, N

    author Rusu, R.B. , author Blodow, N. , author Beetz, M. , year 2009 . title Fast point feature histograms (fpfh) for 3d registration , in: booktitle 2009 IEEE international conference on robotics and automation , organization IEEE . pp. pages 3212--3217 . 10.1109/ROBOT.2009.5152473

  45. [54]

    , author Seeb \"o ck, P

    author Schlegl, T. , author Seeb \"o ck, P. , author Waldstein, S.M. , author Langs, G. , author Schmidt-Erfurth, U. , year 2019 . title f-anogan: Fast unsupervised anomaly detection with generative adversarial networks . journal Medical image analysis volume 54 , pages 30--44...

  46. [55]

    , author Savarese, S

    author Sener, O. , author Savarese, S. , year 2017 . title Active learning for convolutional neural networks: A core-set approach . journal arXiv preprint arXiv:1708.00489 https://doi.org/10.48550/arXiv.1708.00489

  47. [56]

    , author Suchocki, C

    author Sta owska, P. , author Suchocki, C. , author Rutkowska, M. , year 2022 . title Crack detection in building walls based on geometric and radiometric point cloud information . journal Automation in Construction volume 134 , pages 104065 . https://doi.org/10.1016/j.autcon....

  48. [57]

    , author Shazeer, N

    author Vaswani, A. , author Shazeer, N. , author Parmar, N. , author Uszkoreit, J. , author Jones, L. , author Gomez, A.N. , author Kaiser, . , author Polosukhin, I. , year 2017 . title Attention is all you need . journal Advances in neural information processing systems volum...

  49. [58]

    , author Gao, L

    author Wan, Q. , author Gao, L. , author Li, X. , author Wen, L. , year 2021 . title Industrial image anomaly localization based on gaussian clustering of pretrained feature . journal IEEE Transactions on Industrial Electronics volume 69 , pages 6182--6192 . 10.1109/TIE.2021.3094452

  50. [59]

    , author Peng, J

    author Wang, Y. , author Peng, J. , author Zhang, J. , author Yi, R. , author Wang, Y. , author Wang, C. , year 2023 . title Multimodal industrial anomaly detection via hybrid fusion , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognit...

  51. [60]

    , year 2018

    author Weglarczyk, S. , year 2018 . title Kernel density estimation and its application , in: booktitle ITM web of conferences , organization EDP Sciences . p. pages 00037

  52. [61]

    , author Acikgoz, S

    author Ye, C. , author Acikgoz, S. , author Pendrigh, S. , author Riley, E. , author DeJong, M. , year 2018 . title Mapping deformations and inferring movements of masonry arch bridges using point cloud data . journal Engineering Structures volume 173 , pages 530--545 . https:...

  53. [62]

    , author Yoon, S

    author Yi, J. , author Yoon, S. , year 2020 . title Patch svdd: Patch-level svdd for anomaly detection and segmentation , in: booktitle Proceedings of the Asian conference on computer vision . https://doi.org/10.1007/978-3-030-69544-6_23

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.