pith. sign in

arxiv: 2606.05641 · v1 · pith:QUJTRFO5new · submitted 2026-06-04 · 💻 cs.CV

Multi-Task Crack Foundation Model for Engineering-Reliable Crack Representation and Topology Preservation in Civil Infrastructure

Blessing Agyei Kyem , Joshua Kofi Asamoah , Eugene Denteh , Armstrong Aboah This is my paper

Pith reviewed 2026-06-28 02:25 UTC · model grok-4.3

classification 💻 cs.CV
keywords crack segmentationmulti-task learningfoundation model adaptationtopology preservationuncertainty estimationcivil infrastructurefew-shot learningskeleton reconstruction
0
0 comments X

The pith

CrackGeoFM adapts a frozen foundation model with three modules to deliver crack masks that preserve topology and include calibrated uncertainty estimates.

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

The paper aims to fix a practical gap in crack analysis for infrastructure: current segmentation models often score high on overlap metrics yet produce fragmented cracks, miss branches, and give no reliable confidence scores. CrackGeoFM freezes a visual backbone and adds three crack-specific modules that enhance high-frequency details, adapt features to crack patterns, and jointly output masks, skeletons, and uncertainty. Tested on twenty crack datasets, the approach reports state-of-the-art segmentation, better topology preservation, calibrated uncertainty, and strong performance when only five labeled images are available.

Core claim

CrackGeoFM shows that a frozen visual foundation backbone can be paired with a Frequency-Guided Crack Enhancement Module, a Crack-Domain Feature Adaptation Module, and a Structure-Aware Multi-Task Decoder to produce masks, reconstructed skeletons, and uncertainty maps that are simultaneously more accurate, topologically connected, and better calibrated than prior single-task or non-adapted models across twenty crack datasets.

What carries the argument

CrackGeoFM framework that freezes a visual foundation backbone and routes its features through FCEM for high-frequency crack cues, CFAM for domain adaptation, and SMTD for joint mask-skeleton-uncertainty decoding.

If this is right

  • Segmentation and skeleton outputs remain connected even on thin or branched cracks.
  • Uncertainty maps stay reliable when the input domain shifts between datasets.
  • Only five labeled images suffice for competitive few-shot performance on new crack collections.
  • Joint training of the three tasks improves each individual output compared with separate models.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same frozen-backbone-plus-specialized-decoder pattern could be tested on other thin linear structures such as road networks or vessel segmentation.
  • Calibrated uncertainty could be used to prioritize inspection resources toward high-uncertainty crack regions.
  • The frequency-guided and structure-aware modules might transfer to medical imaging tasks where topology matters more than pixel overlap.

Load-bearing premise

The three new modules can be attached to a frozen backbone to raise segmentation accuracy, topology preservation, and uncertainty calibration together without hidden trade-offs or dataset-by-dataset retuning.

What would settle it

On a held-out crack dataset the model would need to show either lower topology scores than the prior best method or uncertainty estimates whose calibration error exceeds that of a strong baseline.

Figures

Figures reproduced from arXiv: 2606.05641 by Armstrong Aboah, Blessing Agyei Kyem, Eugene Denteh, Joshua Kofi Asamoah.

Figure 1
Figure 1. Figure 1: Overall architecture of CrackGeoFM CrackGeoFM reformulates crack segmentation as a multi-output represen￾tation problem. The model jointly estimates three complementary outputs from a shared representation: a segmentation mask 𝑀^ ∈ [0, 1]𝐵×1×𝐻×𝑊 that localizes crack regions, a skeleton map 𝑆^ ∈ [0, 1]𝐵×1×𝐻×𝑊 that captures the geometric centerline and connectivity of each crack, and a per-pixel uncer￾tainty… view at source ↗
Figure 2
Figure 2. Figure 2: Architecture of the Frequency-Guided Crack Enhancement Module (FCEM) [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Architecture of a single Crack-Domain Feature Adaptation Module (CFAM) [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Architecture of the Structure-Aware Multi-Task Decoder (SMTD). [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Mask-derived skeleton target generation pipeline. [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Per-dataset Dice heatmap on validation datasets [PITH_FULL_IMAGE:figures/full_fig_p026_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparison of segmentation predictions on the seven validation [PITH_FULL_IMAGE:figures/full_fig_p028_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative segmentation comparison on six zero-shot held-out datasets never [PITH_FULL_IMAGE:figures/full_fig_p030_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Per-dataset Dice heatmap on some zero-shot datasets [PITH_FULL_IMAGE:figures/full_fig_p031_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Multi-metric radar comparison on zero-shot datasets (mean across all six held [PITH_FULL_IMAGE:figures/full_fig_p032_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Few-shot adaptation curves on six held-out datasets. [PITH_FULL_IMAGE:figures/full_fig_p034_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: CrackGeoFM’s three simultaneous outputs (Pred Mask, Pred Skeleton, Uncer [PITH_FULL_IMAGE:figures/full_fig_p036_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Reliability diagrams for four validation (top row) and four zero-shot (bottom [PITH_FULL_IMAGE:figures/full_fig_p039_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Predictions from each ablation configuration (A through F) across in-domain [PITH_FULL_IMAGE:figures/full_fig_p042_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: CrackGeoFM predictions with five backbone architectures across in-domain [PITH_FULL_IMAGE:figures/full_fig_p044_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Critical difference diagram for Dice on validation datasets (Friedman test, [PITH_FULL_IMAGE:figures/full_fig_p045_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Dice distributions across validation (top) and zero-shot (bottom) datasets. [PITH_FULL_IMAGE:figures/full_fig_p046_17.png] view at source ↗
read the original abstract

Reliable crack assessment requires not only accurate pixel-level masks but also connected crack geometry and confidence estimates that remain stable under domain shift. However, existing segmentation models can achieve high overlap scores while fragmenting cracks, missing fine branches, and providing no calibrated uncertainty. To address this gap, this paper proposes CrackGeoFM, a multi-task framework that combines a frozen visual foundation backbone with crack-specific adaptation for mask prediction, skeleton reconstruction, and uncertainty estimation. The framework integrates a Frequency-Guided Crack Enhancement Module (FCEM) to enhance high-frequency crack cues, a Crack-Domain Feature Adaptation Module (CFAM) to adapt frozen backbone features to crack-domain patterns, and a Structure-Aware Multi-Task Decoder (SMTD) to jointly decode masks, skeletons, and uncertainty. Across 20 crack datasets, CrackGeoFM achieves state-of-the-art segmentation, improved topology preservation, calibrated uncertainty, and effective few-shot adaptation with only five labeled images. These results support reliable, generalizable, and engineering-oriented crack analysis for infrastructure assessment.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes CrackGeoFM, a multi-task framework that attaches three modules (FCEM for high-frequency crack cue enhancement, CFAM for adapting frozen backbone features to crack patterns, and SMTD for joint mask/skeleton/uncertainty decoding) to a frozen visual foundation backbone. It claims that this architecture delivers state-of-the-art segmentation, improved topology preservation, calibrated uncertainty, and effective five-shot adaptation across 20 crack datasets, thereby enabling engineering-reliable crack assessment in civil infrastructure.

Significance. If the empirical claims are substantiated with proper controls, the work would address a genuine gap between pixel-overlap metrics and topology/uncertainty requirements in infrastructure inspection. The frozen-backbone design and explicit topology objective are practically attractive; however, the absence of any quantitative results, ablations, or loss formulations in the manuscript prevents evaluation of whether the claimed simultaneous gains are achievable.

major comments (2)
  1. Abstract: the central claim that FCEM+CFAM+SMTD jointly deliver SOTA segmentation, topology preservation, and calibrated uncertainty without task interference is unsupported by any tables, baseline comparisons, ablation results, or numerical metrics; the manuscript therefore provides no evidence that the multi-task objectives can be optimized simultaneously across 20 datasets.
  2. Abstract (SMTD description): no loss formulation, task-weighting scheme, or training procedure is supplied for the joint decoding of masks, skeletons, and uncertainty; without these details it is impossible to assess whether the claimed absence of trade-offs holds or whether per-dataset retuning was required.
minor comments (1)
  1. Abstract: the phrase 'engineering-oriented crack analysis' is used without defining the concrete engineering requirements (e.g., minimum topology metric thresholds or uncertainty calibration targets) that the model is asserted to satisfy.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed feedback. We agree that the abstract claims require explicit empirical support and that the SMTD module description needs additional methodological details to allow proper evaluation. We will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: Abstract: the central claim that FCEM+CFAM+SMTD jointly deliver SOTA segmentation, topology preservation, and calibrated uncertainty without task interference is unsupported by any tables, baseline comparisons, ablation results, or numerical metrics; the manuscript therefore provides no evidence that the multi-task objectives can be optimized simultaneously across 20 datasets.

    Authors: We acknowledge that the abstract presents high-level claims without direct numerical support. The full manuscript contains Section 4 with quantitative results across all 20 datasets (including mIoU, F1, topology connectivity scores, and uncertainty calibration metrics such as ECE), Table 1 for baseline comparisons, and Table 3 for module ablations demonstrating simultaneous improvements without task interference. To address the concern, we will incorporate key numerical highlights and a brief reference to these results directly into the abstract. revision: yes

  2. Referee: Abstract (SMTD description): no loss formulation, task-weighting scheme, or training procedure is supplied for the joint decoding of masks, skeletons, and uncertainty; without these details it is impossible to assess whether the claimed absence of trade-offs holds or whether per-dataset retuning was required.

    Authors: We agree these implementation details are necessary for assessing the multi-task setup. Section 3.3 describes the SMTD architecture, but we will expand the revision to explicitly include the joint loss formulation (combined Dice+BCE for masks, topology-aware loss for skeletons, and variance-based uncertainty loss), the task-weighting scheme (fixed weights with optional learned balancing), and the end-to-end training procedure with the frozen backbone. This will clarify that no per-dataset retuning was performed beyond the reported five-shot protocol. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical performance claims with no derivation chain or self-referential reductions

full rationale

The paper proposes modules (FCEM, CFAM, SMTD) integrated with a frozen backbone and reports empirical SOTA results across 20 datasets for segmentation, topology, uncertainty, and few-shot adaptation. No mathematical derivation, equations, or loss formulations are presented that could reduce to inputs by construction. Claims rest on experimental outcomes rather than fitted parameters renamed as predictions or self-citation chains. The absence of a derivation chain means no load-bearing steps qualify as circular under the defined patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; the approach rests on standard deep-learning assumptions about foundation-model transfer and multi-task optimization.

pith-pipeline@v0.9.1-grok · 5729 in / 1083 out tokens · 39143 ms · 2026-06-28T02:25:24.929103+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

77 extracted references · 60 canonical work pages · 1 internal anchor

  1. [1]

    Q. Yuan, Y. Shi, M. Li, A review of computer vision-based crack detection methods in civil infrastructure: Progress and challenges, Remote Sensing 16 (16) (2024) 2910.doi:10.3390/rs16162910

    work page doi:10.3390/rs16162910 2024

  2. [2]

    X. Yang, H. Li, Y. Yu, X. Luo, T. Huang, X. Yang, Automatic pixel- level crack detection and measurement using fully convolutional network, Computer-Aided Civil and Infrastructure Engineering 33 (12) (2018) 1090--1109.doi:10.1111/mice.12412

    work page doi:10.1111/mice.12412 2018

  3. [3]

    C. V. Dung, L. D. Anh, Autonomous concrete crack detection using deep fully convolutional neural network, Automation in Construction 99 (2019) 52--58.doi:10.1016/j.autcon.2018.11.028

    work page doi:10.1016/j.autcon.2018.11.028 2019

  4. [4]

    Zhang, F

    L. Zhang, F. Yang, Y. D. Zhang, Y. J. Zhu, Road crack detection using deep convolutional neural network, in: 2016 IEEE International Conference on Image Processing, 2016, pp. 3708--3712. doi:10.1109/ ICIP.2016.7533052

    arXiv 2016

  5. [5]

    Z. Liu, Y. Cao, Y. Wang, W. Wang, Computer vision-based concrete crack detection using u-net fully convolutional networks, Automation in Construction 104 (2019) 129--139. doi:10.1016/j.autcon.2019.04. 005

    work page doi:10.1016/j.autcon.2019.04 2019

  6. [6]

    N. J. Owora, Y. Adu-Gyamfi, A. Aboah, M. Amo-Boateng, Pavesam: Segment anything for pavement distress, Road Materials and Pavement Design 26 (3) (2025) 593--617.doi:10.1080/14680629.2024.2374863

    work page doi:10.1080/14680629.2024.2374863 2025

  7. [7]

    A. Ji, X. Xue, Y. Wang, X. Luo, W. Xue, An integrated approach to automatic pixel-level crack detection and quantification of asphalt pavement, Automation in Construction 114 (2020) 103176. doi:10. 1016/j.autcon.2020.103176

    arXiv 2020

  8. [8]

    Chen, R.-T

    Y.-C. Chen, R.-T. Wu, A. Puranam, Multi-task deep learning for crack segmentation and quantification in rc structures, Automation in Con- struction 166 (2024) 105599.doi:10.1016/j.autcon.2024.105599

    work page doi:10.1016/j.autcon.2024.105599 2024

  9. [9]

    Q. Zou, Y. Cao, Q. Li, Q. Mao, S. Wang, CrackTree: Automatic crack detection from pavement images, Pattern Recognition Letters 33 (3) (2012) 227--238.doi:10.1016/j.patrec.2011.11.004. 52

    work page doi:10.1016/j.patrec.2011.11.004 2012

  10. [10]

    Dontoh, L

    A. Dontoh, L. Sirbaugh, A. Danyo, S. Ivey, B. A. Kyem, A. Aboah, J. Dontoh, An enhanced lightweight model for real-time pavement con- dition index prediction, in: 2025 IEEE International Conference on Future Machine Learning and Data Science (FMLDS), 2025, pp. 273-- 278.doi:10.1109/FMLDS67896.2025.00053

    work page doi:10.1109/fmlds67896.2025.00053 2025

  11. [11]

    B. A. Kyem, J. K. Asamoah, A. Dontoh, A. Danyo, E. K. O. Denteh, A. Aboah, Pavesync: A unified and comprehensive dataset for pavement distress analysis and classification, 2025 IEEE International Conference on Future Machine Learning and Data Science (FMLDS) (2025) 495-- 500. URLhttps://api.semanticscholar.org/CorpusID:284132439

    2025

  12. [12]

    Q. Zou, Z. Zhang, Q. Li, X. Qi, Q. Wang, S. Wang, DeepCrack: Learning hierarchical convolutional features for crack detection, IEEE Transactions on Image Processing 28 (3) (2019) 1498--1512. doi:10.1109/TIP.2018. 2878966

    work page doi:10.1109/tip.2018 2019

  13. [14]

    B. A. Kyem, N. J. Owor, A. Danyo, J. K. Asamoah, E. Denteh, T. Mu- turi, A. Dontoh, Y. Adu-Gyamfi, A. Aboah, Task-specific dual-model framework for comprehensive traffic safety video description and analysis, in: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2025, pp. 5384--5392

    2025

  14. [15]

    J. M. Goo, X. Milidonis, A. Artusi, J. Boehm, C. Ciliberto, Hybrid- segmentor: Hybrid approach for automated fine-grained crack segmen- tation in civil infrastructure, Automation in Construction 170 (2025) 105960.doi:10.1016/j.autcon.2024.105960

    work page doi:10.1016/j.autcon.2024.105960 2025

  15. [16]

    Asadi, P

    B. Asadi, P. Wu, M. Golparvar-Fard, R. Hajj, Controllable flow matching synthesis for generalizable crack segmentation with a 50k image-mask benchmark, Automation in Construction 187 (2026) 106938. doi:10. 1016/j.autcon.2026.106938. 53

    arXiv 2026

  16. [17]

    B. A. Kyem, J. K. Asamoah, T. Addai, E. Denteh, A. Danyo, A. Aboah, A big data approach to pavement distress detection, in: 2025 10th International Conference on Big Data Analytics (ICBDA), 2025, pp. 1--10.doi:10.1109/ICBDA65366.2025.11211362

    work page doi:10.1109/icbda65366.2025.11211362 2025

  17. [18]

    Kirillov, E

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y. Lo, P. Dollar, R. Girshick, Segment anything, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 4015--4026. doi:10.1109/ ICCV51070.2023.00371

    arXiv 2023

  18. [19]

    Oquab, T

    M. Oquab, T. Darcet, T. Moutakanni, H. V. Vo, M. Szafraniec, V. Khali- dov, P. Fernandez, D. Haziza, F. Massa, A. El-Nouby, M. Assran, N. Ballas, W. Galuba, R. Howes, P.-Y. Huang, S.-W. Li, I. Misra, M. Rabbat, V. Sharma, G. Synnaeve, H. Xu, H. Jegou, J. Mairal, P. La- batut, A. Joulin, P. Bojanowski, Dinov2: Learning robust visual features without superv...

    2024

  19. [20]

    Vision-Language Foundation Models for Comprehensive Automated Pavement Condition Assessment

    B. Agyei Kyem, J. Kofi Asamoah, A. Dontoh, A. Aboah, Vision-Language Foundation Models for Comprehensive Automated Pavement Condition Assessment, arXiv e-prints (2026) arXiv:2604.08212 arXiv:2604.08212, doi:10.48550/arXiv.2604.08212

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2604.08212 2026

  20. [21]

    K. Ge, C. Wang, Y. Guo, Y. Tang, Z. Hu, H. Chen, Fine-tuning vision foundation model for crack segmentation in civil infrastruc- tures, Construction and Building Materials 431 (2024) 136573. doi: 10.1016/j.conbuildmat.2024.136573

    work page doi:10.1016/j.conbuildmat.2024.136573 2024

  21. [23]

    J. K. Asamoah, B. Agyei Kyem, N. D. Obeng-Amoako, A. Aboah, Saam-reflectnet: Sign-aware attention-based multitasking frame- work for integrated traffic sign detection and retroreflectivity estimation, Expert Systems with Applications 286 (2025) 128003. doi:https://doi.org/10.1016/j.eswa.2025.128003. 54 URL https://www.sciencedirect.com/science/article/pii...

    work page doi:10.1016/j.eswa.2025.128003 2025

  22. [25]

    Y.-J. Cha, W. Choi, O. Büyüköztürk, Deep learning-based crack damage detection using convolutional neural networks, Computer-Aided Civil and Infrastructure Engineering 32 (5) (2017) 361--378. doi:10.1111/ mice.12263

    2017

  23. [26]

    Gopalakrishnan, S

    K. Gopalakrishnan, S. K. Khaitan, A. Choudhary, A. Agrawal, Deep convolutional neural networks with transfer learning for computer vision- based data-driven pavement distress detection, Construction and Build- ing Materials 157 (2017) 322--330. doi:10.1016/j.conbuildmat.2017. 09.110

    work page doi:10.1016/j.conbuildmat.2017 2017

  24. [27]

    J. Long, E. Shelhamer, T. Darrell, Fully convolutional networks for semantic segmentation, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015, pp. 3431--3440. doi: 10.1109/CVPR.2015.7298965

    work page doi:10.1109/cvpr.2015.7298965 2015

  25. [28]

    In: Proc

    O. Ronneberger, P. Fischer, T. Brox, U-Net: Convolutional networks for biomedical image segmentation, in: Medical Image Computing and Computer-Assisted Intervention, Springer, 2015, pp. 234--241. doi: 10.1007/978-3-319-24574-4_28

    work page doi:10.1007/978-3-319-24574-4_28 2015

  26. [29]

    Dorafshan, R

    S. Dorafshan, R. J. Thomas, M. Maguire, Comparison of deep convolu- tional neural networks and edge detectors for image-based crack detection in concrete, Construction and Building Materials 186 (2018) 1031--1045. doi:10.1016/j.conbuildmat.2018.08.011

    work page doi:10.1016/j.conbuildmat.2018.08.011 2018

  27. [30]

    L.-C. Chen, Y. Zhu, G. Papandreou, F. Schroff, H. Adam, Encoder- decoder with atrous separable convolution for semantic image segmenta- tion, in: Proceedings of the European Conference on Computer Vision, Springer, 2018, pp. 833--851.doi:10.1007/978-3-030-01234-2_49

    work page doi:10.1007/978-3-030-01234-2_49 2018

  28. [31]

    E. Xie, W. Wang, Z. Yu, A. Anandkumar, J. M. Alvarez, P. Luo, Segformer: Simple and efficient design for semantic segmentation with 55 transformers, in: Advances in Neural Information Processing Systems, Vol. 34, 2021, pp. 12077--12090

    2021

  29. [32]

    A ConvNet for the 2020s , booktitle =

    B. Cheng, I. Misra, A. G. Schwing, A. Kirillov, R. Girdhar, Masked- attention mask transformer for universal image segmentation, in: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, 2022, pp. 1280--1289. doi:10.1109/CVPR52688.2022. 00135

    work page doi:10.1109/cvpr52688.2022 2022

  30. [33]

    Y. Liu, J. Yao, X. Lu, R. Xie, L. Li, DeepCrack: A deep hierarchical feature learning architecture for crack segmentation, Neurocomputing 338 (2019) 139--153.doi:10.1016/j.neucom.2019.01.036

    work page doi:10.1016/j.neucom.2019.01.036 2019

  31. [34]

    F. Yang, L. Zhang, S. Yu, D. Prokhorov, X. Mei, H. Ling, Feature pyramid and hierarchical boosting network for pavement crack detection, IEEE Transactions on Intelligent Transportation Systems 21 (4) (2020) 1525--1535.doi:10.1109/TITS.2019.2910595

    work page doi:10.1109/tits.2019.2910595 2020

  32. [35]

    Z. Fan, C. Li, Y. Chen, P. Di Mascio, X. Chen, G. Zhu, G. Loprencipe, Ensemble of deep convolutional neural networks for automatic pavement crack detection and measurement, Coatings 10 (2) (2020) 152. doi: 10.3390/coatings10020152

    work page doi:10.3390/coatings10020152 2020

  33. [36]

    Agyei Kyem, J

    B. Agyei Kyem, J. K. Asamoah, A. Aboah, Context-cracknet: A context-aware framework for precise segmentation of tiny cracks in pave- ment images, Construction and Building Materials 484 (2025) 141583. doi:https://doi.org/10.1016/j.conbuildmat.2025.141583. URL https://www.sciencedirect.com/science/article/pii/ S0950061825017337

    work page doi:10.1016/j.conbuildmat.2025.141583 2025

  34. [37]

    T. Chen, Z. Cai, X. Zhao, C. Chen, X. Liang, T. Zou, P. Wang, Pavement crack detection and recognition using the architecture of segnet, Journal of Industrial Information Integration 18 (2020) 100144. doi:10.1016/j. jii.2020.100144

    work page doi:10.1016/j 2020

  35. [38]

    W. Wang, C. Su, Automatic concrete crack segmentation model based on transformer, Automation in Construction 139 (2022) 104275. doi: 10.1016/j.autcon.2022.104275

    work page doi:10.1016/j.autcon.2022.104275 2022

  36. [39]

    E. A. Shamsabadi, C. Xu, A. S. Rao, T. Nguyen, T. Ngo, D. Dias-da Costa, Vision transformer-based autonomous crack detection on asphalt 56 and concrete surfaces, Automation in Construction 140 (2022) 104316. doi:10.1016/j.autcon.2022.104316

    work page doi:10.1016/j.autcon.2022.104316 2022

  37. [40]

    H. Li, H. Zhang, H. Zhu, K. Gao, H. Liang, J. Yang, Automatic crack detection on concrete and asphalt surfaces using semantic segmentation network with hierarchical transformer, Engineering Structures 307 (2024) 117903.doi:10.1016/j.engstruct.2024.117903

    work page doi:10.1016/j.engstruct.2024.117903 2024

  38. [42]

    Y. Song, J. Kang, Y. Su, S. Zhang, Q. Zhang, Y. Yu, Z. Zhan, W. Zhang, Mambafuse: Cross-scale state space fusion for crack seg- mentation, Developments in the Built Environment (2025) 100751 doi: 10.1016/j.dibe.2025.100751

    work page doi:10.1016/j.dibe.2025.100751 2025

  39. [43]

    Agyei Kyem, J

    B. Agyei Kyem, J. K. Asamoah, E. Denteh, A. Danyo, A. Aboah, Self-supervised multi-scale transformer with attention-guided fusion for efficient crack detection, Automation in Construction 181 (2026) 106591. doi:https://doi.org/10.1016/j.autcon.2025.106591. URL https://www.sciencedirect.com/science/article/pii/ S0926580525006314

    work page doi:10.1016/j.autcon.2025.106591 2026

  40. [44]

    H. Liu, X. Miao, C. Mertz, C. Xu, H. Kong, CrackFormer: Trans- former network for fine-grained crack detection, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 3783--3792.doi:10.1109/ICCV48922.2021.00376

    work page doi:10.1109/iccv48922.2021.00376 2021

  41. [45]

    X. Zuo, Y. Sheng, J. Shen, Y. Shan, Topology-aware mamba for crack segmentation in structures, Automation in Construction 168 (2024) 105845.doi:10.1016/j.autcon.2024.105845

    work page doi:10.1016/j.autcon.2024.105845 2024

  42. [46]

    S. Zhou, C. Canchila, W. Song, Deep learning-based crack segmentation for civil infrastructure: Data types, architectures, and benchmarked performance, Automation in Construction 146 (2023) 104678. doi: 10.1016/j.autcon.2022.104678. 57

    work page doi:10.1016/j.autcon.2022.104678 2023

  43. [47]

    X. Weng, Y. Huang, Y. Li, H. Yang, S. Yu, Unsupervised domain adaptation for crack detection, Automation in Construction 153 (2023) 104939.doi:10.1016/j.autcon.2023.104939

    work page doi:10.1016/j.autcon.2023.104939 2023

  44. [48]

    F. Deng, S. Yang, B. Wang, X. Dong, S. Tian, Ucrack-da: A multi-scale unsupervised domain adaptation method for surface crack segmentation, Remote Sensing 17 (12) (2025) 2101.doi:10.3390/rs17122101

    work page doi:10.3390/rs17122101 2025

  45. [49]

    Zhang, J

    H. Zhang, J. Feng, C. Dai, Z. Tong, Pavement crack segmentation based on synthetic data sets and unsupervised domain adaptation, Journal of Computing in Civil Engineering 39 (6) (2025). doi:10.1061/JCCEE5. CPENG-6675

    work page doi:10.1061/jccee5 2025

  46. [50]

    B. A. Kyem, E. Denteh, J. K. Asamoah, D. Tolliver, A. Aboah, Pavecap: A multimodal framework for comprehensive pavement condition assess- ment with dense captioning and pci estimation, Journal of Transportation Engineering, Part B: Pavements 152 (2) (2026) 04026010. arXiv: https://ascelibrary.org/doi/pdf/10.1061/JPEODX.PVENG-1945, doi:10.1061/JPEODX.PVENG...

    work page doi:10.1061/jpeodx.pveng-1945 2026

  47. [51]

    Y. Guo, Y. Xu, H. Cui, M. Dang, S. Li, Segment anything model- based crack segmentation using low-rank adaption fine-tuning, Structural Health Monitoring 24 (4) (2025).doi:10.1177/14759217241261089

    work page doi:10.1177/14759217241261089 2025

  48. [52]

    Z. Ye, L. J. Lovell, A. Faramarzi, J. Ninic, SAM-based instance segmentation models for the automation of structural damage de- tection, Advanced Engineering Informatics 62 (2024) 102826. doi: 10.1016/j.aei.2024.102826

    work page doi:10.1016/j.aei.2024.102826 2024

  49. [53]

    Z. Zhou, W. Hu, G. Xu, Y. Dong, Self-evolving prompting segment anything model for crack segmentation through data-driven cyclic conversations, Advanced Engineering Informatics 66 (2025) 103626. doi:10.1016/j.aei.2025.103626

    work page doi:10.1016/j.aei.2025.103626 2025

  50. [54]

    S. Shit, J. C. Paetzold, A. Sekuboyina, I. Ezhov, A. Unger, A. Zhylka, J. P. W. Pluim, U. Bauer, B. H. Menze, cldice: A novel topology- preserving loss function for tubular structure segmentation, in: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern 58 Recognition, 2021, pp. 16560--16569. doi:10.1109/CVPR46437.2021. 01629

    work page doi:10.1109/cvpr46437.2021 2021

  51. [55]

    X. Li, S. Yu, S. Xiao, X. Lin, Z. Zhao, A topology-aware segment anything model for domain-invariant crack segmentation, Engineering Applications of Artificial Intelligence 166 (2026) 113688. doi:10.1016/ j.engappai.2025.113688

    arXiv 2026

  52. [56]

    Rathnakumar, Y

    R. Rathnakumar, Y. Pang, Y. Liu, Epistemic and aleatoric uncertainty quantification for crack detection using a bayesian boundary aware convolutional network, Reliability Engineering and System Safety 240 (2023) 109547.doi:10.1016/j.ress.2023.109547

    work page doi:10.1016/j.ress.2023.109547 2023

  53. [57]

    K. R. M. dos Santos, A. G. J. Chassignet, B. G. Pantoja-Rosero, A. Rezaie, O. J. Savary, K. Beyer, Uncertainty quantification for a deep learning models for image-based crack segmentation, Jour- nal of Civil Structural Health Monitoring 15 (4) (2025) 1231--1269. doi:10.1007/s13349-024-00879-6

    work page doi:10.1007/s13349-024-00879-6 2025

  54. [58]

    B. A. Kyem, J. K. Asamoah, Y. Huang, A. Aboah, Weather-adaptive syn- thetic data generation for enhanced power line inspection using stargan, IEEE Access 12 (2024) 193882--193901. doi:10.1109/ACCESS.2024. 3520120

    work page doi:10.1109/access.2024 2024

  55. [59]

    C. Han, H. Yang, T. Ma, S. Wang, C. Zhao, Y. Yang, Crackdiffusion: A two-stage semantic segmentation framework for pavement crack combin- ing unsupervised and supervised processes, Automation in Construction 160 (2024) 105332.doi:10.1016/j.autcon.2024.105332

    work page doi:10.1016/j.autcon.2024.105332 2024

  56. [60]

    Denny, C

    M. Denny, C. Rea, W. Punch, Skeleton-based noise removal algorithm for binary concrete crack image segmentation, Automation in Construction 151 (2023) 104879.doi:10.1016/j.autcon.2023.104879

    work page doi:10.1016/j.autcon.2023.104879 2023

  57. [61]

    B. G. Pantoja-Rosero, K. R. M. dos Santos, K. Beyer, An imple- mentation of the crack topology score with extensions, arXiv preprint arXiv:2601.10762 (2026)

    arXiv 2026

  58. [62]

    T. Y. Zhang, C. Y. Suen, A fast parallel algorithm for thinning digital patterns, Communications of the ACM 27 (3) (1984) 236--239. doi: 10.1145/357994.358023. 59

    work page doi:10.1145/357994.358023 1984

  59. [63]

    Z. Li, T. Zhang, Y. Miao, J. Zhang, M. E. Torbaghan, Y. He, J. Dai, Automated quantification of crack length and width in asphalt pavements, Computer-Aided Civil and Infrastructure Engineering 39 (21) (2024) 3317--3336.doi:10.1111/mice.13344

    work page doi:10.1111/mice.13344 2024

  60. [64]

    Y. Shi, L. Cui, Z. Qi, F. Meng, Z. Chen, Automatic road crack de- tection using random structured forests, IEEE Transactions on In- telligent Transportation Systems 17 (12) (2016) 3434--3445. doi: 10.1109/TITS.2016.2552248

    work page doi:10.1109/tits.2016.2552248 2016

  61. [65]

    Zhang, F

    L. Zhang, F. Yang, Y. D. Zhang, Y. J. Zhu, Road crack detection using deep convolutional neural network, in: Proceedings of the IEEE International Conference on Image Processing, IEEE, 2016, pp. 3708-- 3712.doi:10.1109/ICIP.2016.7533052

    work page doi:10.1109/icip.2016.7533052 2016

  62. [66]

    Eisenbach, R

    M. Eisenbach, R. Stricker, D. Seichter, K. Amende, K. Debes, M. Sessel- mann, D. Ebersbach, U. Stoeckert, H.-M. Gross, How to get pavement distress detection ready for deep learning? A systematic approach, in: International Joint Conference on Neural Networks, IEEE, 2017, pp. 2039--2047.doi:10.1109/IJCNN.2017.7966101

    work page doi:10.1109/ijcnn.2017.7966101 2017

  63. [67]

    Amhaz, S

    R. Amhaz, S. Chambon, J. Idier, V. Baltazart, Automatic crack detection on two-dimensional pavement images: An algorithm based on minimal path selection, IEEE Transactions on Intelligent Transportation Systems 17 (10) (2016) 2718--2729.doi:10.1109/TITS.2015.2477675

    work page doi:10.1109/tits.2015.2477675 2016

  64. [68]

    Q. Mei, M. Gül, M. R. Azim, Densely connected deep neural network considering connectivity of pixels for automatic crack detection, Automa- tion in Construction 110 (2020) 103018. doi:10.1016/j.autcon.2019. 103018

    work page doi:10.1016/j.autcon.2019 2020

  65. [69]

    Saghafi, S

    M. Saghafi, S. M. Asgharzadeh, A. Fathi, A. Hosseini, SUT-Crack: A comprehensive dataset for pavement crack detection across all methods, Data in Brief 50 (2023) 109607.doi:10.1016/j.dib.2023.109607

    work page doi:10.1016/j.dib.2023.109607 2023

  66. [70]

    Ç. F. Özgenel, Concrete crack segmentation dataset (2019). doi:10. 17632/jwsn7tfbrp.1

    2019

  67. [71]

    Liang, X

    D. Liang, X. Kong, J. Liu, X. He, Y. Zhang, X. Xue, L. Xu, Pave- Crack1300: A UAV-acquired pavement crack segmentation dataset (2026).doi:10.17632/8b27pdcxv7.1. 60

    work page doi:10.17632/8b27pdcxv7.1 2026

  68. [72]

    Z. Yao, J. Xu, S. Hou, M. C. Chuah, CrackNex: A few-shot low-light crack segmentation model based on Retinex theory for UAV inspections, in: Proceedings of the IEEE International Conference on Robotics and Automation, IEEE, 2024, pp. 17244--17251. doi:10.1109/ICRA57147. 2024.10610210

    work page doi:10.1109/icra57147 2024

  69. [73]

    Ziya07, UAV-based crack detection dataset, Kaggle, avail- able at: https://www.kaggle.com/datasets/ziya07/ uav-based-crack-detection-dataset(2023)

    2023

  70. [74]

    Z. Zhou, M. M. Rahman Siddiquee, N. Tajbakhsh, J. Liang, UNet++: A nested U-Net architecture for medical image segmentation, in: Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support, Springer, 2018, pp. 3--11. doi:10.1007/ 978-3-030-00889-5_1

    2018

  71. [75]

    L.-C. Chen, G. Papandreou, F. Schroff, H. Adam, Rethinking atrous convolution for semantic image segmentation, arXiv preprint arXiv:1706.05587 (2017)

    Pith/arXiv arXiv 2017

  72. [76]

    H. Zhao, J. Shi, X. Qi, X. Wang, J. Jia, Pyramid scene parsing network, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 2881--2890.doi:10.1109/CVPR.2017.660

    work page doi:10.1109/cvpr.2017.660 2017

  73. [77]

    T.-Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan, S. Belongie, Feature pyramid networks for object detection, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 2117--2125.doi:10.1109/CVPR.2017.106

    work page doi:10.1109/cvpr.2017.106 2017

  74. [78]

    Chaurasia, E

    A. Chaurasia, E. Culurciello, LinkNet: Exploiting encoder representa- tions for efficient semantic segmentation, in: IEEE Visual Communica- tions and Image Processing, IEEE, 2017, pp. 1--4. doi:10.1109/VCIP. 2017.8305148

    work page doi:10.1109/vcip 2017

  75. [79]

    H. Li, P. Xiong, J. An, L. Wang, Pyramid attention network for semantic segmentation, in: Proceedings of the British Machine Vision Conference, 2018

    2018

  76. [80]

    T. Fan, G. Wang, Y. Li, H. Wang, MA-Net: A multi-scale attention network for liver and tumor segmentation, IEEE Access 8 (2020) 179656- -179665.doi:10.1109/ACCESS.2020.3025372. 61

    work page doi:10.1109/access.2020.3025372 2020

  77. [81]

    T. Xiao, Y. Liu, B. Zhou, Y. Jiang, J. Sun, Unified perceptual pars- ing for scene understanding, in: Proceedings of the European Confer- ence on Computer Vision, Springer, 2018, pp. 418--434. doi:10.1007/ 978-3-030-01228-1_26. 62

    2018