Pith. sign in

REVIEW 3 major objections 4 minor 43 references

Training-Free Semantic-Edge Response Decoding of SAM3 for Cross-Domain Infrastructure Crack Segmentation

T0 review · 3 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

Pith's one-line read By decoding SAM3's internal text-conditioned response instead of its mask proposals, SERD raises cross-domain crack segmentation IoU from 54.33% to 58.00% with no training and no target-domain labels.

desk verdict Solid, honest empirical study: decoding SAM3's internal response instead of its proposal masks gains ~3.7 Crack IoU points across unseen datasets; the main flaw is that per-image normalization makes the threshold a relative quantile, so the 'training-free transfer' claim is narrower than the abstract suggests. read the letter →

arxiv 2607.12292 v2 pith:53EQEPJY submitted 2026-07-14 cs.CV

classification cs.CV
keywords cracksegmentationSAM3training-freecross-domainsemanticresponsedecodingSobeledgepriorinfrastructureinspectionzero-shot
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that SAM3's failure to segment thin cracks across domains is largely an output-interface problem: the model's final region proposals suppress, truncate, or distort weak crack evidence that still exists in its internal, text-conditioned pixel response. SERD replaces the proposal interface with that dense response, normalizes it per image, modulates it with a fixed Sobel edge prior, and thresholds it with one global value. With SAM3 frozen and no target-domain annotation, SERD raises average Crack IoU on five unseen datasets from 54.33% to 58.00% compared with native SAM3, and it also improves boundary fidelity, precision, and foreground-area control. The central message is that a foundation model already contains usable crack evidence, and how that evidence is exposed to the downstream task can matter as much as training a better decoder.

What carries the argument

The key object is the internal language-conditioned response R_q: the dense logit field SAM3 generates for the concept 'crack' before it forms discrete proposals. SERD's decoding rule is R_e = Norm(R_q ⊙ (1 + E)), where E is a fixed normalized Sobel gradient magnitude of the grayscale input. The Sobel term acts as a modulator, not an independent detector—it only reweights existing semantic activation, so locations with negligible R_q cannot be recovered from image edges alone. This replaces proposal scoring, filtering, and merging with a single deterministic pixel-level thresholding step, keeping SAM3 fully frozen.

What would settle it

Take a held-out domain with low-contrast or night-time crack images, select SERD's optimal threshold on a source split, freeze it, and measure Crack IoU on the held-out domain. If the threshold sweep on the new domain has no plateau near τ=0.45, or if per-domain optimal thresholds vary widely, then the single-operating-point transfer claim fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that SAM3's internal language-conditioned response R_q—a dense activation field produced for the concept prompt 'crack' before proposal formation—preserves thin, weak, and fragmented cracks that are lost when the model converts that response into scored region proposals. SERD extracts R_q, applies a sigmoid, resizes it, normalizes it per image, modulates it with a normalized Sobel gradient magnitude, and applies a single global threshold τ=0.45. Under this deterministic decoder, average Crack IoU on five unseen test sets reaches 58.00% versus 54.33% for native SAM3; rotating all six datasets through source calibration gives 60.23% mean target-domain IoU and 67.18

Load-bearing premise

The load-bearing premise is that one fixed threshold applied to per-image normalized response values transfers across images, materials, and imaging conditions; if the response distribution shifts between domains, the same τ will systematically over- or under-segment.

Editorial extensions

If this is right

  • Any SAM-family model with a dense text-conditioned response could be repurposed for thin-structure segmentation without training, auxiliary prompting networks, or test-time optimization.
  • Cross-domain crack segmentation can be deployed with one scalar threshold chosen on a single source dataset, with no target-domain annotation needed for calibration.
  • Masks decoded from the internal response are geometrically closer to ground truth: in the paper's macro-average, centerline-length error drops 33.2%, width-distribution error 22.3%, and component-length-distribution error 28.2% relative to native SAM3.
  • Improved boundary and precision behavior makes the decoded masks more usable for downstream image-domain inspection tasks such as length or width estimation, though physical calibration remains a separate step.
  • The decoder adds no meaningful end-to-end latency: SERD averages 550.6 ms per image versus 572.1 ms for native SAM3 in the reported implementation.

Reading between the lines

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

  • The method's ceiling is set by SAM3's representation: because SERD only reweights existing evidence, domains where cracks produce almost no concept-conditioned activation cannot be recovered by any decoder choice—a limit the paper acknowledges.
  • The per-image min-max normalization means the transfer claim rests on one fixed threshold surviving distribution shifts; an adaptive threshold based on response statistics is a natural, testable extension and is one the paper lists as future work.
  • The same response-decoding interface could plausibly extend to other SAM versions or to other thin, line-like defects such as wires, scratches, or vascular structures, provided the internal response retains those patterns—this is a conjecture beyond the paper's crack-only evidence.
  • Because structural calibration only reweights semantic activation, its value is conditional on the semantic response being spatially noisy but locally meaningful; a stronger prior or learned calibration might be needed on highly textured surfaces.
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

3 major / 4 minor

Summary. The paper proposes SERD, a training-free decoder that replaces SAM3's discrete region proposals with the model's internal language-conditioned dense response to the prompt 'crack'. The response is sigmoid-activated, resized, min-max normalized per image (Eq. 4), modulated by a fixed Sobel edge prior (Eq. 8), and thresholded by a single scalar τ. SAM3 remains frozen; no target-domain labels or test-time optimization are used. The method is evaluated on six crack datasets under a unified fixed threshold and under six rotated source-to-target protocols, reporting average Crawl IoU gains over native SAM3 (58.00 vs. 54.33 in the CamCrack789-source protocol; 60.23 vs. 56.95 in the rotated-source protocol), alongside precision, boundary, area, robustness, and latency analyses.

Significance. If the results hold, the paper makes a useful empirical contribution: it shows that for thin, fragmented structures, the internal response of a frozen text-promptable foundation model can be a more informative output interface than its final proposals, and that a simple deterministic decoder can transfer across data domains. The study is unusually transparent: the single threshold is disclosed and source calibration is described, code is promised, the evaluation includes six datasets and rotating source-domain calibration, and the engineering-fidelity analysis in Section 6 provides paired-bootstrap confidence intervals. The main quantitative claim is plausible but not yet fully established, largely because the central IoU comparisons lack uncertainty quantification and the threshold-transfer property is not directly probed on target domains.

major comments (3)
  1. [§3.3 (Eq. 4), §5.5, Table 3] Eq. (4) normalizes each image independently, so τ is not a calibrated confidence but a per-image relative quantile. The paper's central 'one threshold transfers' claim therefore requires evidence that τ remains meaningful when response distributions shift across domains. The rotated-source protocol is helpful, but it reports only aggregate means; the selected thresholds in Table 3 vary from 0.30 to 0.50, and the mean SERD PAR is 1.421, suggesting systematic over-coverage. Fig. 7 sweeps τ only on the CamCrack789 training split. Please add target-domain threshold sweeps or an equivalent analysis, e.g., report for each target dataset the range of τ that stays within 1 IoU point of that target's best τ, and state whether the 3.67-point advantage persists when target-optimal thresholds are used. Without this, the headline gain could depend on the source-calibration choice.
  2. [Tables 1–2, §5.1] The primary IoU comparison is reported as point estimates without confidence intervals, standard errors, or paired significance tests. The gain is averaged over five datasets and varies from +8.94 to +1.19 points across datasets, so the aggregate alone does not establish that the improvement is systematic. The paper already computes paired-bootstrap intervals for the Section 6 morphology metrics; the same procedure should be applied to the main Crack IoU, F1, Precision, and Recall comparisons, with dataset-level and macro-averaged intervals.
  3. [Tables 1 and 3, §3.4] Table 3 shows that direct Response decoding (without Sobel calibration) has essentially the same cross-source mean IoU as SERD (60.15 vs. 60.23), while Table 1 omits Response from the external source-to-target comparison. The Sobel prior's benefit is presented through Precision and Boundary F1, but Tables 4–5 cover only three datasets. Since the method is named and motivated by semantic-edge calibration, include Response in Table 1 or add a cross-domain ablation on the same five target domains, and explicitly state that the headline IoU transfer is attributable to the response interface, not to the Sobel prior.
minor comments (4)
  1. [Abstract/Conclusion] The Boundary F1 value is given as 67.18% in the Abstract and Section 5.1 but 67.17% in the Conclusion. Please harmonize.
  2. [§5.6] The paragraph beginning 'The default decoder contains only...' is duplicated. Remove one instance.
  3. [Table 3] Some rows are difficult to parse because values are run together, e.g., '0.458.4067.83' in the Response row and '0.361.76' in the SERD row. Ensure proper column separation in the final version.
  4. [§5.6, Eq. (12)] Eq. (12) introduces α=1, but α is never used afterward. Either use α in the expression or remove it.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: SERD's only fitted scalar is a source-calibrated threshold; target evaluation is held out and disclosed.

full rationale

The paper does not claim to derive the improvement theoretically; it proposes a deterministic decoding rule (Eqs. 3–9) applied to SAM3's internal language-conditioned response. The only task-level parameter is the scalar threshold τ (Eq. 9), and the paper explicitly calibrates it only on a source training split: 'In source-to-target experiments, a source training split is used only to select the scalar operating threshold, which is then frozen for all target domains' (Section 3.1; see also Table 3 note 'Target-domain labels are not used for threshold selection'). Target-domain test sets are therefore not used to fit the prediction, so the reported 58.00% vs. 54.33% comparison is a genuine transfer evaluation rather than a fitted quantity renamed as a prediction. Equation 4 is per-image min-max normalization, but the paper is explicit that the result 'expresses relative crack-related evidence rather than a calibrated posterior probability' (Section 3.3), so the fixed τ is honestly described as an operating point, not a cross-domain calibrated probability. The Sobel structural prior is fixed ('structural calibration is fixed across images and datasets,' Section 3.4) and is compared against fixed alternatives (Table 6). The claimed benefits are established by held-out benchmarks and ablations, not by a self-citation chain; the only nearby self-citation (CoCo-SAM3, Ref. [30]) is used as related work and is not load-bearing. The Limitations section candidly states that SERD 'cannot reconstruct absent responses'—a boundary condition, not a circular step. No load-bearing premise reduces by construction to the reported outputs.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The central method rests on one scalar threshold and on three domain assumptions about SAM3's internal response and the Sobel prior. There are no invented entities. The paper's evidence for these assumptions is empirical (Figure 1, Table 6) rather than formal.

free parameters (1)
  • response threshold τ = 0.45 (unified); 0.30–0.50 depending on source in Table 3
    Single scalar binarizing the calibrated response; selected on source training split by maximizing foreground IoU (Sec. 4.2). It is the only task-level operating parameter in SERD.
assumptions (3)
  • domain assumption SAM3's grounding module produces a dense concept-conditioned logit field whose sigmoid values reflect crack evidence at pixel level.
    Invoked in Sec. 3.3 (Eq. 3); if the logit field is misaligned or not semantically localized, response decoding has no foundation.
  • domain assumption After per-image normalization, response values are comparable across images and domains, so a single threshold τ transfers.
    Invoked in Eq. (4) and the unified protocol Sec. 4.2; this is the load-bearing calibration assumption.
  • domain assumption Sobel gradient magnitude is a useful fixed prior that improves precision without suppressing valid cracks.
    Invoked in Sec. 3.4 and benchmarked in Table 6; if the prior were domain-specific, the claimed transferability would fail.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Training-Free Semantic-Edge Response Decoding of SAM3 for Cross-Domain Infrastructure Crack Segmentation." pith.science (2026). https://pith.science/paper/53EQEPJY

@misc{pith2026260712292,
  author       = {Pith},
  title        = {Pith review of: Training-Free Semantic-Edge Response Decoding of SAM3 for Cross-Domain Infrastructure Crack Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/53EQEPJY}},
  note         = {Machine review of arXiv:2607.12292}
}
read the original abstract

Cross-project crack segmentation is hindered by variations in materials, imaging conditions, crack morphology, and background interference. Text-promptable foundation models reduce task-specific training, but SAM3's final region proposals may suppress, truncate, or distort weak and fragmented cracks. We propose \textbf{S}emantic-\textbf{E}dge \textbf{R}esponse \textbf{D}ecoding (\textbf{SERD}), a training-free method that replaces the proposal interface with the internal language-conditioned response. SERD normalizes this dense response, calibrates it using a fixed Sobel structural prior, and applies a single threshold to generate the crack mask. SAM3 remains frozen, with no target-domain annotation, auxiliary prompting network, or test-time optimization. When CamCrack789 is used only for threshold selection, SERD achieves 58.00\% average Crack IoU on five unseen datasets, compared with 54.33\% for native SAM3. Across six rotated source-domain settings, SERD obtains 60.23\% mean target-domain IoU and 67.18\% Boundary F1, exceeding SAM3 by 3.27 and 2.70 percentage points. Regional, structural, foreground-area, robustness, and latency analyses show that direct response decoding recovers substantial crack evidence lost during proposal formation, while structural calibration improves precision and boundary localization. The results demonstrate that internal-response decoding provides a simple and transferable interface for cross-domain infrastructure crack segmentation. \textit{Code is available at: \href{https://github.com/xauat-liushipeng/SERD}{GitHub}}.

Figures

Figures reproduced from arXiv: 2607.12292 by the authors.

Figure 1
Figure 1. Native SAM3 output and internal-response decoding. Each row shows the ground truth, [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Native SAM3 inference and the proposed response-decoding interface. Native SAM3 [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Semantic-edge response decoding. The upper row shows the full image and ground truth, [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison between native SAM3 and SERD. Each column presents one [PITH_FULL_IMAGE:figures/full_fig_p019_4.png]
Figure 5
Figure 5. Figure 5: Complete SERD decoding process. From left to right, each row shows the input image, [PITH_FULL_IMAGE:figures/full_fig_p020_5.png]
Figure 6
Figure 6. Figure 6: Local response-profile analysis. The upper and lower plots show profiles sampled along [PITH_FULL_IMAGE:figures/full_fig_p021_6.png]
Figure 7
Figure 7. Figure 7: Sensitivity of native SAM3 and SERD to their respective operating thresholds on the [PITH_FULL_IMAGE:figures/full_fig_p024_7.png]
Figure 8
Figure 8. Figure 8: Representative failure cases of SERD. Weak or diffuse semantic evidence causes missed [PITH_FULL_IMAGE:figures/full_fig_p029_8.png]
Figure 9
Figure 9. Figure 9: Engineering-oriented mask fidelity of SAM3 and SERD. Panels (a)–(c) compare dataset [PITH_FULL_IMAGE:figures/full_fig_p030_9.png]
Figure 10
Figure 10. Figure 10: Cross-dataset consistency of SERD’s morphology-fidelity changes relative to SAM3. [PITH_FULL_IMAGE:figures/full_fig_p031_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 2 linked inside Pith

  1. [1]

    König, M

    J. König, M. Jenkins, M. Mannion, P. Barrie, G. Morison, What’s crack- ing? a review and analysis of deep learning methods for structural crack 33 segmentation, detection and quantification, arXiv preprint arXiv:2202.03714 (2022)

  2. [2]

    Xiang, J

    C. Xiang, J. Guo, R. Cao, L. Deng, A crack-segmentation algorithm fus- ing transformers and convolutional neural networks for complex detection scenarios, Automation in Construction 152 (2023) 104894

  3. [3]

    H. Liu, X. Miao, C. Mertz, C. Xu, H. Kong, Crackformer: Transformer network for fine-grained crack detection, in: Proceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 3783–3792

  4. [4]

    B. G. Pantoja-Rosero, D. Oner, M. Kozinski, R. Achanta, P. Fua, F. Pérez- Cruz, K. Beyer, Topo-loss for continuity-preserving crack detection using deep learning, Construction and Building Materials 344 (2022) 128264

  5. [5]

    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

  6. [6]

    H. Liu, C. Jia, F. Shi, X. Cheng, S. Chen, Scsegamba: Lightweight structure- aware vision mamba for crack segmentation in structures, in: Proceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 29406– 29416

  7. [7]

    Z. Zhao, Z. Ding, P. Niu, W. Sun, F. Guo, Mixercseg: An efficient mixer architecture for crack segmentation via decoupled mamba attention, in: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026, pp. 17969–17978. 34

  8. [8]

    Rostami, P.-H

    G. Rostami, P.-H. Chen, M. S. Hosseini, Segment any crack: Deep semantic segmentation adaptation for crack detection, Journal of Computing in Civil Engineering 40 (3) (2026) 04026020

Show all 43 references
  1. [9]

    Kirillov, E

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Lo, et al., Segment anything, in: Proceed- ings of the IEEE/CVF international conference on computer vision, 2023, pp. 4015–4026

  2. [10]

    N. Ravi, V . Gabeur, Y .-T. Hu, R. Hu, C. Ryali, T. Ma, H. Khedr, R. Rä- dle, C. Rolland, L. Gustafson, et al., Sam 2: Segment anything in images and videos, in: International Conference on Learning Representations, V ol. 2025, 2025, pp. 28085–28128

  3. [11]

    Carion, L

    N. Carion, L. Gustafson, Y .-T. Hu, S. Debnath, R. Hu, D. Suris, C. Ryali, K. V . Alwala, H. Khedr, A. Huang, et al., Sam 3: Segment anything with concepts, in: International Conference on Learning Representations, 2026

  4. [12]

    K. Ge, C. Wang, Y . Guo, Y . Tang, Z. Hu, H. Chen, Fine-tuning vision foun- dation model for crack segmentation in civil infrastructures, Construction and Building Materials 431 (2024) 136573

  5. [13]

    Z. Zhou, W. Hu, G. Xu, Y . Dong, Self-evolving prompting segment anything model for crack segmentation through data-driven cyclic conversations, Ad- vanced Engineering Informatics 68 (2025) 103626

  6. [14]

    Y . Rao, W. Zhao, G. Chen, Y . Tang, Z. Zhu, G. Huang, J. Zhou, J. Lu, Dense- clip: Language-guided dense prediction with context-aware prompting, in: 35 Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 18082–18091

  7. [15]

    Hajimiri, I

    S. Hajimiri, I. Ben Ayed, J. Dolz, Pay attention to your neighbours: Training- free open-vocabulary semantic segmentation, in: Proceedings of the Winter Conference on Applications of Computer Vision, 2025, pp. 5061–5071

  8. [16]

    Kirillov, Y

    A. Kirillov, Y . Wu, K. He, R. Girshick, Pointrend: Image segmentation as rendering, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 9799–9808

  9. [17]

    Y . Yuan, J. Xie, X. Chen, J. Wang, Segfix: Model-agnostic boundary re- finement for segmentation, in: European conference on computer vision, Springer, 2020, pp. 489–506

  10. [18]

    E. A. Shamsabadi, S. M. H. Erfani, C. Xu, D. Dias-da Costa, Efficient semi- supervised surface crack segmentation with small datasets based on consis- tency regularisation and pseudo-labelling, Automation in Construction 158 (2024) 105181

  11. [19]

    Huangfu, Y

    Z. Huangfu, Y . Jiao, F. Wei, G. Shi, H. Dong, A unified approach for weakly supervised crack detection via affine transformation and pseudo label refine- ment, Scientific Reports 15 (1) (2025) 8673

  12. [20]

    Z. Ye, L. Lovell, A. Faramarzi, J. Nini ´c, Sam-based instance segmentation models for the automation of structural damage detection, Advanced Engi- neering Informatics 62 (2024) 102826

  13. [21]

    Y . Wang, J. He, S. Yu, Crack-edgesam self-prompting crack segmentation system for edge devices., CoRR (2024). 36

  14. [22]

    Radford, J

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sas- try, A. Askell, P. Mishkin, J. Clark, et al., Learning transferable visual mod- els from natural language supervision, in: International conference on ma- chine learning, PmLR, 2021, pp. 8748–8763

  15. [23]

    Lüddecke, A

    T. Lüddecke, A. Ecker, Image segmentation using text and image prompts, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), 2022, pp. 7086–7096

  16. [24]

    C. Zhou, C. C. Loy, B. Dai, Extract free dense labels from clip, in: European conference on computer vision, Springer, 2022, pp. 696–712

  17. [25]

    M. Xu, Z. Zhang, F. Wei, H. Hu, X. Bai, Side adapter network for open- vocabulary semantic segmentation, in: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, 2023, pp. 2945–2954

  18. [26]

    Liang, B

    F. Liang, B. Wu, X. Dai, K. Li, Y . Zhao, H. Zhang, P. Zhang, P. Vajda, D. Marculescu, Open-vocabulary semantic segmentation with mask-adapted clip, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 7061–7070

  19. [27]

    S. Cho, H. Shin, S. Hong, A. Arnab, P. H. Seo, S. Kim, Cat-seg: Cost aggre- gation for open-vocabulary semantic segmentation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 4113–4123

  20. [28]

    Zhang, F

    D. Zhang, F. Liu, Q. Tang, Corrclip: Reconstructing patch correlations in clip for open-vocabulary semantic segmentation, in: Proceedings of the 37 IEEE/CVF International Conference on Computer Vision, 2025, pp. 24677– 24687

  21. [29]

    X. Xuan, Z. Deng, K.-L. Ma, Reme: A data-centric framework for training- free open-vocabulary segmentation, in: Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision, 2025, pp. 20954–20965

  22. [30]

    Y . Chen, B. Yang, S. Liu, J. Wang, Coco-sam3: Harnessing con- cept conflict in open-vocabulary semantic segmentation, arXiv preprint arXiv:2604.19648 (2026)

  23. [31]

    F. Yang, L. Zhang, S. Yu, D. Prokhorov, X. Mei, H. Ling, Feature pyra- mid and hierarchical boosting network for pavement crack detection, IEEE transactions on intelligent transportation systems 21 (4) (2019) 1525–1535

  24. [32]

    G. Zhu, J. Liu, Z. Fan, D. Yuan, P. Ma, M. Wang, W. Sheng, K. C. Wang, A lightweight encoder–decoder network for automatic pavement crack detec- tion, Computer-Aided Civil and Infrastructure Engineering 39 (12) (2024) 1743–1765

  25. [33]

    Katsamenis, E

    I. Katsamenis, E. Protopapadakis, N. Bakalos, A. Varvarigos, A. Doulamis, N. Doulamis, A. V oulodimos, A few-shot attention recurrent residual u-net for crack segmentation, in: International Symposium on Visual Computing, Springer, 2023, pp. 199–209

  26. [34]

    H. Liu, C. Jia, F. Shi, X. Cheng, M. Wang, S. Chen, Staircase cascaded fu- sion of lightweight local pattern recognition and long-range dependencies for structural crack segmentation, arXiv preprint arXiv:2408.12815 1 (5) (2024) 7. 38

  27. [35]

    C. Benz, V . Rodehorst, Omnicrack30k: A benchmark for crack segmentation and the reasonable effectiveness of transfer learning, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 3876–3886

  28. [36]

    M. Pu, Y . Huang, Q. Guan, H. Ling, Rindnet: Edge detection for disconti- nuity in reflectance, illumination, normal and depth, in: Proceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 6879– 6888

  29. [37]

    D. Kang, M. Cho, In defense of lazy visual grounding for open-vocabulary semantic segmentation, in: European Conference on Computer Vision, Springer, 2024, pp. 143–164

  30. [38]

    J. Li, Y . Lu, Y . Zhang, Y . Xie, F. Wang, Y . Xie, Y . Qu, Target refocusing via attention redistribution for open-vocabulary semantic segmentation: An explainability perspective, in: Proceedings of the AAAI Conference on Ar- tificial Intelligence, V ol. 40, 2026, pp. 6181–6189

  31. [39]

    Cheng, R

    B. Cheng, R. Girshick, P. Dollár, A. C. Berg, A. Kirillov, Boundary iou: Improving object-centric image segmentation evaluation, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 15334–15342

  32. [40]

    S. Shit, J. C. Paetzold, A. Sekuboyina, I. Ezhov, A. Unger, A. Zhylka, J. P. Pluim, U. Bauer, B. H. Menze, cldice-a novel topology-preserving loss func- tion for tubular structure segmentation, in: Proceedings of the IEEE/CVF 39 conference on computer vision and pattern recogn...

  33. [41]

    Canny, A computational approach to edge detection, IEEE Transactions on pattern analysis and machine intelligence (6) (1986) 679–698

    J. Canny, A computational approach to edge detection, IEEE Transactions on pattern analysis and machine intelligence (6) (1986) 679–698

  34. [42]

    D. Marr, E. Hildreth, Theory of edge detection, Proceedings of the Royal Society of London. Series B. Biological Sciences 207 (1167) (1980) 187– 217

  35. [43]

    R. M. Haralick, S. R. Sternberg, X. Zhuang, Image analysis using math- ematical morphology, IEEE transactions on pattern analysis and machine intelligence (4) (1987) 532–550. 40

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.