Pith. sign in

REVIEW 2 major objections 4 minor 79 references

Under a leak-free cross-validated stacking protocol that closes both the tuning-leak and synthetic-descendant leak, a five-backbone sand-boil ensemble (IoU 0.681) does not beat the best single model (0.694), while the modernized single netw

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 02:37 UTC pith:62BSCZRS

load-bearing objection A careful, honest negative-result study of stacking and synthetic augmentation for a scarce-data segmentation task — worth refereeing, but the architecture-selection split must be clarified before the headline numbers can be trusted. the 2 major comments →

arxiv 2607.25367 v1 pith:62BSCZRS submitted 2026-07-28 cs.CV

Leak-Free Cross-Validated Stacking with Per-Architecture Calibration for Sand-Boil Segmentation in Earthen Levees

Padam Jung Thapa , Anav Katwal , Ayon Dey , Abdullah Bin Naeem , Steve Sloan , Kendall Niles , Md Tamjidul Hoque This is my paper
classification cs.CV
keywords sand boil segmentationearthen levee inspectioncross-validated stackingtemperature calibrationsynthetic data leakageparent-image filterlabel fidelity filteringmask-conditioned synthesis
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Sand boils are early visible signs of hidden erosion inside earthen levees, and deep networks are taught to spot them in inspection photos — but annotated examples are rare, and two common workarounds quietly inflate reported accuracy: tuning ensemble weights on the images that will later be scored, and training on synthetic images that descend from the very photographs held out for testing. This paper builds a segmentation pipeline that closes both loopholes: every synthetic image points to its real parent, a per-fold filter blocks descendants of held-out parents, and five backbones are combined only through out-of-fold predictions, calibrated by one temperature scalar per architecture. On a held-out test set, the modernized single model reaches IoU 0.707 versus 0.608 for the published baseline re-evaluated on the same split. Under the leak-free protocol, however, the calibrated five-model stack reaches only 0.681 against 0.694 for the strongest member — the ensemble does not help, and the paper attributes this to a mean pairwise error correlation of 0.894 among members. The point a sympathetic reader takes away is that honest measurement, not the score itself, is the contribution: it separates transfers that help scarce-data defect segmentation (a modern encoder, label-fidelity-filtered synthetic data) from those that do not (stacking, hard-negative mining, higher resolution).

Core claim

This paper claims that when both common leakage paths are closed, the measured value of ensembling in a scarce-data defect-segmentation regime is negative: the calibrated stack (0.681 IoU) falls short of the best fold-averaged single model (0.694 IoU), a result reproduced by eight meta-learner families and explained by a mean pairwise member error correlation of 0.894. It further claims that the modernized SandBoilNet (ConvNeXt-S encoder with scSE skip attention) reaches 0.707 IoU against 0.608 for the published original re-evaluated on the same held-out split, and that filtering the synthetic augmentation pool for label fidelity lifts the champion to 0.718. The author's own account is that

What carries the argument

The central object is the leak-free cross-validated stacking protocol: each of five encoder–decoder backbones is trained once per fold, predicts only on its held-out fold, and the resulting out-of-fold cube is recalibrated by one temperature scalar per architecture and combined by a per-pixel logistic-regression meta-learner fitted exclusively on those out-of-fold predictions. The companion mechanism is the parent-image filter: each synthetic record carries a source_image_id pointer to its real parent, and a per-fold set operation excludes any synthetic descendant whose parent sits in that fold's validation split, closing the synthetic-descendant leakage path in O(1). A second, complementary

Load-bearing premise

The synthetic training masks are produced by running the published SandBoilNet probability map through a convex-hull cleaning pipeline, and the label-fidelity admission filter checks a held-out segmenter's agreement with those same auto-generated labels — if the baseline's systematic errors are passed on to its successor, the observed synthetic-data gain (0.718 vs 0.707) may reflect reinforcement of those biases rather than genuinely new, correct signal.

What would settle it

Re-train the champion on a synthetic pool whose labels are generated by an independent segmenter or human-verified masks, using the same diffusion source and filter logic; if the IoU gain over the 0.707 control disappears or reverses, the filter was selecting for agreement with the baseline's own errors, not for label fidelity. Concretely, compare the current filtered-pool result against a pool filtered on human-verified IoU rather than the held-out segmenter's agreement with pseudo-labels.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the leak-free protocol is correct, any reported gain from weighted-averaging or naive synthetic augmentation in this domain must be re-read as partly an artifact of tuning on the scored split or of parent leakage.
  • Stacking's measured failure is attributed to member correlation (mean pairwise error correlation 0.894; five backbones share only three encoders), so future ensemble gains will require genuinely independent feature extractors rather than more decoders on the same encoders.
  • Label-fidelity-filtered synthetic data lifts the champion from 0.707 to 0.718 IoU, while unfiltered pools do not; the action is in filtering, not in the generator configuration, since sweeping guidance scale over four values produced no difference.
  • The negative hard-negative result (IoU −0.010, false-positive rate up) suggests that confusion-score-weighted synthetic negatives require a closed loop that re-scores against the model being trained, not the frozen baseline.
  • The modernized encoder (ConvNeXt-S + scSE) transfers to a second scarce-data defect dataset, indicating the architecture ranking is not an artifact of this particular split.

Where Pith is reading between the lines

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

  • A natural extension is to apply the same parent-image filter and out-of-fold discipline to other synthetic-augmentation regimes in medical or infrastructure imaging, where diffusion-generated descendants of real patient or structure images are becoming common; the source_image_id field is cheap to add at generation time.
  • The error-correlation diagnosis suggests a testable recipe: decorrelate by pairing each backbone with a distinct pretrained encoder, and check whether the stacking shortfall (−0.013 IoU) shrinks further than the −0.008 observed when only two encoders were swapped.
  • Since the label-fidelity filter leans on the same baseline pseudo-labels that seed the synthetic masks, an editorial caution is warranted: the observed synthetic-data gain may partly reinforce the baseline's biases; re-validating with human-checked masks or a second independent annotator would disentangle genuine signal from inherited error.
  • MaskCN's zero-annotation labelling, if it generalises, would decouple dataset size from annotation budget for any segmentation task with a well-defined mask shape, potentially shifting the scarce-data bottleneck from labelling to mask design.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 4 minor

Summary. The paper proposes a sand-boil segmentation framework built around a 'leak-free' five-fold cross-validated stacking protocol with per-architecture temperature calibration, a per-fold filter that excludes synthetic descendants of held-out parents, a synthetic hard-negative phase, and a mask-conditioned synthesis route (MaskCN). On 199 training / 46 test images, the Updated SandBoilNet reaches 0.707 IoU versus 0.608 for the published baseline; the calibrated stack reaches 0.681 IoU versus 0.694 for the best fold-averaged member; a label-fidelity-filtered synthetic pool lifts the champion to 0.718 versus a 0.707 control. The paper reports that stacking does not improve on the best single model and that the hard-negative phase does not help, framing these as honest negative results. The central methodological claims are (1) that the protocol closes two named leakage paths and therefore gives unbiased ensemble measurements, and (2) that the updated backbone improves on the published baseline.

Significance. If the protocol is valid, the paper makes a useful contribution to scarce-data defect segmentation: a carefully specified out-of-fold stacking pipeline, a simple parent-tagging filter for synthetic data, per-architecture calibration, extensive meta-learner ablations, bootstrap uncertainty intervals, a cross-dataset check on DeepCrack, and a reproducible artifact with code, checkpoints, and manifests. The reporting of negative results is candid and the sensitivity analysis is more thorough than typical for this area. However, the central validation is undermined by the fact that architecture selection is performed on the held-out test split, which contradicts the 'leak-free/unbiased by construction' claim and calls the headline numbers into question. The synthetic-label circularity is a further load-bearing concern for the synthetic-augmentation contribution.

major comments (2)
  1. [Section V-A, Table I; Algorithm 1] The family champions are selected on the held-out test set: Section V-A states that the search keeps 'the best performer in each family' and that 'Table I reports the resulting family champions on the held-out test set'; Table I's caption identifies a single 85/15 split, and Table VII uses the same kind of split to choose the scSE attention variant. The same 46-image test set is later used to report the champion's 0.707 IoU and the stack-vs-single comparison (0.681 vs 0.694). Algorithm 1's assertion that 'the held-out test set is first touched at line 6' is therefore false for the overall pipeline: model selection touched it earlier. On 46 test images, selecting among 11 architectures plus attention variants on test IoU can inflate the reported numbers and can change which member is 'best' or whether stacking closes the gap. Please state clearly whether the 85/15 split is an internal val
  2. [Section III-B; Section V-G] The synthetic positive masks are generated by thresholding the public SandBoilNet probability map through Convex_Hull_Annotator, and the label-fidelity filter admits a synthetic pair only if a held-out segmenter reaches IoU >= 0.70 against that same pseudo-label. The filtered-pool gain (0.718 vs 0.707, Section V-G) may therefore reflect reinforcement of the baseline's biases rather than genuinely new signal. This is load-bearing for the synthetic-augmentation contribution. Please provide independent validation of a sample of synthetic masks (manual review, or agreement with a segmenter trained without SandBoilNet-derived labels), and report the rejection rate of the fidelity filter. Re-deriving labels with 'the strongest segmenter available here' does not remove the concern if that segmenter inherits the same training labels.
minor comments (4)
  1. [Tables I and II] The 'best single' model differs between Table I (Updated SandBoilNet, 0.707) and Table II (SegFormer, 0.694) because one is a single-split comparison and the other uses fold-averaged predictions. This is stated in captions but should be flagged prominently in the main text to avoid confusion.
  2. [Section III-B] The label-fidelity admission criterion ('all five cross-validation folds of a held-out segmenter reach IoU 0.70 against the image's own mask') is ambiguous: which segmenter, and what does 'all five folds' mean for a held-out segmenter? Please clarify.
  3. [Figure 5] The IoU values shown in Figure 5 come from Table I, which is the architecture-search score on the test split. The figure should be labeled as such so readers do not mistake these for unbiased estimates.
  4. [Section V-G] The three-seed filtered-pool gain of 0.012 ± 0.012 is appropriately described as suggestive, but reporting the individual per-seed values would strengthen this transparency.

Circularity Check

2 steps flagged

Two leaky/circular steps: architecture selection on the held-out test set and a label-fidelity filter defined against baseline-derived pseudo-labels; the real-only stack negative result retains independent content.

specific steps
  1. other [Section V-A (Table I) and Algorithm 1 comment]
    "Table I reports the resulting family champions on the held-out test set, with the original SandBoilNet shown for reference. ... Both the temperature scalars and the meta-learner are fitted only on out-of-fold predictions (lines 5 and 6); the held-out test set is first touched at line 6."

    The architecture search and the scSE attention-variant ablation (Table VII) select models using the same 46-image split later used for the headline 0.707 and the 0.681-vs-0.694 stack comparison. Algorithm 1's assertion that the held-out test set is first touched at line 6 is therefore false for the overall pipeline: model selection touched it earlier. Reporting test IoU after selecting the best of 11 architectures and attention variants on that test set makes the reported numbers a selected maximum, not an unbiased estimate. Even if the 'single 85/15 split' is an internal validation split, the text as written identifies the held-out test set as the selection ground truth.

  2. self definitional [Section III-B (synthetic labels and label-fidelity filter); Section V-G]
    "The synthetic ground-truth mask is produced by running the synthetic image through the Convex_Hull_Annotator pipeline of [3], [62], which thresholds the public SandBoilNet probability map... Candidates are admitted only where all five cross-validation folds of a held-out segmenter reach an intersection-over-union of 0.70 against the image's own mask, a label-fidelity filter..."

    The 'image's own mask' is the pseudo-label obtained by thresholding the public SandBoilNet probability map. The label-fidelity admission filter therefore measures agreement with the same baseline's output, not with independent ground truth. A pool selected for agreement with baseline-derived labels is self-consistent by construction, so the 0.718-vs-0.707 gain attributed to 'filtered for label fidelity' may partly reflect reinforcement of the baseline's systematic errors rather than genuinely correct new signal. The final evaluation on human-labeled test images keeps this from being a fully forced circularity.

full rationale

The paper's central real-only stacking result (calibrated stack 0.681 vs best single 0.694, replicated across eight meta-learners) is an empirical finding with independent content: it is not obtained by fitting a parameter to the same target being predicted, and the out-of-fold protocol is a legitimate construction. However, the paper's 'leak-free' claim is overstated because architecture selection and the scSE attention choice are reported against the held-out test set before the final numbers are measured on that same set. In addition, the synthetic-data contribution relies on a label-fidelity filter whose ground truth is itself derived from the public SandBoilNet probability map, making the 'fidelity' criterion self-referential. The companion-paper self-citation [5] is also load-bearing for the synthetic augmentation claims, though not for the real-only stack conclusion. These issues warrant a moderate circularity/leakage score; they do not reduce the entire derivation to its inputs.

Axiom & Free-Parameter Ledger

7 free parameters · 6 axioms · 0 invented entities

No new physical or ontological entities are introduced; MaskCN is a synthesis method rather than a postulated entity. The central claims rest on standard stacking theory plus several domain assumptions about pseudo-label quality, test-set representativeness, transfer learning, and generator provenance.

free parameters (7)
  • Per-architecture temperature T_a = T_a = 1.10, 1.50, 1.70, 1.20, 1.10 (Table VI)
    Fitted by grid search on 1M out-of-fold pixels (Eq. 2); recalibrates each backbone's probabilities before stacking.
  • Meta-learner weights w and bias b = Normalized weights 0.201, 0.225, 0.273, 0.152, 0.149; bias not stated
    Fitted by l2-regularized logistic regression on 2M out-of-fold pixels (Eq. 3); these determine the stacked output.
  • Decision threshold tau = Not reported
    Tuned at the OOF operating point; the paper itself notes in Appendix G that tau is a fitted quantity.
  • Synthetic admission IoU threshold = 0.70
    Hand-selected label-fidelity filter used to admit synthetic candidates into each training pool.
  • Loss weights w_bce and w_dice = 0.5 each
    Hand-set in Eq. 5 to balance binary cross-entropy and Dice losses.
  • Training hyperparameters (learning rates, weight decay, warmup, early stopping) = LR 6e-5 to 1e-4; weight decay 1e-4 to 1e-2; patience 30
    Per-architecture choices listed in Appendix E; these materially affect model quality.
  • Number of folds K and partition seed = K=5, seed=42
    The fixed fold partition determines the out-of-fold cube and all fitted quantities downstream.
axioms (6)
  • standard math Wolpert stacking validity: a meta-learner fitted only on out-of-fold predictions is unbiased
    Used throughout Section III-F; assumes no additional leakage path besides the ones the filter addresses.
  • domain assumption Synthetic pseudo-labels produced by the public SandBoilNet probability map are accurate enough to train on
    Section III-B generates synthetic ground-truth masks via Convex_Hull_Annotator thresholding of SandBoilNet outputs; systematic errors would bias the synthetic-data gains.
  • domain assumption ImageNet-pretrained encoders transfer to levee-inspection imagery
    All encoders start from ImageNet weights and the paper states this is decisive at 199-image scale (Section III-D).
  • domain assumption The 46-image held-out split is representative and sufficiently clean
    All headline comparisons rely on this split; the bootstrap CI width of roughly ±0.05 IoU shows sampling uncertainty is material.
  • domain assumption The diffusion generator's training set excluded the held-out test images
    Unstated in the paper; the parent filter only blocks direct synthetic descendants, not indirect leakage through the DreamBooth LoRA or ControlNet training data.
  • domain assumption MaskCN's mask-ControlNet reliably renders a sand boil inside the target mask
    MaskCN's 'label by construction' claim depends on the generator faithfully realizing arbitrary target masks (Section III-C).

pith-pipeline@v1.3.0-alltime-deepseek · 33212 in / 14000 out tokens · 137736 ms · 2026-08-01T02:37:26.878585+00:00 · methodology

0 comments
read the original abstract

Sand boils, points where water seeping beneath an earthen levee re-emerges at the surface, are early warnings of internal erosion, and deep segmentation networks are increasingly used to find them in inspection photographs. Annotated examples are scarce, and two common ways of working around that scarcity quietly inflate reported accuracy: tuning ensemble weights on the same images later used to score them, and training on synthetic images derived from the very photographs held out for testing. We present a sand-boil segmentation framework that closes both loopholes. Every synthetic image carries a pointer to its real parent, and a per-fold filter excludes any image whose parent is held out; five encoder-decoder backbones are trained under five-fold cross-validation, calibrated by one temperature scalar each, and combined by a per-pixel meta-learner fitted only on out-of-fold predictions. On the held-out test set the proposed Updated SandBoilNet reaches an intersection-over-union of 0.707 over three seeds, against 0.608 for the published original re-evaluated on the same split. Under the stacking protocol the calibrated stack reaches 0.681 against 0.694 for the strongest fold-averaged member, so it does not improve on the best single model; eight meta-learner families reproduce that outcome, which we trace to a mean pairwise error correlation of 0.894 among members. A synthetic pool filtered for label fidelity lifts the champion to 0.718 over three seeds against a 0.707 control. We also introduce a mask-conditioned synthesis route that makes the conditioning mask the label by construction, giving labelled training images at zero annotation cost.

Figures

Figures reproduced from arXiv: 2607.25367 by Abdullah Bin Naeem, Anav Katwal, Ayon Dey, Kendall Niles, Md Tamjidul Hoque, Padam Jung Thapa, Steve Sloan.

Figure 1
Figure 1. Figure 1: The segmentation framework at a glance. Real and synthetic data pass through the leak-free per-fold filter; five backbones are trained [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Representative real sand boils (top) with their hand-annotated ground-truth masks overlaid (bottom). The dataset spans a wide [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Real sand boils (top) beside their best-fidelity synthetic descendants (bottom), produced by the companion generation paper [5] and [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Mask-conditioned synthesis (MaskCN). A mask-ControlNet [41] trained jointly with the sand-boil DreamBooth-LoRA [42], [43] [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The five backbone families retained as base learners, with [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Training phase. The real training set is split into five [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The leak-free per-fold filter as a constraint on the fold [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Test-time inference. For each of the 46 held-out real test [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Synthetic hard-negative mining as a curriculum. An adapter [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 11
Figure 11. Figure 11: Per-image IoU distributions for the best single base learner, [PITH_FULL_IMAGE:figures/full_fig_p014_11.png] view at source ↗
Figure 10
Figure 10. Figure 10: Ensembling strategies against the best single model at each [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
Figure 12
Figure 12. Figure 12: Meta-learner weights coloured by calibration temperature: [PITH_FULL_IMAGE:figures/full_fig_p015_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Calibration reliability of the most over-confident architec [PITH_FULL_IMAGE:figures/full_fig_p016_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Qualitative comparison on held-out real test images, drawn from those the proposed model leads and spanning its IoU range from [PITH_FULL_IMAGE:figures/full_fig_p017_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Precision–recall (top) and ROC (bottom) curves across the [PITH_FULL_IMAGE:figures/full_fig_p020_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: MaskCN as a diversity engine. Eighteen distinct, fully labelled sand boils synthesised from a single mask bank, ordered small-to-large [PITH_FULL_IMAGE:figures/full_fig_p021_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: The heavy train-time augmentation pipeline (the [PITH_FULL_IMAGE:figures/full_fig_p021_17.png] view at source ↗

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

79 extracted references · 3 canonical work pages

  1. [1]

    Investigation of the performance of the new orleans flood protection systems in hurricane katrina on august 29, 2005,

    R.Seed,R.Bea,A.Athanasopoulos-Zekkosetal.,“Investigation of the performance of the new orleans flood protection systems in hurricane katrina on august 29, 2005,” Independent Levee Investigation Team, University of California Berkeley, Tech. Rep., 2006

    2005

  2. [2]

    Deep learning approach for accurate segmen- tation of sand boils in levee systems,

    M. Panta, M. T. Hoque, K. N. Niles, J. Tom, M. Abdelguerfi, and M. Flanagin, “Deep learning approach for accurate segmen- tation of sand boils in levee systems,”IEEE Access, vol. 11, pp. 126263–126282, 2023

    2023

  3. [3]

    Toward robust semantic segmentation in levee in- frastructure monitoring: Enhancing accuracy with high-fidelity synthetic data and ensemble learning,

    P. J. Thapa, “Toward robust semantic segmentation in levee in- frastructure monitoring: Enhancing accuracy with high-fidelity synthetic data and ensemble learning,” Master’s Thesis, Uni- versity of New Orleans, New Orleans, LA, USA, 2025, uni- versity of New Orleans Theses and Dissertations No. 3231. https://scholarworks.uno.edu/td/3231/

    2025

  4. [4]

    Stacked generalization,

    D. H. Wolpert, “Stacked generalization,”Neural Networks, vol. 5, no. 2, pp. 241–259, 1992

    1992

  5. [5]

    Multi-conditioned diffusion synthesis of sand boils for low-resource earthen-levee inspection,

    P. J. Thapa, A. B. Naeem, A. Dey, A. Katwal, and M. T. Hoque, “Multi-conditioned diffusion synthesis of sand boils for low-resource earthen-levee inspection,”arXiv preprint arXiv:2607.08794, 2026. 23

    Pith/arXiv arXiv 2026

  6. [6]

    Deeplearning approaches in pavement distress identification: A review,

    S.Guan,H.Liu,H.R.Pourreza,andH.Mahyar,“Deeplearning approaches in pavement distress identification: A review,”arXiv preprint arXiv:2308.00828, 2023

    Pith/arXiv arXiv 2023

  7. [7]

    IterLUNet: Deep learning architecture for pixel-wise crack detection in levee systems,

    M. Panta, M. T. Hoque, M. Abdelguerfi, and M. C. Flanagin, “IterLUNet: Deep learning architecture for pixel-wise crack detection in levee systems,”IEEE Access, vol. 11, pp. 12249– 12262, 2023

    2023

  8. [8]

    Application of deeplearningforsegmentingseepagesinleveesystems,

    M. Panta, P. J. Thapa, M. T. Hoque, K. N. Niles, S. Sloan, M. Flanagin, K. Pathak, and M. Abdelguerfi, “Application of deeplearningforsegmentingseepagesinleveesystems,”Remote Sensing, vol. 16, no. 13, p. 2441, 2024

    2024

  9. [9]

    Application of deep learning for segmenting seepages in levee systems,

    M. Panta, P. J. Thapa, M. T. Hoque, K. N. Niles, S. Sloan, M. Flanagin, K. Pathak, and M. Abdelguerfi, “Application of deep learning for segmenting seepages in levee systems,” Engineer Research and Development Center (US), Vicksburg, MS, Miscellaneous Paper ERDC MP-24-8, Nov. 2024. [Online]. Available: https://hdl.handle.net/11681/49453

    2024

  10. [10]

    Imbalance-aware culvert-sewer defect segmentation using an enhanced feature pyramid network,

    R. Alshawi, M. M. Ferdaus, M. Abdelguerfi, K. Niles, K. Pathak, and S. Sloan, “Imbalance-aware culvert-sewer defect segmentation using an enhanced feature pyramid network,”arXiv preprint arXiv:2408.10181, 2024. [Online]. Available: https://arxiv.org/abs/2408.10181

    Pith/arXiv arXiv 2024

  11. [11]

    OilSpillNet: A deep learning framework for accurate detection and segmentation of marine oil spills from imagery,

    M. Elsheref, P. J. Thapa, A. Katwal, A. B. Naeem, M. A. Tarr, and M. T. Hoque, “OilSpillNet: A deep learning framework for accurate detection and segmentation of marine oil spills from imagery,”Journal of Environmental Chemical Engineer- ing, vol. 14, no. 2, p. 121714, Apr. 2026

    2026

  12. [12]

    From manual to uav-based inspection: Efficient detection of levee seepage hazards driven by thermal infrared image and deep learning,

    B. Chen, Q. Duan, and L. Luo, “From manual to uav-based inspection: Efficient detection of levee seepage hazards driven by thermal infrared image and deep learning,”International Journal of Disaster Risk Reduction, vol. 114, p. 104982, 2024. [Online]. Available: https://doi.org/10.1016/j.ijdrr.2024.104982

    arXiv 2024

  13. [13]

    Rapid and automatic uav detection of river embankment piping,

    Q. Duan, B. Chen, and L. Luo, “Rapid and automatic uav detection of river embankment piping,”Water Resources Research, vol. 61, no. 2, 2025. [Online]. Available: https: //doi.org/10.1029/2024WR038931

    work page doi:10.1029/2024wr038931 2025

  14. [14]

    Automated detection of embankment piping and leakage hazards using uav visible light imagery: A frequency-enhanced deep learning approach,

    Z. Wang, R. Liet al., “Automated detection of embankment piping and leakage hazards using uav visible light imagery: A frequency-enhanced deep learning approach,”Remote Sensing, vol. 17, no. 21, p. 3602, 2025. [Online]. Available: https: //doi.org/10.3390/rs17213602

    work page doi:10.3390/rs17213602 2025

  15. [15]

    Pruned u-net with multi-scale feature fusion and attention for real-time uav remote sensing of levee defects,

    B. Wu, B. Chen, X. Jiang, and Z. Liu, “Pruned u-net with multi-scale feature fusion and attention for real-time uav remote sensing of levee defects,”Scientific Reports, vol. 15, p. 42354, 2025. [Online]. Available: https://doi.org/10.1038/ s41598-025-26431-0

    2025

  16. [16]

    Machine learning applications in detecting sand boils from images,

    A. Kuchi, M. T. Hoque, M. Abdelguerfi, and M. C. Flanagin, “Machine learning applications in detecting sand boils from images,”Array, vol. 3, p. 100012, 2019

    2019

  17. [17]

    A depth-wise separable U-Net architecture with multiscale filters to detect sinkholes,

    R. Alshawi, M. T. Hoque, and M. C. Flanagin, “A depth-wise separable U-Net architecture with multiscale filters to detect sinkholes,”Remote Sensing, vol. 15, no. 5, p. 1384, 2023

    2023

  18. [18]

    Earthen levee monitoring with synthetic aperture radar,

    J. V. Aanstoos, K. Hasan, C. G. O’Hara, S. Prasad, L. Dab- biru, M. Mahrooghy, B. Gokaraju, and R. Nobrega, “Earthen levee monitoring with synthetic aperture radar,” in2011 IEEE Applied Imagery Pattern Recognition Workshop (AIPR), 2011

    2011

  19. [19]

    A survey on deep learning techniques for image and video semantic seg- mentation,

    A. Garcia-Garcia, S. Orts-Escolano, S. Opreaet al., “A survey on deep learning techniques for image and video semantic seg- mentation,”Applied Soft Computing, vol. 70, pp. 41–65, 2018

    2018

  20. [20]

    Fully convolutional networks for semantic segmentation,

    J. Long, E. Shelhamer, and T. Darrell, “Fully convolutional networks for semantic segmentation,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015

    2015

  21. [21]

    Pyramid scene parsing network,

    H. Zhao, J. Shi, X. Qi, X. Wang, and J. Jia, “Pyramid scene parsing network,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017

    2017

  22. [22]

    SegNet: A deep convolutional encoder-decoder architecture for image seg- mentation,

    V. Badrinarayanan, A. Kendall, and R. Cipolla, “SegNet: A deep convolutional encoder-decoder architecture for image seg- mentation,”IEEE Transactions on Pattern Analysis and Ma- chine Intelligence, vol. 39, no. 12, pp. 2481–2495, 2017

    2017

  23. [23]

    Rethink- ing atrous convolution for semantic image segmentation,

    L.-C.Chen, G.Papandreou,F. Schroff, and H.Adam, “Rethink- ing atrous convolution for semantic image segmentation,”arXiv preprint arXiv:1706.05587, 2017

    Pith/arXiv arXiv 2017

  24. [24]

    U-Net:Convolutional networks for biomedical image segmentation,

    O.Ronneberger,P.Fischer,andT.Brox,“U-Net:Convolutional networks for biomedical image segmentation,” inMedical Image Computing and Computer-Assisted Intervention (MICCAI), 2015

    2015

  25. [25]

    MultiResUNet: Rethinking the u-net architecture for multimodal biomedical image segmenta- tion,

    N. Ibtehaz and M. S. Rahman, “MultiResUNet: Rethinking the u-net architecture for multimodal biomedical image segmenta- tion,”Neural Networks, vol. 121, pp. 74–87, 2020

    2020

  26. [26]

    UNet++: A nested U-Net architecture for medical image segmentation,

    Z. Zhou, M. M. Rahman Siddiquee, N. Tajbakhsh, and J. Liang, “UNet++: A nested U-Net architecture for medical image segmentation,” inDeep Learning in Medical Image Analysis (DLMIA), 2018

    2018

  27. [27]

    Attention U- Net: Learning where to look for the pancreas,

    O. Oktay, J. Schlemper, L. L. Folgocet al., “Attention U- Net: Learning where to look for the pancreas,”arXiv preprint arXiv:1804.03999, 2018

    Pith/arXiv arXiv 2018

  28. [28]

    Deepcrack: A deep hierarchical feature learning architecture for crack segmentation,

    Y. Liu, J. Yao, X. Lu, R. Xie, and L. Li, “Deepcrack: A deep hierarchical feature learning architecture for crack segmentation,”Neurocomputing, vol. 338, pp. 139–153, 2019. [Online]. Available: https://doi.org/10.1016/j.neucom.2019.01. 036

    work page doi:10.1016/j.neucom.2019.01 2019

  29. [29]

    Crackformer network for pavement crack segmentation,

    H. Liu, J. Yang, X. Miao, C. Mertz, and H. Kong, “Crackformer network for pavement crack segmentation,”IEEE Transactions onIntelligentTransportationSystems,2023.[Online].Available: https://doi.org/10.1109/TITS.2023.3266776

    arXiv 2023

  30. [30]

    Masked-attention mask transformer for universal image segmentation,

    B. Cheng, I. Misra, A. G. Schwing, A. Kirillov, and R. Girdhar, “Masked-attention mask transformer for universal image segmentation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022. [Online]. Available: https://arxiv.org/abs/2112.01527

    Pith/arXiv arXiv 2022

  31. [31]

    Fine-tuning vision foundation model for crack segmentation in civil infrastructures,

    K. Ge, C. Wang, Y. Guo, Y. Tang, Z. Hu, and H. Chen, “Fine-tuning vision foundation model for crack segmentation in civil infrastructures,”arXiv preprint arXiv:2312.04233, 2023. [Online]. Available: https://arxiv.org/abs/2312.04233

    Pith/arXiv arXiv 2023

  32. [32]

    Application of segment anything model for civil infrastructure defect assessment,

    M. Ahmadi, A. G. Lonbar, H. K. Naeini, A. T. Beris, M. Nouri, A. S. Javidi, and A. Sharifi, “Application of segment anything model for civil infrastructure defect assessment,” arXiv preprint arXiv:2304.12600, 2023. [Online]. Available: https://arxiv.org/abs/2304.12600

    Pith/arXiv arXiv 2023

  33. [33]

    Crackseg9k: A collection and benchmark for crack segmentation datasets and frameworks,

    S. Kulkarni, S. Singh, D. Balakrishnan, S. Sharma, S. Devunuri, and S. C. R. Korlapati, “Crackseg9k: A collection and benchmark for crack segmentation datasets and frameworks,” in European Conference on Computer Vision (ECCV) Workshops, ser. Lecture Notes in Computer Science, vol. 13807, 2022. [Online]. Available: https://arxiv.org/abs/2208.13054

    Pith/arXiv arXiv 2022

  34. [34]

    Omnicrack30k: A benchmark for crack segmentation and the reasonable effectiveness of transfer learning,

    C. Benz and V. Rodehorst, “Omnicrack30k: A benchmark for crack segmentation and the reasonable effectiveness of transfer learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops,

  35. [35]

    Few-shot semantic segmentation: a review of methodologies, benchmarks, and open challenges,

    N. Catalano and M. Matteucci, “Few-shot semantic segmentation: a review of methodologies, benchmarks, and open challenges,”arXiv preprint arXiv:2304.05832, 2023. [Online]. Available: https://arxiv.org/abs/2304.05832

    Pith/arXiv arXiv 2023

  36. [36]

    Learning with limited annotations: A survey on deep semi-supervised learning for medical image segmentation,

    R. Jiao, Y. Zhang, L. Ding, B. Xue, J. Zhang, R. Cai, and C. Jin, “Learning with limited annotations: A survey on deep semi-supervised learning for medical image segmentation,” Computers in Biology and Medicine, 2023. [Online]. Available: https://arxiv.org/abs/2207.14191

    Pith/arXiv arXiv 2023

  37. [37]

    Datasetgan: Efficient labeled data factory with minimal human effort,

    Y. Zhang, H. Ling, J. Gao, K. Yin, J.-F. Lafleche, A. Barriuso, A. Torralba, and S. Fidler, “Datasetgan: Efficient labeled data factory with minimal human effort,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021. [Online]. Available: https: //arxiv.org/abs/2104.06490

    Pith/arXiv arXiv 2021

  38. [38]

    Establishment and evaluationofconditionalGAN-basedimagedatasetforsemantic segmentation of structural cracks,

    T. Jin, X. W. Ye, and Z. X. Li, “Establishment and evaluationofconditionalGAN-basedimagedatasetforsemantic segmentation of structural cracks,”Engineering Structures, vol. 285, p. 116058, 2023. [Online]. Available: https://doi.org/10. 1016/j.engstruct.2023.116058

    arXiv 2023

  39. [39]

    Diffumask: Synthesizing images with pixel-level annotations for semantic segmentation using diffusion models,

    W.Wu,Y.Zhao,M.Z.Shou,H.Zhou,andC.Shen,“Diffumask: Synthesizing images with pixel-level annotations for semantic segmentation using diffusion models,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023. [Online]. Available: https://arxiv.org/abs/2303. 11681

    2023

  40. [41]

    Adding conditional control to text-to-image diffusion models,

    L. Zhang, A. Rao, and M. Agrawala, “Adding conditional control to text-to-image diffusion models,” inProceedings of 24 the IEEE/CVF International Conference on Computer Vision (ICCV), 2023

    2023

  41. [42]

    Dreambooth: Fine tuning text-to-image diffu- sion models for subject-driven generation,

    N. Ruiz, Y. Li, V. Jampani, Y. Pritch, M. Rubinstein, and K. Aberman, “Dreambooth: Fine tuning text-to-image diffu- sion models for subject-driven generation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023

    2023

  42. [43]

    LoRA: Low-rank adaptation of large language models,

    E. J. Hu, Y. Shen, P. Wallis, Z. Allen-Zhu, Y. Li, S. Wang, L. Wang, and W. Chen, “LoRA: Low-rank adaptation of large language models,” inInternational Conference on Learning Representations (ICLR), 2022

    2022

  43. [44]

    Latent diffusion models to enhance the performance of visual defect segmentation networks in steel surface inspection,

    J. Leiñena, F. A. Saiz, and I. Barandiaran, “Latent diffusion models to enhance the performance of visual defect segmentation networks in steel surface inspection,” Sensors, vol. 24, no. 18, p. 6016, 2024. [Online]. Available: https://doi.org/10.3390/s24186016

    doi:10.3390/s24186016 2024

  44. [45]

    Synthetic data from diffusion models improves ImageNet clas- sification,

    S. Azizi, S. Kornblith, C. Saharia, M. Norouzi, and D. J. Fleet, “Synthetic data from diffusion models improves ImageNet clas- sification,”Transactions on Machine Learning Research, 2023

    2023

  45. [46]

    Effective data augmentation with diffusion models,

    B. Trabucco, K. Doherty, M. Gurinas, and R. Salakhutdinov, “Effective data augmentation with diffusion models,” in International Conference on Learning Representations (ICLR),

  46. [47]

    Simple copy-paste is a strong data augmentation method for instance segmentation,

    G. Ghiasi, Y. Cui, A. Srinivas, R. Qian, T.-Y. Lin, E. D. Cubuk, Q. V. Le, and B. Zoph, “Simple copy-paste is a strong data augmentation method for instance segmentation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021. [Online]. Available: https://arxiv.org/abs/2012.07177

    Pith/arXiv arXiv 2021

  47. [48]

    Cutmix: Regularization strategy to train strong classifiers with localizable features,

    S. Yun, D. Han, S. J. Oh, S. Chun, J. Choe, and Y. J. Yoo, “Cutmix: Regularization strategy to train strong classifiers with localizable features,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019. [Online]. Available: https://arxiv.org/abs/1905.04899

    Pith/arXiv arXiv 2019

  48. [49]

    Available: https://arxiv.org/abs/2302.07944

    [Online]. Available: https://arxiv.org/abs/2302.07944

    Pith/arXiv arXiv

  49. [50]

    Leakage and the reproducibility crisis in machine-learning-based science,

    S. Kapoor and A. Narayanan, “Leakage and the reproducibility crisis in machine-learning-based science,”Patterns, vol. 4, no. 9, p. 100804, 2023. [Online]. Available: https://doi.org/10.1016/j. patter.2023.100804

    arXiv 2023

  50. [51]

    Don’t push the button! exploring data leakage risks in machine learning and transfer learning,

    A. Apicella, F. Isgrò, and R. Prevete, “Don’t push the button! exploring data leakage risks in machine learning and transfer learning,”Artificial Intelligence Review, vol. 58, 2025. [Online]. Available: https://doi.org/10.1007/s10462-025-11326-3

    doi:10.1007/s10462-025-11326-3 2025

  51. [52]

    Syntheticallyenhanced: unveiling synthetic data’s potential in medical imaging research,

    B.Khosravi,F.Li,T.Dapamedeetal.,“Syntheticallyenhanced: unveiling synthetic data’s potential in medical imaging research,”eBioMedicine, vol. 104, p. 105174, 2024. [Online]. Available: https://www.thelancet.com/journals/ebiom/article/ PIIS2352-3964(24)00209-3/fulltext

    2024

  52. [53]

    Improvement of concrete crack segmentation performance using stacking ensemble learning,

    T. Lee, J.-H. Kim, S.-J. Lee, S.-K. Ryu, and B.-C. Joo, “Improvement of concrete crack segmentation performance using stacking ensemble learning,”Applied Sciences, vol. 13, no. 4, p. 2367, 2023. [Online]. Available: https://doi.org/10. 3390/app13042367

    2023

  53. [54]

    Simple and scalable predictive uncertainty estimation using deep ensembles,

    B. Lakshminarayanan, A. Pritzel, and C. Blundell, “Simple and scalable predictive uncertainty estimation using deep ensembles,” inAdvances in Neural Information Processing Systems (NeurIPS), 2017. [Online]. Available: https://arxiv. org/abs/1612.01474

    Pith/arXiv arXiv 2017

  54. [55]

    A performance improvement strategy for concrete damage detection using stacking ensemble learning of multiple semantic segmentation networks,

    S. Li and X. Zhao, “A performance improvement strategy for concrete damage detection using stacking ensemble learning of multiple semantic segmentation networks,”Sensors, vol. 22, no. 9, p. 3341, 2022. [Online]. Available: https://doi.org/10. 3390/s22093341

    2022

  55. [56]

    On calibrating semantic segmentation models: Analyses and an algorithm,

    D. Wang, B. Gong, and L. Wang, “On calibrating semantic segmentation models: Analyses and an algorithm,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023. [Online]. Available: https://arxiv.org/abs/2212.12053

    Pith/arXiv arXiv 2023

  56. [57]

    Training region- based object detectors with online hard example mining,

    A. Shrivastava, A. Gupta, and R. Girshick, “Training region- based object detectors with online hard example mining,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016

    2016

  57. [58]

    On calibra- tion of modern neural networks,

    C. Guo, G. Pleiss, Y. Sun, and K. Q. Weinberger, “On calibra- tion of modern neural networks,” inInternational Conference on Machine Learning (ICML), 2017

    2017

  58. [59]

    Segment anything,

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafsonet al., “Segment anything,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023

    2023

  59. [60]

    Synthetic sand boil dataset for levee monitoring,

    P. J. Thapa and M. T. Hoque, “Synthetic sand boil dataset for levee monitoring,” IEEE DataPort, Sep. 2024. [Online]. Available: https://ieee-dataport.org/ documents/synthetic-sand-boil-dataset-levee-monitoring

    2024

  60. [61]

    Levee owner’s manual for non- federal flood control works,

    U. A. C. of Engineers, “Levee owner’s manual for non- federal flood control works,” https://www.mvr.usace.army.mil/ Portals/48/docs/EC/LSP/LeveeOwnersManual.pdf, accessed 2025-02-17

    2025

  61. [62]

    Convex_Hull_Annotator: Automated convex- hull mask annotation from a SandBoilNet probability map,

    P. J. Thapa, “Convex_Hull_Annotator: Automated convex- hull mask annotation from a SandBoilNet probability map,” https://github.com/padam56/Convex_Hull_Annotator, 2024, companion tool to [3]; applies thresholding, connected- component analysis, and per-component convex hulls to a CNN- predicted sandboil probability map

    2024

  62. [63]

    DPM- Solver: A fast ODE solver for diffusion probabilistic model sampling in around 10 steps,

    C. Lu, Y. Zhou, F. Bao, J. Chen, C. Li, and J. Zhu, “DPM- Solver: A fast ODE solver for diffusion probabilistic model sampling in around 10 steps,”Advances in Neural Information Processing Systems (NeurIPS), 2022

    2022

  63. [64]

    Synthetic animal burrow dataset for levee monitoring,

    P. J. Thapa and M. T. Hoque, “Synthetic animal burrow dataset for levee monitoring,” IEEE DataPort, Jun. 2025. [Online]. Available: https://ieee-dataport.org/ documents/synthetic-animal-burrow-dataset-levee-monitoring

    2025

  64. [65]

    A ConvNet for the 2020s,

    Z. Liu, H. Mao, C.-Y. Wu, C. Feichtenhofer, T. Darrell, and S. Xie, “A ConvNet for the 2020s,” inProceedings of the IEEE/CVFConferenceonComputerVisionandPatternRecog- nition (CVPR), 2022

    2022

  65. [66]

    SegFormer: Simple and efficient design for semantic segmentation with transformers,

    E. Xie, W. Wang, Z. Yu, A. Anandkumar, J. M. Alvarez, and P. Luo, “SegFormer: Simple and efficient design for semantic segmentation with transformers,” inAdvances in Neural Infor- mation Processing Systems (NeurIPS), 2021

    2021

  66. [67]

    Concurrent spatial and channel squeeze & excitation in fully convolutional net- works,

    A. G. Roy, N. Navab, and C. Wachinger, “Concurrent spatial and channel squeeze & excitation in fully convolutional net- works,” inMedical Image Computing and Computer Assisted Intervention (MICCAI), 2018

    2018

  67. [68]

    Feature pyramid networks for object detection,

    T.-Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature pyramid networks for object detection,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017

    2017

  68. [69]

    MA-Net: A multi-scale attention network for liver and tumor segmentation,

    T. Fan, G. Wang, Y. Li, and H. Wang, “MA-Net: A multi-scale attention network for liver and tumor segmentation,”IEEE Access, vol. 8, pp. 179656–179665, 2020

    2020

  69. [70]

    EfficientNet: Rethinking model scaling for convolutional neural networks,

    M. Tan and Q. V. Le, “EfficientNet: Rethinking model scaling for convolutional neural networks,” inProceedings of the 36th International Conference on Machine Learning (ICML), 2019

    2019

  70. [71]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018. [Online]. Available: https://arxiv.org/abs/1709.01507

    Pith/arXiv arXiv 2018

  71. [72]

    XGBoost: A scalable tree boosting system,

    T. Chen and C. Guestrin, “XGBoost: A scalable tree boosting system,” inProceedings of the 22nd ACM SIGKDD Interna- tional Conference on Knowledge Discovery and Data Mining (KDD), 2016

    2016

  72. [73]

    Encoder-decoder with atrous separable convolution for seman- tic image segmentation,

    L.-C. Chen, Y. Zhu, G. Papandreou, F. Schroff, and H. Adam, “Encoder-decoder with atrous separable convolution for seman- tic image segmentation,” inEuropean Conference on Computer Vision (ECCV), 2018

    2018

  73. [74]

    Segmentation models PyTorch,

    P. Iakubovskii, “Segmentation models PyTorch,” https:// github.com/qubvel/segmentation_models.pytorch, 2019

    2019

  74. [75]

    Albumentations: Fast and flexible image augmentations,

    A. Buslaev, V. I. Iglovikov, E. Khvedchenya, A. Parinov, M. Druzhinin, and A. A. Kalinin, “Albumentations: Fast and flexible image augmentations,”Information, vol. 11, no. 2, p. 125, 2020

    2020

  75. [76]

    PyTorch: An imperative style, high-performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G.Chanan,T.Killeen,Z.Lin,N.Gimelshein,L.Antiga,A.Des- maison, A. Köpf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala, “PyTorch: An imperative style, high-performance deep learning library,” inAdvances in Neural Information Processing Systems (Ne...

    2019

  76. [77]

    Sam 2: Segment anything in images and videos,

    N. Ravi, V. Gabeur, Y.-T. Huet al., “Sam 2: Segment anything in images and videos,”arXiv preprint arXiv:2408.00714, 2024. [Online]. Available: https://arxiv.org/abs/2408.00714

    Pith/arXiv arXiv 2024

  77. [79]

    Scikit-learn: Ma- chine learning in Python,

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and É. Duchesnay, “Scikit-learn: Ma- chine learning in Python,”Journal of Machine Learning Re- search, vol. 12, pp. 2825–2830, 2011

    2011

  78. [2023]

    Available: https://arxiv.org/abs/2311.17121

    [Online]. Available: https://arxiv.org/abs/2311.17121

    Pith/arXiv arXiv

  79. [2024]

    Available: https://openaccess.thecvf.com/ content/CVPR2024W/VAND/papers/Benz_OmniCrack30k_ A_Benchmark_for_Crack_Segmentation_and_the_ Reasonable_Effectiveness_CVPRW_2024_paper.pdf

    [Online]. Available: https://openaccess.thecvf.com/ content/CVPR2024W/VAND/papers/Benz_OmniCrack30k_ A_Benchmark_for_Crack_Segmentation_and_the_ Reasonable_Effectiveness_CVPRW_2024_paper.pdf