Pith. sign in

REVIEW 3 major objections 5 minor 64 references

InfraDiffusion: zero-shot depth map restoration with diffusion models and prompted segmentation from sparse infrastructure point clouds

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read InfraDiffusion claims that sparse, noisy depth maps of masonry infrastructure can be restored zero-shot by a natural-image diffusion model, with boundary masks preventing hallucinated bricks, and that the restored maps turn a failing brick-

desk verdict The restoration method is plausible, but the headline mIoU gains are circular because the brick masks are drawn on the restored images themselves. read the letter →

arxiv 2509.03324 v1 pith:H7LVU2SS submitted 2025-09-03 cs.CV

Yixiong Jing , Cheng Zhang , Haibing Wu , Guangming Wang , Olaf Wysocki , Brian Sheil This is my paper
classification cs.CV
keywords zero-shotimagerestorationdiffusionmodelspointclouddepthmapsbrick-levelsegmentationmasonryinfrastructureboundarymaskingDDNMvirtualcameraprojection
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

InfraDiffusion claims that sparse, noisy depth maps projected from masonry point clouds can be cleaned up by a pre-trained diffusion model working in a zero-shot mode, with no task-specific training and no clean ground-truth depth maps. The pipeline crops point-cloud patches, projects them onto a virtual pinhole camera, and adapts the Denoising Diffusion Null-space Model (DDNM) to inpaint and denoise the depth image, using a boundary mask to stop the model from inventing bricks outside the scanned surface. The paper argues that the restored depth maps make brick-level segmentation work: with the Segment Anything Model (SAM) and a single positive prompt, mean IoU on one tunnel section rises from 0.064 to 0.708, with similar jumps on two masonry bridges. If correct, inspectors could find brick- and mortar-scale defects directly from LiDAR scans in low-light environments where RGB imaging fails.

What carries the argument

The load-bearing mechanism is the boundary-constrained range–null space update, written as y_hat = M ⊙ (y_0|t − Σ_t A†(A y_0|t − y_tilde)). Here A is the binary mask of valid point projections, A† is its Moore–Penrose pseudoinverse, y_0|t is the clean image predicted by the diffusion model, and M is the bounding-box mask of projected pixels. The range-space correction keeps the restored image consistent with observed points; the null-space term, filled by the pre-trained diffusion model, synthesizes plausible brick texture; and M confines that synthesis to physically valid regions, preventing spurious bricks at structural edges.

What would settle it

Inspect the annotation protocol for the brick masks. If they were created on the restored depth maps, annotate the same patches from the raw point cloud or from independent photographs and rerun SAM on both raw and restored depth maps. If the mIoU advantage of restored inputs disappears under independently sourced masks, the restoration did not in fact enable segmentation.

Watch

Extended reading notes

Core claim

The central claim is that a diffusion prior trained on ordinary images transfers to depth maps of masonry structures well enough to restore them for fine-grained segmentation, as long as restoration is confined to valid projection regions. The paper adapts the zero-shot DDNM by multiplying its range–null space correction by a boundary mask built from the projected point cloud; vanilla DDNM already keeps restored pixels consistent with observed depth values, but without the mask it invents brick-like texture beyond structural edges. InfraDiffusion restrains that generation to the mask. Across three tunnel chunks and two masonry bridges, the restored depth maps show clean, continuous brick bou

Load-bearing premise

The reported segmentation gains assume the brick ground-truth masks were annotated independently of the restored depth maps; if the masks were drawn on the restored images, the main comparison becomes circular.

Editorial extensions

If this is right

  • Brick-level segmentation from point clouds becomes feasible without collecting large annotated 3D datasets or retraining segmentation networks.
  • The same zero-shot restoration could apply to other sparse geometric modalities generated by virtual cameras, such as depth maps of rock faces, tunnels, pipes, or heritage structures.
  • Inspection workflows could reuse a single pre-trained diffusion prior across different infrastructure types, changing only camera parameters and mask geometry.
  • Because prompt-based SAM replaces trained segmentation models, annotation effort shifts from labelling thousands of bricks to placing a few point prompts per brick.

Reading between the lines

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

  • The paper leaves the ground-truth annotation provenance unspecified; if the brick masks were traced on restored depth maps rather than on the original data, the mIoU comparison would partly measure self-agreement rather than restoration quality. An independent annotation source would settle this.
  • The noise level σ_y = 0.16 is chosen visually on these datasets, so the zero-shot claim may not transfer to noisier or denser scans without re-tuning that single parameter.
  • Because the diffusion prior was trained on natural RGB images, the restored depth map is a stylised geometric interpretation, not a measured surface; downstream quantitative measurements such as crack widths or mortar-loss volumes would need validation against direct measurements.
  • A natural next test is to feed restored depth maps into a metric that uses absolute depth values rather than visual segmentation, since the current evaluation shows only that brick boundaries become segmentable.
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 / 5 minor

Summary. The paper proposes InfraDiffusion, a zero-shot framework that projects sparse and noisy masonry infrastructure point clouds into depth maps via virtual cameras and restores them by adapting the Denoising Diffusion Null-space Model (DDNM) with a novel boundary mask. The restored depth maps are then fed into SAM with point prompts for brick-level segmentation. The authors claim significant mIoU improvements after restoration, e.g., from 0.064 to 0.708 on Wheatley Tunnel TC15, and present qualitative results on one tunnel and two bridges, plus an ablation showing that the boundary mask suppresses spurious structures outside valid projections.

Significance. If the quantitative claims were reliable, the framework would be a useful contribution: it leverages pre-trained diffusion priors without task-specific training, releases code and data, and addresses a real need in masonry infrastructure inspection where RGB imagery is often unavailable. The DDNM adaptation is clearly described, and the boundary-mask ablation demonstrates a plausible improvement over vanilla DDNM. However, the central quantitative validation is undermined by the GT annotation protocol, which appears to derive masks from the restored images themselves, and by the empirically tuned noise parameter sigma_y selected on test patches. These issues make the reported mIoU gains not trustworthy, so the paper's stated significance is not currently established.

major comments (3)
  1. [§4.4, Figure 11/12] The manuscript never states on which image the brick GT masks were annotated. Section 4.4 says only that labels were created on the sampled patches for bricks with 'clear and well-defined boundaries,' while Figure 11 and 12 label the GT panels as 'GT (y0)'. If the masks were drawn on the restored output y0, the main mIoU comparison is circular: positive prompts are derived from GT polygons on y0, SAM is evaluated on y0 against masks derived from y0, and the noisy input y_j is evaluated against masks that originate from a different, cleaner image. The reported 0.064→0.708 improvement would then largely reflect annotation-source mismatch rather than restoration quality. The authors must specify the annotation image exactly; if it is y0, the evaluation must be redone with masks annotated independently (e.g., on the original point cloud or the degraded input).
  2. [§4.2, Figure 9] The measurement noise standard deviation sigma_y = 0.16 is selected by visual inspection on test patches (Figure 9) and then applied to all subsequent experiments. The same patches are used in the quantitative mIoU evaluation of Table 5. This constitutes selection on the test set and can inflate the reported gains. The authors should provide a principled selection procedure, report results across the full range of sigma_y values, or use a separate validation set for any tuning, so that the reported numbers reflect out-of-sample behavior.
  3. [§4.4, Table 4 and Table 5] The text states that 'The experiments use the same patch indices listed in Table 4' for the SAM evaluation. Table 4 lists only one arch patch and one wall/pier patch per infrastructure (e.g., TC15: patches 1828 and 969). If the mIoU values in Table 5 are computed from only two patches per dataset, they are not representative of the 100/60/140 patches described in Section 3, and the conclusions are statistically fragile. The authors should clarify how many patches were used and report mIoU over the full patch set with appropriate variance.
minor comments (5)
  1. [§2.2.3] Typo: 'The coral idea' should be 'The core idea'.
  2. [§1] The paragraph 'Unlike 2D vision...' is duplicated verbatim.
  3. [§4.2] The sentence 'The different orientations of ~y_j between arch and wall/pier arise...' is duplicated.
  4. [Table 5] The column header 'Wheatley Tunnel (TC20)' appears twice; the third column should presumably be 'Wheatley Tunnel (TC25)'.
  5. [§2.2.2/§2.2.4] References to 'DDPM (Wang et al., 2022)' should cite DDNM (Wang et al., 2022); DDPM refers to Ho et al. (2020).

Circularity Check

2 steps flagged · score 6.0 of 10

Main mIoU claim is partly circular: GT masks are attached to the restored image y0 and σy is tuned on the same test patches, while the restoration itself is a self-contained zero-shot derivation.

  1. self definitional [Section 4.4, Figure 11 caption, Table 5]
    "GT annotations are created only for bricks with clear and well-defined boundaries... The labelling is performed on the sampled patches introduced in the Dataset section. ... (a)–(d) show GT ( y0), SAM predictions on y0, ˜yj, and SAM predictions on ˜yj for test chunk 15."

    The ground-truth masks used to compute mIoU in Table 5 are displayed as 'GT (y0)' in Figure 11, and the paper never states that the masks were annotated on the noisy input ˜yj. If the masks were drawn on the restored image y0, then SAM on y0 is measured against masks derived from the same image it is segmenting, while SAM on ˜yj is measured against masks from a different, cleaner image. The mIoU improvement (0.064 → 0.708) therefore partly reflects annotation-source alignment rather than restoration fidelity. Since Section 4.2 states that quantitative depth GT is unavailable, there is no independent check that y0 is geometrically consistent with the original point cloud.

  2. fitted input called prediction [Section 4.2, Figure 9, Tables 4–5]
    "Based on this observation, we select σy = 0.16 for all subsequent experiments. This choice represents a practical balance between noise suppression and boundary preservation. We note, however, that the optimal value is empirically and visually determined and may vary across datasets due to differences in point density, registration errors, and sensor noise."

    The noise-level parameter σy was selected by visual inspection on the exact test patches later used for quantitative evaluation (e.g., Wheatley Tunnel TC15 arch patch 1828 and wall patch 969; anonymised UK bridge arch patch 703 and pier patch 145, listed in Table 4). The same patches then contribute to the mIoU numbers in Table 5. Thus the headline 'zero-shot' improvements are not out-of-sample: the restoration strength was tuned to make those specific test images look best before measuring segmentation on those same images. This is a test-set selection bias that inflates the reported mIoU gains.

full rationale

The restoration pipeline itself—virtual camera projection, range-null-space DDNM with boundary masking—is a self-contained adaptation of published zero-shot methods and does not reduce by construction to its inputs. The depth-restoration derivation is not circular. However, the paper's central quantitative claim, the mIoU jump from 0.064 to 0.708 on Wheatley Tunnel TC15, is not an independent benchmark. Section 4.2 says quantitative depth GT is unavailable, so the only quantitative validation is SAM segmentation evaluated against manual brick masks. Figure 11 labels the GT panel as 'GT (y0)', and Section 4.4 says labeling was performed on the sampled patches without ever stating that masks were drawn on the noisy input ˜yj; the 'clear and well-defined boundaries' criterion is far more natural on the restored image than on the noisy one. If the masks were annotated on y0, then SAM-on-y0 is measured against masks derived from the very image it segments, while SAM-on-˜yj is held to masks from a different, cleaner image; the reported improvement then partly reflects annotation-source alignment rather than restoration quality. Additionally, σy=0.16 was selected by visual inspection on the exact test patches (Figure 9, Table 4) before reporting mIoU on those same patches, so the 'zero-shot' claim is qualified by test-set tuning. No load-bearing self-citation chain was found: the authors' prior works are cited for component-level segmentation background, not for the central restoration result. The circularity is confined to the evaluation protocol rather than the method derivation, hence a score of 6 rather than 8 or 10.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim rests on standard IR assumptions (linear degradation with known mask and Gaussian noise), on the cross-domain transfer of a natural-image diffusion prior to masonry depth maps, and on SAM's ability to segment depth images. Several geometric and noise parameters are fitted to the test data, and the evaluation benchmark itself is not clearly independent of the restored outputs.

free parameters (4)
  • sigma_y (measurement noise std) = 0.16
    Selected by visual inspection on the same test patches (Fig. 9); the paper states the optimal value is empirically and visually determined.
  • patch side length s = 0.8 m
    Chosen by trial-and-error on the test infrastructure point clouds to balance local completeness and boundary clarity.
  • patch thickness t = 0.25 m
    Chosen by trial-and-error to capture geometric variation near corners and non-shell-like regions.
  • camera distance d = 0.8 m
    Selected to maximize coverage of the patch while preserving necessary detail of individual bricks.
assumptions (3)
  • domain assumption The degradation model for projected depth maps is linear: y = A y + n with known binary mask A and i.i.d. Gaussian noise n of known variance.
    Invoked in Section 2.2.1 to cast restoration as inpainting plus denoising; real TLS noise and occlusion are not necessarily Gaussian.
  • domain assumption A diffusion model trained on natural RGB images provides a strong and valid prior for masonry depth maps.
    The method relies on OpenAI's ImageNet-trained unconditional diffusion model; the paper asserts transfer without quantitative depth validation.
  • domain assumption SAM can segment bricks from depth maps when prompted with centroids.
    SAM was trained on RGB images; its use on restored depth maps is assumed to produce meaningful masks that correspond to real brick geometry.

how reviews work

0 comments
Cite this review

Pith. "Pith review of InfraDiffusion: zero-shot depth map restoration with diffusion models and prompted segmentation from sparse infrastructure point clouds." pith.science (2026). https://pith.science/paper/H7LVU2SS

@misc{pith2026250903324,
  author       = {Pith},
  title        = {Pith review of: InfraDiffusion: zero-shot depth map restoration with diffusion models and prompted segmentation from sparse infrastructure point clouds},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/H7LVU2SS}},
  note         = {Machine review of arXiv:2509.03324}
}
read the original abstract

Point clouds are widely used for infrastructure monitoring by providing geometric information, where segmentation is required for downstream tasks such as defect detection. Existing research has automated semantic segmentation of structural components, while brick-level segmentation (identifying defects such as spalling and mortar loss) has been primarily conducted from RGB images. However, acquiring high-resolution images is impractical in low-light environments like masonry tunnels. Point clouds, though robust to dim lighting, are typically unstructured, sparse, and noisy, limiting fine-grained segmentation. We present InfraDiffusion, a zero-shot framework that projects masonry point clouds into depth maps using virtual cameras and restores them by adapting the Denoising Diffusion Null-space Model (DDNM). Without task-specific training, InfraDiffusion enhances visual clarity and geometric consistency of depth maps. Experiments on masonry bridge and tunnel point cloud datasets show significant improvements in brick-level segmentation using the Segment Anything Model (SAM), underscoring its potential for automated inspection of masonry assets. Our code and data is available at https://github.com/Jingyixiong/InfraDiffusion-official-implement.

Figures

Figures reproduced from arXiv: 2509.03324 by the authors.

Figure 1
Figure 1. Overview of the InfraDiffusion pipeline. [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Overview of the patch extraction and projection using the pinhole camera model. [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Boundary mask Mj extraction. (a) represents a virtual camera projection near the infrastructure boundary; (b) shows depth mask ˜yj with missing regions shown by the solid red box; (c) visualises the boundary mask Mj where the foreground and background are shown in green and red respectively. Due to the fixed field of view of the virtual camera, the resulting image plane may include regions not covered by any 3D poin… view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Wheatley Tunnel overview and selected test chunks. (a) Schematic plan view [PITH_FULL_IMAGE:figures/full_fig_p017_4.png]
Figure 5
Figure 5. Figure 5: Semantic segmentation of the Wheatley Tunnel (TC15) point cloud. The arch [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]
Figure 6
Figure 6. Figure 6: Point clouds and semantic segmentation of anonymised UK bridge and Hertford [PITH_FULL_IMAGE:figures/full_fig_p019_6.png]
Figure 7
Figure 7. Figure 7: IR results for arch components across five infrastructures with [PITH_FULL_IMAGE:figures/full_fig_p021_7.png]
Figure 8
Figure 8. Figure 8: IR results for wall/pier components across five infrastructures with [PITH_FULL_IMAGE:figures/full_fig_p021_8.png]
Figure 9
Figure 9. Figure 9: Effect of varying σy ∈ {0, 0.16, 0.32, 0.64} on IR quality. (a)–(e) and (f)–(j) show y˜j and y0 for the arch and walls of Wheatley Tunnel (TC15) (patch 1828 and patch 969, respectively). (k)–(o) and (p)–(t) show corresponding results for the arch and pier of the anonym…
Figure 10
Figure 10. Figure 10: Comparison between DDNM and InfraDiffusion on wall/pier patches. (a)–(d) [PITH_FULL_IMAGE:figures/full_fig_p024_10.png]
Figure 11
Figure 11. Figure 11: Zero-shot segmentation of arch patches from Wheatley Tunnel using SAM with [PITH_FULL_IMAGE:figures/full_fig_p025_11.png]
Figure 12
Figure 12. Figure 12: Zero-shot segmentation of wall patches from Wheatley Tunnel using SAM with [PITH_FULL_IMAGE:figures/full_fig_p026_12.png]
Figure 13
Figure 13. Figure 13: Zero-shot segmentation of masonry bridge patches using SAM with ViT-H by [PITH_FULL_IMAGE:figures/full_fig_p028_13.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

64 extracted references · 44 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  3. [3]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  4. [4]

    , author DeJong, M.J

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

    work page doi:10.1002/stc.2187 2018

  5. [5]

    , author Soga, K

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

    work page doi:10.1016/j.conbuildmat.2017.05.075 2017

  6. [6]

    , year 1974

    author Andrews, H.C. , year 1974 . title Digital image restoration: A survey . journal Computer volume 7 , pages 36--45 . 10.1109/MC.1974.6323527

    work page arXiv 1974

  7. [7]

    , year 2022

    author Brackenbury, D. , year 2022 . title Automated Image-Based Inspection of Masonry Arch Bridges . Ph.D. thesis. Apollo - University of Cambridge Repository. https://www.repository.cam.ac.uk/handle/1810/338321, :10.17863/CAM.85731

    work page doi:10.17863/cam.85731 2022

  8. [8]

    , author Dong, Q

    author Cao, C. , author Dong, Q. , author Fu, Y. , year 2022 . title Learning prior feature and attention enhanced image inpainting , in: booktitle European conference on computer vision , organization Springer . pp. pages 306--322 . 10.1007/978-3-031-19784-0_18

    work page doi:10.1007/978-3-031-19784-0_18 2022

Show all 64 references
  1. [9]

    , author Xiong, Z

    author Chen, C. , author Xiong, Z. , author Tian, X. , author Zha, Z.J. , author Wu, F. , year 2019 . title Real-world image denoising with deep boosting . journal IEEE Transactions on Pattern Analysis and Machine Intelligence volume 42 , pages 3071--3087 . 10.1109/TPAMI.2019.2921548

    2019

  2. [10]

    , author Kang, Z

    author Chen, Q. , author Kang, Z. , author Cao, Z. , author Xie, X. , author Guan, B. , author Pan, Y. , author Chang, J. , year 2024 . title Combining cylindrical voxel and mask r-cnn for automatic detection of water leakages in shield tunnel point clouds . journal Remote Sen...

    2024 doi

  3. [11]

    , author Ma, H

    author Chen, X. , author Ma, H. , author Wan, J. , author Li, B. , author Xia, T. , year 2017 . title Multi-view 3d object detection network for autonomous driving , in: booktitle Proceedings of the IEEE conference on Computer Vision and Pattern Recognition , pp. pages 1907--1...

    2017 doi

  4. [12]

    , author Qiu, W.G

    author Cheng, Y.J. , author Qiu, W.G. , author Duan, D.Y. , year 2019 . title Automatic creation of as-is building information model from single-track railway tunnel point clouds . journal Automation in Construction volume 106 , pages 102911

    2019

  5. [13]

    , author Lu, M

    author Dai, F. , author Lu, M. , year 2013 . title Three-dimensional modeling of site elements by analytically processing image data contained in site photos . journal Journal of construction engineering and management volume 139 , pages 881--894 . https://doi.org/10.1061/(ASC...

    2013

  6. [14]

    , author Bal, I.E

    author Dais, D. , author Bal, I.E. , author Smyrou, E. , author Sarhosis, V. , year 2021 . title Automatic crack classification and segmentation on masonry surfaces using convolutional neural networks and transfer learning . journal Automation in Construction volume 125 , page...

    2021

  7. [15]

    , author Nichol, A

    author Dhariwal, P. , author Nichol, A. , year 2021 . title Diffusion models beat gans on image synthesis . journal Advances in neural information processing systems volume 34 , pages 8780--8794

    2021

  8. [16]

    , author Loy, C.C

    author Dong, C. , author Loy, C.C. , author He, K. , author Tang, X. , year 2014 . title Learning a deep convolutional network for image super-resolution , in: booktitle European conference on computer vision , organization Springer . pp. pages 184--199 . 10.1007/978-3-319-10593-2_13

    2014 doi

  9. [17]

    , author Loy, C.C

    author Dong, C. , author Loy, C.C. , author He, K. , author Tang, X. , year 2015 . title Image super-resolution using deep convolutional networks . journal IEEE transactions on pattern analysis and machine intelligence volume 38 , pages 295--307 . 10.1109/TPAMI.2015.2439281

    2015

  10. [18]

    , author Gu, S

    author Fritsche, M. , author Gu, S. , author Timofte, R. , year 2019 . title Frequency separation for real-world super-resolution , in: booktitle 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW) , organization IEEE . pp. pages 3599--3608 . 10.1109/TCS...

    2019

  11. [19]

    , author Mele, E

    author Giordano, A. , author Mele, E. , author De Luca, A. , year 2002 . title Modelling of historical masonry structures: comparison of different approaches through a case study . journal Engineering Structures volume 24 , pages 1057--1069 . https://doi.org/10.1016/S0141-0296...

    2002 doi

  12. [20]

    , author Reinsch, C

    author Golub, G.H. , author Reinsch, C. , year 1971 . title Singular value decomposition and least squares solutions , in: booktitle Linear algebra . publisher Springer , pp. pages 134--151 . https://link.springer.com/content/pdf/10.1007/978-3-662-39778-7_10.pdf

    1971 doi

  13. [21]

    , author Yang, D

    author Guo, D. , author Yang, D. , author Zhang, H. , author Song, J. , author Zhang, R. , author Xu, R. , author Zhu, Q. , author Ma, S. , author Wang, P. , author Bi, X. , et al., year 2025 . title Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement lea...

    2025 doi arXiv

  14. [22]

    , author Napolitano, R.K

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

    2021 doi

  15. [23]

    , author Jang, M

    author Han, J. , author Jang, M. , author Han, H. , author Shin, D.H. , year 2025 . title Automatic extraction of bridge dimensional information using 3d point cloud data . journal KSCE Journal of Civil Engineering , pages 100312 https://doi.org/10.1016/j.kscej.2025.100312

    2025

  16. [24]

    , author Jain, A

    author Ho, J. , author Jain, A. , author Abbeel, P. , year 2020 . title Denoising diffusion probabilistic models . journal Advances in neural information processing systems volume 33 , pages 6840--6851 . 10.5555/3495724.3496298

    2020

  17. [26]

    , author Sheil, B

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

    2023 doi

  18. [27]

    , author Sheil, B

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

    2024 doi

  19. [28]

    , author Zhong, J.X

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

    2024

  20. [29]

    , author Seyedzadeh, S

    author Katsigiannis, S. , author Seyedzadeh, S. , author Agapiou, A. , author Ramzan, N. , year 2023 . title Deep learning for crack detection on masonry fa c ades using limited data and transfer learning . journal Journal of Building Engineering volume 76 , pages 107105 . htt...

    2023

  21. [30]

    , author Hussain, M

    author Khanam, R. , author Hussain, M. , year 2024 . title Yolov11: An overview of the key architectural enhancements . journal arXiv preprint arXiv:2410.17725 https://arxiv.org/abs/2410.17725

    2024 arXiv

  22. [31]

    , author Mintun, E

    author Kirillov, A. , author Mintun, E. , author Ravi, N. , author Mao, H. , author Rolland, C. , author Gustafson, L. , author Xiao, T. , author Whitehead, S. , author Berg, A.C. , author Lo, W.Y. , et al., year 2023 . title Segment anything , in: booktitle Proceedings of the...

    2023 doi arXiv

  23. [32]

    , author Lilge, S

    author Krawciw, A. , author Lilge, S. , author Barfoot, T.D. , year 2024 . title Lasersam: Zero-shot change detection using visual segmentation of spinning lidar . journal arXiv preprint arXiv:2402.10321 https://doi.org/10.48550/arXiv.2402.10321

    2024 doi arXiv

  24. [33]

    , year 2007

    author Lemos, J.V. , year 2007 . title Discrete element modeling of masonry structures . journal International Journal of Architectural Heritage volume 1 , pages 190--213 . https://doi.org/10.1080/15583050601176868

    2007 doi

  25. [34]

    , author He, J

    author Lin, X. , author He, J. , author Chen, Z. , author Lyu, Z. , author Dai, B. , author Yu, F. , author Qiao, Y. , author Ouyang, W. , author Dong, C. , year 2024 . title Diffbir: Toward blind image restoration with generative diffusion prior , in: booktitle European confe...

    2024 doi

  26. [35]

    , author Peters, J

    author Lo, K.S.H. , author Peters, J. , author Spellman, E. , year 2024 . title Roofdiffusion: Constructing roofs from severely corrupted point data via diffusion , in: booktitle European Conference on Computer Vision , organization Springer . pp. pages 38--57 . https://doi.or...

    2024 doi arXiv

  27. [36]

    , year 2013

    author Lourenço, P.B. , year 2013 . title Computational strategies for masonry structures : multi-scale modeling, dynamics, engineering applications and other challenges . https://api.semanticscholar.org/CorpusID:111285063

    2013

  28. [37]

    , author Sarhosis, V

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

    2022

  29. [38]

    , author Armesto, J

    author Lubowiecka, I. , author Armesto, J. , author Arias, P. , author Lorenzo, H. , year 2009 . title Historic bridge modelling using laser scanning, ground penetrating radar and finite element methods in the context of structural dynamics . journal Engineering Structures vol...

    2009 doi

  30. [39]

    , author Ye, Q

    author Luo, J. , author Ye, Q. , author Zhang, S. , author Yang, Z. , year 2023 . title Indoor mapping using low-cost mls point clouds and architectural skeleton constraints . journal Automation in Construction volume 150 , pages 104837 . https://doi.org/10.1016/j.autcon.2023.104837

    2023

  31. [40]

    , author Huang, X

    author Lyu, Y. , author Huang, X. , author Zhang, Z. , year 2020 . title Learning to segment 3d point clouds in 2d image space , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. pages 12255--12264 . 10.1109/CVPR42600.2020.01227

    2020

  32. [41]

    , author Grobler, H

    author Massa, K.J. , author Grobler, H. , year 2024 . title Adapting projection-based lidar semantic segmentation to natural domains . journal Journal of Visual Communication and Image Representation volume 100 , pages 104111 . https://doi.org/10.1016/j.jvcir.2024.104111

    2024

  33. [42]

    , author Laddha, A

    author Meyer, G.P. , author Laddha, A. , author Kee, E. , author Vallespi-Gonzalez, C. , author Wellington, C.K. , year 2019 . title Lasernet: An efficient probabilistic 3d object detector for autonomous driving , in: booktitle Proceedings of the IEEE/CVF conference on compute...

    2019

  34. [43]

    , author Louren c o, P.B

    author Milani, G. , author Louren c o, P.B. , author Tralli, A. , year 2006 . title Homogenised limit analysis of masonry walls, part i: Failure surfaces . journal Computers & structures volume 84 , pages 166--180 . https://doi.org/10.1016/j.compstruc.2005.09.005

    2006 doi

  35. [44]

    , year 2004

    author Orb \'a n, Z. , year 2004 . title Assessment, reliability and maintenance of masonry arch railway bridges in europe . journal Arch Bridges IV--Advances in Assessment, Structural Design and Construction. Eds: P. Roca and C. Molins, Barcelona volume 2004 , pages 152--161

    2004

  36. [45]

    , author Coco, J

    author Qu, T. , author Coco, J. , author Rönnäng, M. , author Sun, W. , year 2014 . title Challenges and Trends of Implementation of 3D Point Cloud Technologies in Building Information Modeling (BIM): Case Studies . pp. pages 809--816 . https://ascelibrary.org/doi/abs/10.1061/...

    2014 doi

  37. [46]

    , year 1972

    author Richardson, W.H. , year 1972 . title Bayesian-based iterative method of image restoration . journal Journal of the optical society of America volume 62 , pages 55--59 . https://doi.org/10.1364/JOSA.62.000055

    1972 doi

  38. [47]

    , author Blattmann, A

    author Rombach, R. , author Blattmann, A. , author Lorenz, D. , author Esser, P. , author Ommer, B. , year 2022 . title High-resolution image synthesis with latent diffusion models , in: booktitle Proceedings of the IEEE/CVF conference on computer vision and pattern recognitio...

    2022

  39. [48]

    , author Ho, J

    author Saharia, C. , author Ho, J. , author Chan, W. , author Salimans, T. , author Fleet, D.J. , author Norouzi, M. , year 2022 . title Image super-resolution via iterative refinement . journal IEEE transactions on pattern analysis and machine intelligence volume 45 , pages 4...

    2022

  40. [49]

    , author Abed, F.M

    author Shanoer, M.M. , author Abed, F.M. , year 2018 . title Evaluate 3d laser point clouds registration for cultural heritage documentation . journal The Egyptian Journal of Remote Sensing and Space Science volume 21 , pages 295--304 . https://doi.org/10.1016/j.ejrs.2017.11.007

    2018 doi

  41. [50]

    , author Meng, C

    author Song, J. , author Meng, C. , author Ermon, S. , year 2020 . title Denoising diffusion implicit models . journal arXiv preprint arXiv:2010.02502 https://doi.org/10.48550/arXiv.2010.02502

    2020 doi arXiv

  42. [51]

    , author Cho, Y.K

    author Wang, C. , author Cho, Y.K. , year 2015 . title Smart scanning and near real-time 3d surface modeling of dynamic construction equipment from a point cloud . journal Automation in Construction volume 49 , pages 239--249 . https://doi.org/10.1016/j.autcon.2014.06.003

    2015 doi

  43. [52]

    , author Wang, Y

    author Wang, L. , author Wang, Y. , author Dong, X. , author Xu, Q. , author Yang, J. , author An, W. , author Guo, Y. , year 2021 a. title Unsupervised degradation representation learning for blind super-resolution , in: booktitle Proceedings of the IEEE/CVF conference on com...

    2021

  44. [53]

    , year 2023

    author Wang, W. , year 2023 . title Advanced auto labeling solution with added features . https://github.com/CVHub520/X-AnyLabeling

    2023

  45. [54]

    , author Xie, L

    author Wang, X. , author Xie, L. , author Dong, C. , author Shan, Y. , year 2021 b. title Real-esrgan: Training real-world blind super-resolution with pure synthetic data , in: booktitle Proceedings of the IEEE/CVF international conference on computer vision , pp. pages 1905--...

    2021

  46. [55]

    , author Yu, J

    author Wang, Y. , author Yu, J. , author Zhang, J. , year 2022 . title Zero-shot image restoration using denoising diffusion null-space model . journal arXiv preprint arXiv:2212.00490 https://doi.org/10.48550/arXiv.2212.00490

    2022 doi arXiv

  47. [56]

    , author Yu, Y

    author Wang, Y. , author Yu, Y. , author Yang, W. , author Guo, L. , author Chau, L.P. , author Kot, A.C. , author Wen, B. , year 2023 . title Exposurediffusion: Learning to expose for low-light image enhancement , in: booktitle Proceedings of the IEEE/CVF International Confer...

    2023 doi arXiv

  48. [57]

    , author Zhou, X

    author Wu, B. , author Zhou, X. , author Zhao, S. , author Yue, X. , author Keutzer, K. , year 2019 a. title Squeezesegv2: Improved model structure and unsupervised domain adaptation for road-object segmentation from a lidar point cloud , in: booktitle 2019 international confe...

    2019

  49. [58]

    , author Ao, X

    author Wu, H. , author Ao, X. , author Chen, Z. , author Liu, C. , author Xu, Z. , author Yu, P. , year 2019 b. title Concrete spalling detection for metro tunnel from point cloud based on roughness descriptor . journal Journal of Sensors volume 2019 , pages 8574750 . https://...

    2019 doi

  50. [59]

    , author Acikgoz, S

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

    2018 doi

  51. [60]

    , author Lin, W

    author Ye, Z. , author Lin, W. , author Faramarzi, A. , author Xie, X. , author Nini \'c , J. , year 2025 . title Sam4tun: No-training model for tunnel lining point cloud component segmentation . journal Tunnelling and Underground Space Technology volume 158 , pages 106401 . h...

    2025

  52. [61]

    , author Lovell, L

    author Ye, Z. , author Lovell, L. , author Faramarzi, A. , author Nini \'c , J. , year 2024 . title Sam-based instance segmentation models for the automation of structural damage detection . journal Advanced Engineering Informatics volume 62 , pages 102826 . https://doi.org/10...

    2024

  53. [62]

    , author Zhou, C

    author You, K. , author Zhou, C. , author Ding, L. , author Wang, Y. , year 2025 . title Construction robotics in extreme environments: From earth to space . journal Engineering https://doi.org/10.1016/j.eng.2024.11.037

    2025 doi

  54. [63]

    , author Ji, A

    author Zhang, Z. , author Ji, A. , author Wang, K. , author Zhang, L. , year 2022 . title Unrollingnet: An attention-based deep learning approach for the segmentation of large-scale point clouds of tunnels . journal Automation in Construction volume 142 , pages 104456 . https:...

    2022

  55. [64]

    , author Park, J

    author Zhou, Q.Y. , author Park, J. , author Koltun, V. , year 2018 . title Open3d: A modern library for 3d data processing . journal arXiv preprint arXiv:1801.09847 https://doi.org/10.48550/arXiv.1801.09847

    2018 doi arXiv

  56. [65]

    , author Zhao, W

    author Zhu, Y. , author Zhao, W. , author Li, A. , author Tang, Y. , author Zhou, J. , author Lu, J. , year 2024 . title Flowie: Efficient image enhancement via rectified flow , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , p...

    2024 doi arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.