pith. machine review for the scientific record. sign in

arxiv: 2604.20474 · v1 · submitted 2026-04-22 · 💻 cs.CV

Recognition: unknown

Random Walk on Point Clouds for Feature Detection

Bao Guo, Jian Gao, Shunli Zhang, Yuhe Zhang, Zhikun Tu, Zhi Li

Authors on Pith no claims yet

Pith reviewed 2026-05-10 01:13 UTC · model grok-4.3

classification 💻 cs.CV
keywords point cloudfeature detectionrandom walkneighborhood descriptorgraph-based analysis3D shape outlinecomputer graphicsfeature extraction
0
0 comments X

The pith

A random walk performed on disk-sampled neighborhood graphs extracts feature points from point clouds by jointly modeling spatial distribution, topology, and local geometry.

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

The paper introduces RWoDSN as a two-stage process for locating the points that outline a point-cloud model's shape. It first builds a Disk Sampling Neighborhood descriptor that keeps matrix structure and normal relations while spanning sharp-to-smooth, large-to-small, and textural-to-detailed transitions. It then runs a random walk on the resulting graph so that each point's score reflects spatial, topological, and geometric properties at once. The authors report that this yields a recall of 0.769 (22 percent above the prior best) at a precision of 0.784 and beats both classical and learned competitors on eight standard metrics.

Core claim

Feature extraction is treated as a context-dependent graph problem: a Disk Sampling Neighborhood is formed around each point to preserve neighborhood relations in matrix form, after which a random walk on that neighborhood graph produces a score that simultaneously encodes spatial distribution, topological connectivity, and geometric variation, allowing reliable selection of the points that define the overall shape.

What carries the argument

The Disk Sampling Neighborhood graph, on which a random walk aggregates spatial, topological, and geometric information to rank each point's importance.

If this is right

  • The method produces a recall of 0.769 and precision of 0.784 while handling transitions across scales and feature types.
  • It outperforms both traditional hand-crafted descriptors and deep-learning baselines on eight evaluation metrics.
  • Feature points located this way can serve directly as input for downstream tasks such as registration, reconstruction, and CAD operations.
  • Because the walk operates on an explicitly constructed graph, the approach avoids the need for large training sets or post-processing heuristics.

Where Pith is reading between the lines

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

  • If the graph construction generalizes across acquisition modalities, the same pipeline could be applied to noisy or incomplete scans without retraining.
  • The explicit separation of neighborhood definition from the walk step makes it straightforward to substitute alternative neighborhood samplers and measure their isolated effect on detection quality.
  • Because the scoring is deterministic once the graph is built, the method could be inserted as a lightweight preprocessing stage before learned descriptors are applied.

Load-bearing premise

That the random walk on the DSN graph reliably surfaces feature points by incorporating spatial, topological, and geometric cues without any dataset-specific parameter tuning.

What would settle it

On any standard point-cloud feature-detection benchmark, a measured recall lower than the current best published method would falsify the performance claim.

Figures

Figures reproduced from arXiv: 2604.20474 by Bao Guo, Jian Gao, Shunli Zhang, Yuhe Zhang, Zhikun Tu, Zhi Li.

Figure 1
Figure 1. Figure 1: The results of the baseline methods and the proposed RWoDSN on three typical models with imbalanced sampling patterns (highlighted [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The illustration of the proposed DSN, kNN and the ball neighborhood. 2 [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The illustration of the DSN construction for point [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The DSN of four different points: point A is a corner point, point B is an edge point, point C is a non-feature point near the edge, and point D is a non-feature point on a smooth surface. 3.2. RWoDSN for feature detection A point can be classified as a “feature point” or “non-feature point”, by analyzing its neighborhood. Similarly, the analysis can be extended to the DSNs to detect feature points. Theref… view at source ↗
Figure 5
Figure 5. Figure 5: The illustration of vertex_Matrix, d i, j r and d i, j c : (a) shows the nesting disks, which can be considered as an N × 2π/φ matrix, as shown in (b). (c) shows an example of a non-feature point on a smooth surface, (d) shows an example of a non-feature point located near the edge, and (e) shows an example of an edge point. Therefore, two vectors are built for storing the differences between d i, j r and … view at source ↗
Figure 6
Figure 6. Figure 6: The resulting graph-based DSNs of the points in Fig.2. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The results of our RWoDSN using different parameters (with models and ground truth are shown in Fig.9). 4.4. Quantitative results In this section, we quantitatively evaluate the feature detection results by obtaining the outputs of 6000 models from the ABC dataset using our RWoDSN and the selected baselines [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The results of the baseline methods compared with the proposed RWoDSN. [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative results of our RWoDSN applied to several models from the ABC dataset and real-scanned models. [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: The results of our RWoDSN applied to noisy and simplified models. [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: The results of surface segmentation. 4.8. Efficiency In this section, we compare the efficiency of the baseline methods with the proposed method. Since the input point clouds for PIENet are down-sampled to 8096 points, PIENet[13] is not included in this comparison. All testing was 16 [PITH_FULL_IMAGE:figures/full_fig_p016_11.png] view at source ↗
read the original abstract

The points on the point clouds that can entirely outline the shape of the model are of critical importance, as they serve as the foundation for numerous point cloud processing tasks and are widely utilized in computer graphics and computer-aided design. This study introduces a novel method, RWoDSN, for extracting such feature points, incorporating considerations of sharp-to-smooth transitions, large-to-small scales, and textural-to-detailed features. We approach feature extraction as a two-stage context-dependent analysis problem. In the first stage, we propose a novel neighborhood descriptor, termed the Disk Sampling Neighborhood (DSN), which, unlike traditional spatially and geometrically invariant approaches, preserves a matrix structure while maintaining normal neighborhood relationships. In the second stage, a random walk is performed on the DSN (RWoDSN), yielding a graph-based DSN that simultaneously accounts for the spatial distribution, topological properties, and geometric characteristics of the local surface surrounding each point. This enables the effective extraction of feature points. Experimental results demonstrate that the proposed RWoDSN method achieves a recall of 0.769-22% higher than the current state-of-the-art-alongside a precision of 0.784. Furthermore, it significantly outperforms several traditional and deep-learning techniques across eight evaluation metrics.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 1 minor

Summary. The manuscript introduces RWoDSN, a two-stage method for feature point extraction from point clouds. Stage one defines a Disk Sampling Neighborhood (DSN) descriptor that preserves matrix structure while retaining normal neighborhood relations, unlike traditional invariant approaches. Stage two performs a random walk on the DSN to produce a graph-based representation that simultaneously encodes spatial distribution, topological properties, and geometric characteristics of local surfaces. The authors report that this yields a recall of 0.769 (22% above current SOTA) and precision of 0.784, with significant outperformance over traditional and deep-learning baselines across eight evaluation metrics.

Significance. If the central claims hold after proper validation, the work offers a potentially useful graph-based perspective on multi-aspect feature detection in point clouds, which could benefit downstream tasks in computer graphics and CAD. The DSN construction is a concrete, novel contribution worth exploring for its matrix-preserving property. No machine-checked proofs or parameter-free derivations are present, but the two-stage framing and random-walk integration represent an original synthesis if the implementation details prove reproducible.

major comments (3)
  1. Abstract: The superiority claims (recall 0.769 with 22% improvement, precision 0.784, outperformance on eight metrics) are presented without any reference to the datasets used, the specific baselines compared, implementation details of those baselines, or statistical measures such as error bars or significance tests. This information is load-bearing for the empirical claims and must be supplied for the results to be verifiable.
  2. Abstract (second-stage description): The random walk is asserted to 'simultaneously account for the spatial distribution, topological properties, and geometric characteristics' yet no transition probabilities, walk length, absorption criteria, or explicit mapping from DSN matrix to weighted graph are provided. Without these, it is unclear whether the procedure truly integrates curvature or scale transitions or instead relies on unstated design choices that would make the reported metrics conditional on tuning.
  3. Abstract: The claim that DSN 'preserves a matrix structure while maintaining normal neighborhood relationships' and is 'unlike traditional spatially and geometrically invariant approaches' requires a concrete comparison (e.g., to standard k-NN or ball neighborhoods) and a demonstration that no scale or sampling parameters are implicitly fitted to the evaluation data; otherwise the independence from post-hoc adjustments cannot be assessed.
minor comments (1)
  1. Abstract contains a typographical error: 'state-of-the-art-alongside' should read 'state-of-the-art alongside'.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We agree that the abstract requires additional context to make the empirical claims more verifiable and will revise it to address the points raised while preserving its conciseness. We respond to each major comment below.

read point-by-point responses

  1. Referee: Abstract: The superiority claims (recall 0.769 with 22% improvement, precision 0.784, outperformance on eight metrics) are presented without any reference to the datasets used, the specific baselines compared, implementation details of those baselines, or statistical measures such as error bars or significance tests. This information is load-bearing for the empirical claims and must be supplied for the results to be verifiable.

    Authors: We will revise the abstract to explicitly reference the benchmark datasets and the specific traditional and deep-learning baselines used for comparison. Implementation details of the baselines are already provided in the Experiments section; we will add a brief note in the abstract directing readers there. Regarding statistical measures, we will incorporate error bars or significance tests from our existing analysis (or compute them if needed) to support the reported metrics. revision: yes

  2. Referee: Abstract (second-stage description): The random walk is asserted to 'simultaneously account for the spatial distribution, topological properties, and geometric characteristics' yet no transition probabilities, walk length, absorption criteria, or explicit mapping from DSN matrix to weighted graph are provided. Without these, it is unclear whether the procedure truly integrates curvature or scale transitions or instead relies on unstated design choices that would make the reported metrics conditional on tuning.

    Authors: The transition probabilities, walk length, absorption criteria, and the explicit mapping from the DSN matrix to the weighted graph are fully specified in the Method section. To improve the abstract, we will add a concise clause referencing these elements and how they enable integration of spatial, topological, and geometric properties, while keeping the abstract brief and pointing to the detailed description. revision: yes

  3. Referee: Abstract: The claim that DSN 'preserves a matrix structure while maintaining normal neighborhood relationships' and is 'unlike traditional spatially and geometrically invariant approaches' requires a concrete comparison (e.g., to standard k-NN or ball neighborhoods) and a demonstration that no scale or sampling parameters are implicitly fitted to the evaluation data; otherwise the independence from post-hoc adjustments cannot be assessed.

    Authors: We will revise the abstract to include a direct comparison of DSN against standard k-NN and ball neighborhoods, emphasizing the matrix-structure preservation. We will also add a statement clarifying that DSN construction uses fixed sampling without fitting to evaluation data, with supporting analysis and parameter settings provided in the Method section to demonstrate independence from post-hoc adjustments. revision: yes

Circularity Check

0 steps flagged

No circularity: method is a two-stage descriptor plus walk with external experimental validation

full rationale

The paper presents RWoDSN as a novel DSN matrix descriptor followed by a random walk that produces a graph encoding spatial/topological/geometric properties, then validates via recall/precision on benchmark data. No equations or steps reduce by construction to fitted inputs or self-citations; the central claim is an empirical procedure whose performance is measured against independent test sets rather than derived tautologically from its own parameters. No self-definitional, fitted-prediction, or uniqueness-imported patterns appear in the provided description.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The method rests on the unproven premise that random walks on the constructed DSN graph capture the desired feature properties across scales and transitions; no explicit free parameters or invented physical entities are named in the abstract, but the DSN itself is a new constructed representation.

axioms (1)
  • domain assumption Random walks on a graph derived from local neighborhoods can encode spatial, topological, and geometric information sufficient to distinguish feature points.
    Invoked in the second stage description of RWoDSN.
invented entities (1)
  • Disk Sampling Neighborhood (DSN) no independent evidence
    purpose: A neighborhood descriptor that preserves matrix structure and normal relationships unlike spatially or geometrically invariant methods.
    Introduced as the core of the first stage; no independent evidence outside the paper.

pith-pipeline@v0.9.0 · 5530 in / 1452 out tokens · 32945 ms · 2026-05-10T01:13:03.998117+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

49 extracted references · 45 canonical work pages

  1. [1]

    Deep learning for 3d point clouds: A survey

    Guo Y, Wang H, Hu Q, Liu H, Liu L, Bennamoun M. Deep learning for 3d point clouds: A survey. IEEE Trans Pattern Anal 2020;43(12):4338–64.http://dx.doi.org/10.1109/TPAMI.2020.3005434. 17

    work page doi:10.1109/tpami.2020.3005434 2020

  2. [2]

    Feature extraction from point clouds

    Gumhold S, Wang X, MacLeod RS, et al. Feature extraction from point clouds. In: 10th International Meshing Roundtable. 2001, p. 293–305.https://api.semanticscholar.org/CorpusID:18343879

    2001

  3. [3]

    Detection of closed sharp edges in point clouds using normal estimation and graph theory

    Demarsin K, Vanderstraeten D, V olodine T, Roose D. Detection of closed sharp edges in point clouds using normal estimation and graph theory. Comput Aided Design 2007;39(4):276–83.http://dx.doi.org/10.10 16/j.cad.2006.12.005

    2007

  4. [4]

    A statistical approach for extraction of feature lines from point clouds

    Zhang Y, Geng G, Wei X, Zhang S, Li S. A statistical approach for extraction of feature lines from point clouds. Comput Graph-UK 2016;56:31–45.http://dx.doi.org/10.1016/j.cag.2016.01.004

    work page doi:10.1016/j.cag.2016.01.004 2016

  5. [5]

    Information Fusion 124, 103332 (2025) https://doi.org/10.1016/j.inffus.2025.103332

    Hu E, Sun L. Daniel: A fast and robust consensus maximization method for point cloud registration with high outlier ratios. Inform Sciences 2022;614:563–79.http://dx.doi.org/https://doi.org/10.1016/j.in s.2022.10.086

    work page doi:10.1016/j.in 2022

  6. [6]

    Type-based outlier removal framework for point clouds

    Ge L, Feng J. Type-based outlier removal framework for point clouds. Inform Sciences 2021;580:436–59. http://dx.doi.org/https://doi.org/10.1016/j.ins.2021.08.090

    work page doi:10.1016/j.ins.2021.08.090 2021

  7. [7]

    Deep-learning-based stair detection using 3d point cloud data for preventing walking accidents of the visually impaired

    Matsumura H, Premachandra C. Deep-learning-based stair detection using 3d point cloud data for preventing walking accidents of the visually impaired. IEEE Access 2022;10:56249–55.http://dx.doi.org/10.1109 /ACCESS.2022.3178154

    work page arXiv 2022

  8. [8]

    Depth-gyro sensor-based extended face orientation estimation using deep learn- ing

    Premachandra C, Funahashi Y. Depth-gyro sensor-based extended face orientation estimation using deep learn- ing. IEEE Sensors Journal 2023;23(17):20199–206.http://dx.doi.org/10.1109/JSEN.2023.3296531

    work page doi:10.1109/jsen.2023.3296531 2023

  9. [9]

    Sharp feature detection in point clouds

    Weber C, Hahmann S, Hagen H. Sharp feature detection in point clouds. In: 2010 Shape Modeling International Conference. IEEE; 2010, p. 175–86.http://dx.doi.org/10.1109/SMI.2010.32

    work page doi:10.1109/smi.2010.32 2010

  10. [10]

    Computer Graphics Forum29(2), 419–428 (2010) https://doi.org/10.1111/j.1467-8659.2009.01611.x

    Weinkauf T, Guenther D. Separatrix persistence: extraction of salient edges on surfaces using topological methods. Comput Graph Forum 2009;28(5):1519–28.http://dx.doi.org/10.1111/j.1467-8659.2009 .01528.x

    work page doi:10.1111/j.1467-8659.2009 2009

  11. [11]

    Ec-net: an edge-aware point set consolidation network

    Yu L, Li X, Fu CW, Cohen-Or D, Heng PA. Ec-net: an edge-aware point set consolidation network. In: Proceedings of the European Conference on Computer Vision (ECCV). 2018, p. 386–402.http://dx.doi.o rg/10.1007/978-3-030-01234-2_24

    work page doi:10.1007/978-3-030-01234-2_24 2018

  12. [12]

    Learning part boundaries from 3d point clouds

    Loizou M, Averkiou M, Kalogerakis E. Learning part boundaries from 3d point clouds. Comput Graph Forum 2020;39(5):183–195.http://dx.doi.org/10.1111/cgf.14078.http://dx.doi.org/10.1111/cgf.1 4078

    work page doi:10.1111/cgf.14078.http://dx.doi.org/10.1111/cgf.1 2020

  13. [13]

    Pie-net: Parametric inference of point cloud edges

    Wang X, Xu Y, Xu K, Tagliasacchi A, Zhou B, Mahdavi-Amiri A, et al. Pie-net: Parametric inference of point cloud edges. In: NeurIPS Proceedings 2020; vol. 33. 2020, p. 20167–78.http://dx.doi.org/10.48550/ar Xiv.2007.04883

    work page doi:10.48550/ar 2020

  14. [14]

    Spiral aggregation map (splam): A new descriptor for robust template matching with fast algorithm

    Shih HC, Yu KC. Spiral aggregation map (splam): A new descriptor for robust template matching with fast algorithm. Pattern Recogn 2015;48(5):1707–23.http://dx.doi.org/10.1016/j.patcog.2014.11.004

    work page doi:10.1016/j.patcog.2014.11.004 2015

  15. [15]

    Curvature-based approach for multi-scale feature extraction from 3d meshes and unstructured point clouds

    Ho HT, Gibbins D. Curvature-based approach for multi-scale feature extraction from 3d meshes and unstructured point clouds. IET Comput Vis 2009;3(4):201–12.http://dx.doi.org/10.1049/iet-cvi.2009.0044

    work page doi:10.1049/iet-cvi.2009.0044 2009

  16. [16]

    Extracting Sharp Features from RGB-D Images

    Cao Y, Ju T, Xu J, Hu S. Extracting Sharp Features from RGB-D Images. Comput Graph Forum 2017;36(8):138–52.http://dx.doi.org/10.1111/cgf.13069

    work page doi:10.1111/cgf.13069 2017

  17. [17]

    Robust smooth feature extraction from point clouds

    Daniels JI, Ha LK, Ochotta T, Silva CT. Robust smooth feature extraction from point clouds. In: IEEE International Conference on Shape Modeling and Applications 2007 (SMI’07). 2007, p. 123–36.http: //dx.doi.org/10.1109/SMI.2007.32. 18

    work page doi:10.1109/smi.2007.32 2007

  18. [18]

    Pattern Recog- nition153, 110500 (2024).https://doi.org/https://doi.org/10.1016/j

    Wang Y, Feng HY, Étienne Delorme F, Engin S. An adaptive normal estimation method for scanned point clouds with sharp features. Comput Aided Design 2013;45(11):1333–48.http://dx.doi.org/10.1016/j .cad.2013.06.003

    work page doi:10.1016/j 2013

  19. [19]

    ACM Trans

    Huang QX, Flöry S, Gelfand N, Hofer M, Pottmann H. Reassembling fractured objects by geometric matching. ACM Trans Graph 2006;25(3):569–78.http://dx.doi.org/10.1145/1141911.1141925

    work page doi:10.1145/1141911.1141925 2006

  20. [20]

    Multi-scale tensor voting for feature extraction from unstructured point clouds

    Park MK, Lee SJ, Lee KH. Multi-scale tensor voting for feature extraction from unstructured point clouds. Graph Models 2012;74(4):197–208.http://dx.doi.org/10.1016/j.gmod.2012.04.008

    work page doi:10.1016/j.gmod.2012.04.008 2012

  21. [21]

    Multi- scale feature extraction on point-sampled surfaces

    Pauly M, Keiser R, Gross M. Multi-scale feature extraction on point-sampled surfaces. Comput Graph Forum 2003;22(3):281–9.http://dx.doi.org/10.1111/1467-8659.00675

    work page doi:10.1111/1467-8659.00675 2003

  22. [22]

    Computer Graphics Forum 25(3), 547–556 (2006) https://doi.org/10.1111/ j.1467-8659.2006.00974.x

    Jenke P, Wand M, Bokeloh M, Schilling A, Straßer W. Bayesian point cloud reconstruction. Comput Graph Forum 2006;25(3):379–88.http://dx.doi.org/10.1111/j.1467-8659.2006.00957.x

    work page doi:10.1111/j.1467-8659.2006.00957.x 2006

  23. [23]

    Fast and robust edge extraction in unorganized point clouds

    Bazazian D, Casas JR, Ruiz-Hidalgo J. Fast and robust edge extraction in unorganized point clouds. In: 2015 International Conference on Digital Image Computing: Techniques and Applications (DICTA). 2015, p. 1–8. http://dx.doi.org/10.1109/DICTA.2015.7371262

    work page doi:10.1109/dicta.2015.7371262 2015

  24. [24]

    Edge and corner detection for unorganized 3d point clouds with application to robotic welding

    Ahmed SM, Tan YZ, Chew CM, Mamun AA, Wong FS. Edge and corner detection for unorganized 3d point clouds with application to robotic welding. In: 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). 2018, p. 7350–5.http://dx.doi.org/10.1109/IROS.2018.8593910

    work page doi:10.1109/iros.2018.8593910 2018

  25. [25]

    Robust and accurate feature detection on point clouds

    Liu Z, Xin X, Xu Z, Zhou W, Wang C, Chen R, et al. Robust and accurate feature detection on point clouds. Comput Aided Design 2023;164:103592.http://dx.doi.org/https://doi.org/10.1016/j.cad.2023 .103592

    work page doi:10.1016/j.cad.2023 2023

  26. [26]

    Sglbp: subgraph-based local binary patterns for feature extraction on point clouds

    Guo B, Zhang Y, Gao J, Li C, Hu Y. Sglbp: subgraph-based local binary patterns for feature extraction on point clouds. Comput Graph Forum 2022;41(6):51–66.http://dx.doi.org/10.1111/cgf.14500

    work page doi:10.1111/cgf.14500 2022

  27. [27]

    V oronoi-based curvature and feature estimation from point clouds

    Mérigot Q, Ovsjanikov M, Guibas LJ. V oronoi-based curvature and feature estimation from point clouds. IEEE Trans Vis Comput Graphic 2011;17(6):743–56.http://dx.doi.org/10.1109/TVCG.2010.261

    work page doi:10.1109/tvcg.2010.261 2011

  28. [28]

    V olumetric and multi-view cnns for object classification on 3d data

    Qi CR, Su H, Nießner M, Dai A, Yan M, Guibas LJ. V olumetric and multi-view cnns for object classification on 3d data. In: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2016, p. 5648–56. http://dx.doi.org/10.1109/CVPR.2016.609

    work page doi:10.1109/cvpr.2016.609 2016

  29. [29]

    Shape completion using 3d-encoder-predictor cnns and shape synthesis

    Dai A, Qi CR, Nießner M. Shape completion using 3d-encoder-predictor cnns and shape synthesis. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2017, p. 6545–54.http://dx.doi.o rg/10.1109/CVPR.2017.693

    work page doi:10.1109/cvpr.2017.693 2017

  30. [30]

    Jsenet: joint semantic segmentation and edge detection network for 3d point clouds

    Hu Z, Zhen M, Bai X, Fu H, Tai Cl. Jsenet: joint semantic segmentation and edge detection network for 3d point clouds. In: Proceedings of the European Conference on Computer Vision (ECCV). 2020, p. 222–39. http://dx.doi.org/10.1007/978-3-030-58565-5_14

    work page doi:10.1007/978-3-030-58565-5_14 2020

  31. [31]

    Edc-net: edge detection capsule network for 3d point clouds

    Bazazian D, Parés ME. Edc-net: edge detection capsule network for 3d point clouds. Applied Sciences 2021;11(4):1833.http://dx.doi.org/10.3390/app11041833

    work page doi:10.3390/app11041833 2021

  32. [32]

    Pcednet: a lightweight neural network for fast and interactive edge detection in 3d point clouds

    Himeur CE, Lejemble T, Pellegrini T, Paulin M, Barthe L, Mellado N. Pcednet: a lightweight neural network for fast and interactive edge detection in 3d point clouds. ACM Trans Graph 2021;41(1):1–21.http://dx.d oi.org/10.1145/3481804

    work page doi:10.1145/3481804 2021

  33. [33]

    Def: Deep estimation of sharp geometric features in 3d shapes

    Matveev A, Rakhimov R, Artemov A, Bobrovskikh G, Egiazarian V, Bogomolov E, et al. Def: Deep estimation of sharp geometric features in 3d shapes. ACM Trans Graph 2022;41(4).http://dx.doi.org/10.1145/352 8223.3530140. 19

    work page doi:10.1145/352 2022

  34. [34]

    Vision transformers are parameter- efficient audio-visual learners

    Zhu X, Du D, Chen W, Zhao Z, Nie Y, Han X. Nerve: Neural volumetric edges for parametric curve extraction from point cloud. In: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2023, p. 13601–10.http://dx.doi.org/10.1109/CVPR52729.2023.01307

    work page doi:10.1109/cvpr52729.2023.01307 2023

  35. [35]

    Vision transformers are parameter- efficient audio-visual learners

    Ye Y, Yi R, Gao Z, Zhu C, Cai Z, Xu K. Nef: Neural edge fields for 3d parametric curve reconstruction from multi-view images. In: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2023, p. 8486–95.http://dx.doi.org/10.1109/CVPR52729.2023.00820

    work page doi:10.1109/cvpr52729.2023.00820 2023

  36. [36]

    Random walks and random environments

    Hughes BD. Random walks and random environments. Oxford University Press; 1996

    1996

  37. [37]

    Random walks for image segmentation

    Grady L. Random walks for image segmentation. IEEE Trans Pattern Anal 2006;28(11):1768–83.http: //dx.doi.org/10.1109/TPAMI.2006.233

    work page doi:10.1109/tpami.2006.233 2006

  38. [38]

    Rapid and effective segmentation of 3d models using random walks

    Lai YK, Hu SM, Martin RR, Rosin PL. Rapid and effective segmentation of 3d models using random walks. Comput Aided Geom D 2009;26(6):665–79.http://dx.doi.org/10.1016/j.cagd.2008.09.007

    work page doi:10.1016/j.cagd.2008.09.007 2009

  39. [39]

    Community detection using restrained random-walk similarity

    Okuda M, Satoh S, Sato Y, Kidawara Y. Community detection using restrained random-walk similarity. IEEE Trans Pattern Anal 2021;43(1):89–103.http://dx.doi.org/10.1109/TPAMI.2019.2926033

    work page doi:10.1109/tpami.2019.2926033 2021

  40. [40]

    Meshwalker: deep mesh understanding by random walks

    Lahav A, Tal A. Meshwalker: deep mesh understanding by random walks. ACM Trans Graph 2020;39(6):1–13. http://dx.doi.org/10.1145/3414685.3417806

    work page doi:10.1145/3414685.3417806 2020

  41. [41]

    Cloudwalker: Random walks for 3d point cloud shape analysis

    Mesika A, Ben-Shabat Y, Tal A. Cloudwalker: Random walks for 3d point cloud shape analysis. Comput Graph-UK 2022;106(C):110–8.http://dx.doi.org/10.1016/j.cag.2022.06.001

    work page doi:10.1016/j.cag.2022.06.001 2022

  42. [42]

    Editing conditional radiance fields

    Xiang T, Zhang C, Song Y, Yu J, Cai W. Walk in the cloud: Learning curves for point clouds shape analysis. In: 2021 IEEE/CVF International Conference on Computer Vision (ICCV). 2021, p. 895–904.http://dx.d oi.org/10.1109/ICCV48922.2021.00095

    work page doi:10.1109/iccv48922.2021.00095 2021

  43. [43]

    2021 , url =

    Li R, Lin G, Xie L. Self-point-flow: Self-supervised scene flow estimation from point clouds with optimal transport and random walk. In: 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2021, p. 15572–81.http://dx.doi.org/10.1109/CVPR46437.2021.01532

    work page doi:10.1109/cvpr46437.2021.01532 2021

  44. [44]

    Contour detection in unstructured 3d point clouds

    Hackel T, Wegner JD, Schindler K. Contour detection in unstructured 3d point clouds. In: 2016 IEEE Confer- ence on Computer Vision and Pattern Recognition (CVPR). 2016, p. 1610–8.http://dx.doi.org/10.1109 /CVPR.2016.178

    2016

  45. [45]

    Numerical analysis of differential operators on raw point clouds

    Digne JJ, Morel JM. Numerical analysis of differential operators on raw point clouds. Numerische Mathematik 2014;127(2):255–89.http://dx.doi.org/10.1007/s00211-013-0584-y

    work page doi:10.1007/s00211-013-0584-y 2014

  46. [46]

    Stable and efficient differential estimators on oriented point clouds

    Lejemble T, Coeurjolly D, Barthe L, Mellado N. Stable and efficient differential estimators on oriented point clouds. Comput Graph Forum 2021;40(5):205–16.http://dx.doi.org/10.1111/cgf.14368

    work page doi:10.1111/cgf.14368 2021

  47. [47]

    Abc: A big cad model dataset for geometric deep learning

    Koch S, Matveev A, Jiang Z, Williams F, Artemov A, Burnaev E, et al. Abc: A big cad model dataset for geometric deep learning. In: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2019, p. 9593–603.http://dx.doi.org/10.1109/CVPR.2019.00983

    work page doi:10.1109/cvpr.2019.00983 2019

  48. [48]

    3d is here: Point cloud library (pcl)

    Rusu RB, Cousins S. 3d is here: Point cloud library (pcl). In: 2011 IEEE International Conference on Robotics and Automation. 2011, p. 1–4.http://dx.doi.org/10.1109/ICRA.2011.5980567

    work page doi:10.1109/icra.2011.5980567 2011

  49. [49]

    Feature curve extraction on triangle meshes

    Moscoso Thompson E, Arvanitis G, Moustakas K, Hoang-Xuan N, Nguyen ER, Tran M, et al. Feature curve extraction on triangle meshes. In: 12th EG Workshop 3D Object Retrieval (2019). 2019, p. 1–8.http: //dx.doi.org/10.2312/3dor.20191066. 20

    work page doi:10.2312/3dor.20191066 2019