pith. sign in

arxiv: 2604.02497 · v1 · submitted 2026-04-02 · 💻 cs.CV

Delaunay Canopy: Building Wireframe Reconstruction from Airborne LiDAR Point Clouds via Delaunay Graph

Donghyun Kim , Chanyoung Kim , Youngjoong Kwon , Seong Jae Hwang This is my paper

Pith reviewed 2026-05-13 21:44 UTC · model grok-4.3

classification 💻 cs.CV
keywords wireframe reconstructionDelaunay graphairborne LiDARpoint cloud processingbuilding geometrygeometric prior3D manifold reconstruction
0
0 comments X

The pith

Delaunay graph prior defines an adaptive search space that enables accurate wireframe reconstruction from noisy airborne LiDAR point clouds.

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

The paper proposes Delaunay Canopy to reconstruct building wireframes from airborne LiDAR point clouds. Conventional methods struggle with noise, sparsity, and internal corners because they lack an adaptive search space that exploits the full 3D geometry. The approach builds a Delaunay graph from the points and uses it as a geometric prior. A dedicated scoring step recovers the underlying manifold while producing local curvature signatures that steer subsequent corner and wire selection. Experiments on the Building3D Tallinn and entry-level datasets show state-of-the-art accuracy across varied building shapes.

Core claim

The Delaunay graph serves as a geometric prior that defines a geometrically adaptive search space. Delaunay Graph Scoring reconstructs the underlying geometric manifold and produces region-wise curvature signatures. These signatures, together with corner and wire selection modules, focus computation on high-probability elements and thereby produce accurate wireframe predictions even in regions previously intractable due to noise, sparsity, or internal corners.

What carries the argument

Delaunay Graph Scoring, which reconstructs the geometric manifold from the Delaunay graph and yields region-wise curvature signatures to guide corner and wire selection.

If this is right

  • Wireframe output becomes reliable for structural analysis in dense urban LiDAR surveys that contain sparse or noisy roof regions.
  • Topology-centric building models can be extracted directly from airborne scans without first generating dense meshes.
  • Internal corners and complex roof junctions become recoverable rather than systematically omitted.
  • The same graph-based prior can be reused for related tasks such as facade line extraction or roof segmentation on the same point clouds.

Where Pith is reading between the lines

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

  • Curvature signatures extracted from the Delaunay graph may transfer to other point-cloud tasks such as terrain edge detection or road network reconstruction.
  • The method implies that any sufficiently dense 3D point set could benefit from an initial Delaunay triangulation before learned refinement steps.
  • Because the prior is purely geometric and parameter-light, the approach could scale to very large city-scale LiDAR collections with modest additional compute.

Load-bearing premise

The Delaunay graph derived from the input point cloud supplies a sufficiently adaptive search space that captures the true 3D geometry even when the cloud is noisy, sparse, or contains internal corners.

What would settle it

A controlled ablation on the Building3D Tallinn dataset in which the Delaunay graph and its scoring step are replaced by a uniform or random search space, yet the method still matches or exceeds the reported accuracy.

Figures

Figures reproduced from arXiv: 2604.02497 by Chanyoung Kim, Donghyun Kim, Seong Jae Hwang, Youngjoong Kwon.

Figure 1
Figure 1. Figure 1: We present Delaunay Canopy for building wireframe reconstruction. Compared to traditional methods with overly large search space (left) and 2D height map pro￾jection that loses 3D context (right), our approach scores a Delaunay graph with ge￾ometric priors to select corners and wires, yielding an adaptive (i.e., geometrically context-aware) search space that enables precise wireframe reconstruction. Abstra… view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of geometric context between 2D heightmap and Delaunay scoring. While the 2D heightmap collapses 3D structural details and obscures internal corners (red box), our Delaunay scoring effectively pre￾serves the underlying geometric context. To address this, BWFormer [19] introduced a 2D-to-3D corner de￾tection approach in which the point cloud is projected onto a 2D height map ( [PITH_FULL_IMAGE:f… view at source ↗
Figure 3
Figure 3. Figure 3: Visualization of the edge-wise dihedral angle and vertex-wise corner [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Overall pipeline of Delaunay Canopy. The pipeline begins by utilizing Delaunay triangulation to construct the point cloud into a Delaunay graph. The method then sequentially computes metrics that represent the local curvature around each element, proceeding in the order of face, edge, and vertex. This process encapsulates rich 3D information and introduces a judicious search space into the pipeline. edges … view at source ↗
Figure 5
Figure 5. Figure 5: Comparison between Corner Score Sampling (ours) and Farthest [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparison with other methods. [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparison with the strongest baseline BWFormer [19] [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Robustness to real-world ar￾tifacts. Our method effectively recovers structures even in regions with severe noise or scan voids [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The resulting mesh, reconstructed from the wireframe output generated by the Delaunay Canopy. The wireframe demonstrably encapsulates sufficient geometric information, allowing the resultant mesh to exhibit a highly faithful form. ometric ambiguity, or topologically closed structures. A comprehensive analysis and visualization of these failure cases are provided in the supplementary. 5 Conclusion In this w… view at source ↗
Figure 10
Figure 10. Figure 10: Visualization of performance differences between challenging (perturbed point cloud) and original inputs. The number next to ∆ indicates the absolute performance difference, and the percentage (%) shows the relative decrease compared to the original performance. The notation MODEL∗ signifies a model acting as the feature extractor within the Point2Roof [14] pipeline. D Discussion D.1 Limitations While Del… view at source ↗
Figure 11
Figure 11. Figure 11: Visualization of failure cases. Precise wireframe reconstruction is demonstra￾bly compromised under several key scenarios. (a) When critical regions containing valid corner and wire features are unscanned and consequently absent from the point cloud. (b & c) When the intrinsic surface topology of the point cloud exhibits excessive smoothness. (d) When processing objects characterized by a closed topology,… view at source ↗
Figure 12
Figure 12. Figure 12: Qualitative comparison with the strongest baseline BWFormer [19]. The or￾ange bounding boxes specifically denote the regions where our method achieves a demonstrably superior reconstruction result [PITH_FULL_IMAGE:figures/full_fig_p026_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Wireframe reconstruction results of Delaunay Canopy on Building3D dataset [PITH_FULL_IMAGE:figures/full_fig_p027_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Wireframe reconstruction results of Delaunay Canopy on Building3D dataset [PITH_FULL_IMAGE:figures/full_fig_p028_14.png] view at source ↗
read the original abstract

Reconstructing building wireframe from airborne LiDAR point clouds yields a compact, topology-centric representation that enables structural understanding beyond dense meshes. Yet a key limitation persists: conventional methods have failed to achieve accurate wireframe reconstruction in regions afflicted by significant noise, sparsity, or internal corners. This failure stems from the inability to establish an adaptive search space to effectively leverage the rich 3D geometry of large, sparse building point clouds. In this work, we address this challenge with Delaunay Canopy, which utilizes the Delaunay graph as a geometric prior to define a geometrically adaptive search space. Central to our approach is Delaunay Graph Scoring, which not only reconstructs the underlying geometric manifold but also yields region-wise curvature signatures to robustly guide the reconstruction. Built on this foundation, our corner and wire selection modules leverage the Delaunay-induced prior to focus on highly probable elements, thereby shaping the search space and enabling accurate prediction even in previously intractable regions. Extensive experiments on the Building3D Tallinn city and entry-level datasets demonstrate state-of-the-art wireframe reconstruction, delivering accurate predictions across diverse and complex building geometries.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 1 minor

Summary. The paper proposes 'Delaunay Canopy' for reconstructing building wireframes from airborne LiDAR point clouds. It uses the Delaunay graph as a geometric prior to define an adaptive search space, introduces Delaunay Graph Scoring to reconstruct the geometric manifold and provide region-wise curvature signatures, and employs corner and wire selection modules to focus on probable elements. The method is claimed to achieve state-of-the-art performance on the Building3D Tallinn city and entry-level datasets, handling diverse and complex building geometries including noisy, sparse regions and internal corners.

Significance. If the results hold, the approach could represent a meaningful advance in wireframe reconstruction by providing a geometrically adaptive framework that overcomes limitations of previous methods in challenging conditions. This would enable more accurate topology-centric representations from large-scale LiDAR data, with implications for urban modeling and structural understanding. The use of established Delaunay triangulation as input is a strength, but the novel scoring and selection steps need validation.

major comments (2)
  1. [Abstract] Abstract: The abstract asserts state-of-the-art results on two datasets but supplies no quantitative metrics, ablation studies, or error analysis, which prevents verification of the central claim that the method delivers accurate predictions across diverse geometries.
  2. [Method] Method (Delaunay Graph Scoring and selection modules): The claim that the Delaunay graph defines a sufficiently adaptive search space is load-bearing for performance in sparse/noisy regions with internal corners. The manuscript does not demonstrate how region-wise curvature signatures explicitly suppress spurious long edges (common in Delaunay triangulation of gapped LiDAR data), which risks combinatorial expansion of the search space rather than contraction to probable elements.
minor comments (1)
  1. [Abstract] Abstract: The invented term 'Delaunay Canopy' is used without a one-sentence definition or motivation, reducing immediate clarity for readers.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments. We address each major point below and have revised the manuscript accordingly to enhance clarity and support for our claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The abstract asserts state-of-the-art results on two datasets but supplies no quantitative metrics, ablation studies, or error analysis, which prevents verification of the central claim that the method delivers accurate predictions across diverse geometries.

    Authors: We agree that incorporating quantitative highlights would make the abstract more verifiable. In the revised manuscript we have updated the abstract to report key metrics (e.g., wireframe precision/recall gains on the Building3D Tallinn and entry-level sets) while retaining the high-level narrative. Full ablation studies and error analyses remain in Sections 4 and 5. revision: yes

  2. Referee: [Method] Method (Delaunay Graph Scoring and selection modules): The claim that the Delaunay graph defines a sufficiently adaptive search space is load-bearing for performance in sparse/noisy regions with internal corners. The manuscript does not demonstrate how region-wise curvature signatures explicitly suppress spurious long edges (common in Delaunay triangulation of gapped LiDAR data), which risks combinatorial expansion of the search space rather than contraction to probable elements.

    Authors: We appreciate the referee’s emphasis on this mechanistic detail. Delaunay Graph Scoring assigns region-wise curvature signatures that penalize edges whose length and orientation deviate from the locally estimated manifold; spurious long edges across gaps receive markedly lower scores and are subsequently pruned by the corner and wire selection modules. To make this suppression explicit we have added a new explanatory paragraph and an accompanying figure in Section 3.2 of the revision that visualizes score distributions on gapped regions. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation builds on standard Delaunay input with independent scoring and selection steps

full rationale

The paper's chain starts from the input point cloud, computes the Delaunay graph (a fixed external algorithm), then applies new Delaunay Graph Scoring to produce curvature signatures, followed by separate corner and wire selection modules. No equation or step reduces a claimed prediction to a parameter fitted from the same output, nor does any load-bearing premise collapse to a self-citation or self-definition. The adaptive-search-space claim is presented as an empirical property of the full pipeline rather than a definitional identity.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the domain assumption that Delaunay graphs capture building geometry sufficiently well to guide reconstruction in noisy data; no free parameters or invented entities are explicitly quantified in the abstract.

axioms (1)
  • domain assumption Delaunay graph provides a geometrically adaptive search space suitable for large sparse building point clouds
    Invoked as the foundation for defining the search space and guiding corner/wire selection
invented entities (1)
  • Delaunay Canopy no independent evidence
    purpose: Framework that combines Delaunay graph prior with scoring and selection modules for wireframe reconstruction
    New named method introduced to address limitations of conventional approaches

pith-pipeline@v0.9.0 · 5506 in / 1214 out tokens · 39682 ms · 2026-05-13T21:44:11.632390+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

45 extracted references · 45 canonical work pages

  1. [1]

    In: NeurIPS Datasets and Benchmarks Track (2021)

    Baruch, G., Chen, Z., Dehghan, A., Dimry, T., Feigin, Y., Fu, P., Gebauer, T., Joffe, B., Kurz, D., Schwartz, A., Shulman, E.: ARKitscenes - a diverse real-world dataset for 3d indoor scene understanding using mobile RGB-d data. In: NeurIPS Datasets and Benchmarks Track (2021)

    work page 2021

  2. [2]

    ACM Transactions on Graphics (TOG)39(5), 1–14 (2020)

    Bauchet, J.P., Lafarge, F.: Kinetic shape reconstruction. ACM Transactions on Graphics (TOG)39(5), 1–14 (2020)

    work page 2020

  3. [3]

    IEEE transactions on visualization and computer graphics (2002)

    Bernardini, F., Mittleman, J., Rushmeier, H., Silva, C., Taubin, G.: The ball- pivoting algorithm for surface reconstruction. IEEE transactions on visualization and computer graphics (2002)

    work page 2002

  4. [4]

    ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences4, 161–166 (2017)

    Buyuksalih, I., Bayburt, S., Buyuksalih, G., Baskaraca, A., Karim, H., Rahman, A.A.: 3d modelling and visualization based on the unity game engine–advantages and challenges. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences4, 161–166 (2017)

    work page 2017

  5. [5]

    In: Edsger Wybe Dijkstra: his life, work, and legacy, pp

    Dijkstra, E.W.: A note on two problems in connexion with graphs. In: Edsger Wybe Dijkstra: his life, work, and legacy, pp. 287–290 (2022)

    work page 2022

  6. [6]

    ACM Transactions on Graphics (ToG)35(4), 1–13 (2016)

    Dou, M., Khamis, S., Degtyarev, Y., Davidson, P., Fanello, S.R., Kowdle, A., Es- colano, S.O., Rhemann, C., Kim, D., Taylor, J., et al.: Fusion4d: Real-time perfor- mance capture of challenging scenes. ACM Transactions on Graphics (ToG)35(4), 1–13 (2016)

    work page 2016

  7. [7]

    INTERNATIONAL ARCHIVES OF PHOTOGRAMMETRY REMOTE SENS- ING AND SPATIAL INFORMATION SCIENCES34(4), 225–230 (2002)

    Forberg, A., Mayer, H.: Generalization of 3d building data based on scale-spaces. INTERNATIONAL ARCHIVES OF PHOTOGRAMMETRY REMOTE SENS- ING AND SPATIAL INFORMATION SCIENCES34(4), 225–230 (2002)

    work page 2002

  8. [8]

    In: CVPR (2012)

    Geiger, A., Lenz, P., Urtasun, R.: Are we ready for autonomous driving? the kitti vision benchmark suite. In: CVPR (2012)

    work page 2012

  9. [9]

    Semantic3D.net: A new Large-scale Point Cloud Classification Benchmark

    Hackel, T., Savinov, N., Ladicky, L., Wegner, J.D., Schindler, K., Pollefeys, M.: Semantic3d. net: A new large-scale point cloud classification benchmark. arXiv preprint arXiv:1704.03847 (2017)

    work page Pith review arXiv 2017

  10. [10]

    Remote Sensing14(9), 2254 (2022)

    Huang, J., Stoter, J., Peters, R., Nan, L.: City3d: Large-scale building reconstruc- tion from airborne lidar point clouds. Remote Sensing14(9), 2254 (2022)

    work page 2022

  11. [11]

    In: CVPR (2024)

    Huang, S., Wang, R., Guo, B., Yang, H.: Pbwr: Parametric-building-wireframe reconstruction from aerial lidar point clouds. In: CVPR (2024)

    work page 2024

  12. [12]

    In: CVPR (2018)

    Huang, X., Cheng, X., Geng, Q., Cao, B., Zhou, D., Wang, P., Lin, Y., Yang, R.: The apolloscape dataset for autonomous driving. In: CVPR (2018)

    work page 2018

  13. [13]

    In: CVPR (2019)

    Koch, S., Matveev, A., Jiang, Z., Williams, F., Artemov, A., Burnaev, E., Alexa, M., Zorin, D., Panozzo, D.: Abc: A big cad model dataset for geometric deep learning. In: CVPR (2019)

    work page 2019

  14. [14]

    ISPRS Journal of Photogrammetry and Remote Sensing (2022)

    Li, L., Song, N., Sun, F., Liu, X., Wang, R., Yao, J., Cao, S.: Point2roof: End-to- end 3d building roof modeling from airborne lidar point clouds. ISPRS Journal of Photogrammetry and Remote Sensing (2022)

    work page 2022

  15. [15]

    In: CVPR (2024)

    Li, W., Yang, H., Hu, Z., Zheng, J., Xia, G.S., He, C.: 3d building reconstruction from monocular remote sensing images with multi-level supervisions. In: CVPR (2024)

    work page 2024

  16. [16]

    In: CVPR (2023)

    Lin, H., Zheng, X., Li, L., Chao, F., Wang, S., Wang, Y., Tian, Y., Ji, R.: Meta architecture for point cloud analysis. In: CVPR (2023)

    work page 2023

  17. [17]

    arXiv preprint arXiv:2103.02766 , year=

    Liu, Y., D’Aronco, S., Schindler, K., Wegner, J.D.: Pc2wf: 3d wireframe recon- struction from raw point clouds. arXiv preprint arXiv:2103.02766 (2021)

    work page arXiv 2021

  18. [18]

    ISPRS Journal of Photogrammetry and Remote Sensing (2024) 16 D

    Liu, Y., Obukhov, A., Wegner, J.D., Schindler, K.: Point2building: Reconstructing buildings from airborne lidar point clouds. ISPRS Journal of Photogrammetry and Remote Sensing (2024) 16 D. Kim et al

    work page 2024

  19. [19]

    In: CVPR (2025)

    Liu, Y., Zhu, L., Ye, H., Huang, S., Gao, X., Zheng, X., Shen, S.: Bwformer: Build- ing wireframe reconstruction from airborne lidar point cloud with transformer. In: CVPR (2025)

    work page 2025

  20. [20]

    In: AAAI (2021)

    Luo, Y., Mi, Z., Tao, W.: Deepdt: Learning geometry from delaunay triangulation for surface reconstruction. In: AAAI (2021)

    work page 2021

  21. [21]

    Rethinking network design and local geometry in point cloud: A simple residual mlp framework.arXiv preprint arXiv:2202.07123, 2022

    Ma, X., Qin, C., You, H., Ran, H., Fu, Y.: Rethinking network design and lo- cal geometry in point cloud: A simple residual mlp framework. arXiv preprint arXiv:2202.07123 (2022)

    work page arXiv 2022

  22. [22]

    In: CVPR (2020)

    Mahmud, J., Price, T., Bapat, A., Frahm, J.M.: Boundary-aware 3d building re- construction from a single overhead image. In: CVPR (2020)

    work page 2020

  23. [23]

    In: CVPR (2019)

    Mao, J., Wang, X., Li, H.: Interpolated convolutional networks for 3d point cloud understanding. In: CVPR (2019)

    work page 2019

  24. [24]

    Frontiers in Robotics and AI7, 600387 (2021)

    Mateo, C.M., Corrales, J.A., Mezouar, Y.: Hierarchical, dense and dynamic 3d re- construction based on vdb data structure for robotic manipulation tasks. Frontiers in Robotics and AI7, 600387 (2021)

    work page 2021

  25. [25]

    In: 2011 10th IEEE international symposium on mixed and augmented reality

    Newcombe, R.A., Izadi, S., Hilliges, O., Molyneaux, D., Kim, D., Davison, A.J., Kohi, P., Shotton, J., Hodges, S., Fitzgibbon, A.: Kinectfusion: Real-time dense surface mapping and tracking. In: 2011 10th IEEE international symposium on mixed and augmented reality. pp. 127–136. Ieee (2011)

    work page 2011

  26. [26]

    In: Computer Vision–ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part II

    Pang, Y., Wang, W., Tay, F.E., Liu, W., Tian, Y., Yuan, L.: Masked autoencoders for point cloud self-supervised learning. In: Computer Vision–ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part II. pp. 604–621. Springer (2022)

    work page 2022

  27. [27]

    In: CVPR (2017)

    Qi, C.R., Su, H., Mo, K., Guibas, L.J.: Pointnet: Deep learning on point sets for 3d classification and segmentation. In: CVPR (2017)

    work page 2017

  28. [28]

    Qian, G., Li, Y., Peng, H., Mai, J., Hammoud, H., Elhoseiny, M., Ghanem, B.: Pointnext: Revisiting pointnet++ with improved training and scaling strategies (2022)

    work page 2022

  29. [29]

    In: CVPR (2021)

    Rakotosaona, M.J., Guerrero, P., Aigerman, N., Mitra, N.J., Ovsjanikov, M.: Learning delaunay surface elements for mesh reconstruction. In: CVPR (2021)

    work page 2021

  30. [30]

    In: NeurIPS Datasets and Benchmarks Track (2021)

    Ramakrishnan, S.K., Gokaslan, A., Wijmans, E., Maksymets, O., Clegg, A., Turner, J.M., Undersander, E., Galuba, W., Westbury, A., Chang, A.X., Savva, M., Zhao, Y., Batra, D.: Habitat-matterport 3d dataset (HM3d): 1000 large-scale 3d environments for embodied AI. In: NeurIPS Datasets and Benchmarks Track (2021)

    work page 2021

  31. [31]

    In: ECCV (2022)

    Sarlin, P.E., Dusmanu, M., Schönberger, J.L., Speciale, P., Gruber, L., Larsson, V., Miksik, O., Pollefeys, M.: Lamar: Benchmarking localization and mapping for augmented reality. In: ECCV (2022)

    work page 2022

  32. [32]

    In: ICCV (2021)

    Selvaraju, P., Nabail, M., Loizou, M., Maslioukova, M., Averkiou, M., Andreou, A., Chaudhuri, S., Kalogerakis, E.: Buildingnet: Learning to label 3d buildings. In: ICCV (2021)

    work page 2021

  33. [33]

    Son, S., Gadelha, M., Zhou, Y., Xu, Z., Lin, M., Zhou, Y.: DMesh: A differentiable mesh representation (2024)

    work page 2024

  34. [34]

    In: CVPR (2023)

    Wang, R., Huang, S., Yang, H.: Building3d: A urban-scale dataset and benchmarks for learning roof structures from point clouds. In: CVPR (2023)

    work page 2023

  35. [35]

    arXiv preprint arXiv:1612.00423 (2016)

    Wang, S., Bai, M., Mattyus, G., Chu, H., Luo, W., Yang, B., Liang, J., Cheverie, J., Fidler, S., Urtasun, R.: Torontocity: Seeing the world with a million eyes. arXiv preprint arXiv:1612.00423 (2016)

    work page arXiv 2016

  36. [36]

    ACM Transactions on Graphics (tog)38(5), 1–12 (2019) Delaunay Canopy 17

    Wang, Y., Sun, Y., Liu, Z., Sarma, S.E., Bronstein, M.M., Solomon, J.M.: Dynamic graph cnn for learning on point clouds. ACM Transactions on Graphics (tog)38(5), 1–12 (2019) Delaunay Canopy 17

    work page 2019

  37. [37]

    In: CVPR (2019)

    Wu, W., Qi, Z., Fuxin, L.: Pointconv: Deep convolutional networks on 3d point clouds. In: CVPR (2019)

    work page 2019

  38. [38]

    In: CVPR (2024)

    Wu, X., Jiang, L., Wang, P.S., Liu, Z., Liu, X., Qiao, Y., Ouyang, W., He, T., Zhao, H.: Point transformer v3: Simpler faster stronger. In: CVPR (2024)

    work page 2024

  39. [39]

    Wu, X., Lao, Y., Jiang, L., Liu, X., Zhao, H.: Point transformer v2: Grouped vector attention and partition-based pooling (2022)

    work page 2022

  40. [40]

    In: CVPR (2015)

    Wu, Z., Song, S., Khosla, A., Yu, F., Zhang, L., Tang, X., Xiao, J.: 3d shapenets: A deep representation for volumetric shapes. In: CVPR (2015)

    work page 2015

  41. [41]

    In: ICCV (2021)

    Xiang, T., Zhang, C., Song, Y., Yu, J., Cai, W.: Walk in the cloud: Learning curves for point clouds shape analysis. In: ICCV (2021)

    work page 2021

  42. [42]

    In: ICCV (2023)

    Zhang, C., Yuan, G., Tao, W.: Dmnet: Delaunay meshing network for 3d shape representation. In: ICCV (2023)

    work page 2023

  43. [43]

    Advances in neural information processing systems35, 27061–27074 (2022)

    Zhang, R., Guo, Z., Gao, P., Fang, R., Zhao, B., Wang, D., Qiao, Y., Li, H.: Point- m2ae: multi-scale masked autoencoders for hierarchical point cloud pre-training. Advances in neural information processing systems35, 27061–27074 (2022)

    work page 2022

  44. [44]

    In: ICCV (2021)

    Zhao, H., Jiang, L., Jia, J., Torr, P.H., Koltun, V.: Point transformer. In: ICCV (2021)

    work page 2021

  45. [45]

    In: ICLR (2021) 18 D

    Zhu, X., Su, W., Lu, L., Li, B., Wang, X., Dai, J.: Deformable detr: Deformable transformers for end-to-end object detection. In: ICLR (2021) 18 D. Kim et al. A Further Implementation Specifics A.1 Training Details for Corner and Wire Selection To train our corner and wire selection modules, we formulate the objective func- tion based on set prediction. L...

    work page 2021