pith. sign in

arxiv: 2605.02098 · v2 · pith:IEBVXZKRnew · submitted 2026-05-03 · 💻 cs.CV

From Spherical to Gaussian: A Comparative Analysis of Point Cloud Cropping Strategies in Large-Scale 3D Environments

Maximilian Kellner , Dominik Merkle , Michael Brunklaus , Alexander Reiterer This is my paper

Pith reviewed 2026-05-22 09:47 UTC · model grok-4.3

classification 💻 cs.CV
keywords point cloud cropping3D semantic segmentationGaussian croppingspherical cropslarge-scale point cloudsoutdoor environmentsdeep learning
0
0 comments X

The pith

Gaussian cropping strategies improve semantic segmentation accuracy over spherical crops on large outdoor 3D point clouds.

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

Large 3D point clouds must be split into smaller subclouds for neural network processing, but conventional spherical crops discard surrounding geometric context. The paper tests exponential, Gaussian, and linear alternatives that support larger spatial extents while holding point counts roughly constant. Across three model architectures and multiple indoor and outdoor datasets, the non-spherical methods raise performance, with Gaussian cropping producing the strongest gains and new state-of-the-art numbers on expansive outdoor scenes. A sympathetic reader would care because better context preservation at fixed computational cost could raise reliability in downstream tasks that rely on accurate 3D scene labels.

Core claim

Replacing spherical cropping with Gaussian, exponential, or linear strategies allows subclouds to cover larger physical areas at comparable point counts; when these subclouds are fed to standard 3D segmentation networks, accuracy rises, most markedly on large-scale outdoor environments, and new state-of-the-art results are reached.

What carries the argument

Gaussian cropping geometry that samples points according to a Gaussian distribution to enlarge the covered volume without increasing the total point count.

If this is right

  • Gaussian cropping yields higher segmentation accuracy than spherical cropping on outdoor datasets.
  • The same gain appears across multiple network architectures without architecture-specific redesign.
  • Indoor scenes benefit less, indicating the advantage scales with scene extent.
  • New state-of-the-art numbers are obtained on standard large-scale outdoor benchmarks simply by changing the cropping routine.

Where Pith is reading between the lines

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

  • If larger context at fixed point count is the driver, then multi-scale or adaptive cropping schedules could further reduce context loss without extra memory.
  • The same geometric principle might transfer to other dense 3D tasks such as instance segmentation or surface reconstruction where boundary context matters.
  • Parameter sweeps over the Gaussian spread could reveal an optimal scale per environment type without changing network capacity.

Load-bearing premise

Observed performance differences arise from the shape of the crop region itself rather than from uncontrolled variations in point density, code implementation, or per-dataset hyperparameter tuning.

What would settle it

Re-run the same models on the same data after forcing every cropping method to produce identical point-density statistics and identical code paths; if the accuracy gap vanishes, the claim is falsified.

Figures

Figures reproduced from arXiv: 2605.02098 by Alexander Reiterer, Dominik Merkle, Maximilian Kellner, Michael Brunklaus.

Figure 1
Figure 1. Figure 1: Different architectures tested on S3DIS [2] Area view at source ↗
Figure 2
Figure 2. Figure 2: Probability of a point being selected depending on view at source ↗
Figure 4
Figure 4. Figure 4: Point cardinality for subclouds using a voxel size of 2 cm on the S3DIS dataset. Shaded regions represent the min-max view at source ↗
Figure 5
Figure 5. Figure 5: Influence on training using different point cropping view at source ↗
Figure 5
Figure 5. Figure 5: Influence on training using different point cropping [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Performance validation using different cropping view at source ↗
Figure 7
Figure 7. Figure 7: Performance analysis of voxel size (cm) variation [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The influence of the probability between all points view at source ↗
read the original abstract

Large-scale 3D point clouds can consist of hundreds of millions of points. Even after downsampling, these point clouds are too large for modern 3D neural networks. In order to develop a semantic understanding of the scene, the point clouds are divided into smaller subclouds that can be processed. Typically, this division is done using spherical crops, resulting in a loss of surrounding geometric context. To address this issue, we propose alternative methods that produce subclouds with larger crop sizes while maintaining a similar number of points. Specifically, we compare exponential, Gaussian, and linear cropping methods with the spherical method. We evaluated three 3D deep learning model architectures using multiple indoor and outdoor environment datasets. Our results demonstrate that altering the cropping strategy can enhance model performance, especially for large-scale outdoor scenes, yielding new state-of-the-art results. Code is available at https://github.com/mvg-inatech/point_cloud_cropping

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 / 2 minor

Summary. The manuscript compares spherical cropping to exponential, Gaussian, and linear alternatives for dividing large 3D point clouds into subclouds suitable for neural network processing. The alternatives are designed to retain more surrounding context while keeping a comparable number of points. Experiments across three 3D architectures and multiple indoor/outdoor datasets report performance gains, especially for large-scale outdoor scenes, with claims of new state-of-the-art results.

Significance. If performance differences can be isolated to crop geometry rather than sampling density or implementation details, the work offers a lightweight way to improve context preservation in large-scale point cloud tasks such as semantic segmentation. Open-sourced code aids reproducibility and potential follow-up studies.

major comments (2)
  1. [Abstract and experimental protocol] Abstract and experimental protocol: The claim that alternative cropping methods maintain 'a similar number of points' is not accompanied by a description of the point selection procedure inside each crop region. Without specifying whether uniform sampling, farthest-point sampling, rejection sampling, or another method is applied uniformly across spherical, exponential, Gaussian, and linear crops, differences in local density profiles cannot be ruled out as a confound. This directly affects attribution of gains to crop shape, especially for the outdoor SOTA results where scale amplifies any uncontrolled variation.
  2. [Results section] Results section: While gains are reported across models and datasets, the manuscript provides no error bars, statistical significance tests, or ablations that hold point count and augmentation pipeline fixed while varying only crop boundary. This leaves open whether the central empirical claim rests on geometry or on secondary factors in the data preparation pipeline.
minor comments (2)
  1. [Methods] The mathematical definitions of the exponential, Gaussian, and linear cropping functions would benefit from explicit equations in the methods section to allow exact reproduction.
  2. [Figures] Figure captions describing crop visualizations should note the sampling density used inside each shape to aid interpretation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript comparing cropping strategies for large-scale point clouds. We address each major comment below and will revise the manuscript to improve clarity and empirical rigor where feasible.

read point-by-point responses

  1. Referee: [Abstract and experimental protocol] The claim that alternative cropping methods maintain 'a similar number of points' is not accompanied by a description of the point selection procedure inside each crop region. Without specifying whether uniform sampling, farthest-point sampling, rejection sampling, or another method is applied uniformly across spherical, exponential, Gaussian, and linear crops, differences in local density profiles cannot be ruled out as a confound.

    Authors: We agree that explicit details on point selection are necessary to attribute performance differences to crop geometry. In our implementation, cropping strategies define geometric boundaries, and every point falling strictly inside the boundary is retained with no further subsampling (uniform, FPS, or otherwise) applied inside the region. Crop parameters for the exponential, Gaussian, and linear methods are chosen via dataset-wide statistics to yield average point counts comparable to spherical crops. The identical inclusion rule is used for all four strategies. We will add a precise description of this procedure to the experimental protocol section in the revision. revision: yes

  2. Referee: [Results section] While gains are reported across models and datasets, the manuscript provides no error bars, statistical significance tests, or ablations that hold point count and augmentation pipeline fixed while varying only crop boundary. This leaves open whether the central empirical claim rests on geometry or on secondary factors in the data preparation pipeline.

    Authors: We recognize that error bars, significance testing, and targeted ablations would strengthen isolation of the crop-boundary effect. The reported experiments already kept the augmentation pipeline and target point-count distribution fixed across methods, varying only boundary shape. However, the computational cost of retraining multiple 3D architectures on large outdoor datasets precluded repeated runs. In the revision we will add a dedicated ablation that strictly controls point count and pipeline while varying only the boundary, and we will report variance from a limited set of additional seeds where compute permits; full statistical testing may remain constrained by resources. revision: partial

Circularity Check

0 steps flagged

No circularity in empirical cropping comparison

full rationale

The paper is an empirical comparative study that evaluates spherical, exponential, Gaussian, and linear cropping strategies on public indoor and outdoor point cloud datasets using three standard 3D network architectures. Performance claims rest on direct experimental measurements of model accuracy after applying each cropping method while attempting to hold point count approximately constant. No derivation chain, fitted parameters renamed as predictions, self-referential equations, or load-bearing self-citations appear in the provided text. The methodology is self-contained against external benchmarks because results are obtained by running the same models on fixed public data splits with the proposed cropping variants, and code is released for independent verification.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No new free parameters, axioms, or invented entities are introduced; the work relies on standard assumptions of point-cloud deep learning and existing model architectures.

pith-pipeline@v0.9.0 · 5702 in / 940 out tokens · 43270 ms · 2026-05-22T09:47:34.566841+00:00 · methodology

Review history (2 revisions) →

discussion (0)

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

Reference graph

Works this paper leans on

79 extracted references · 79 canonical work pages · 17 internal anchors

  1. [1]

    C. R. Qi, H. Su, K. Mo, L. J. Guibas, Point- Net: Deep Learning on Point Sets for 3D Clas- sification and Segmentation, arXiv:1612.00593 [cs] (Feb. 2016).doi:10.48550/arXiv.1612. 00593. URLhttp://arxiv.org/abs/1612.00593

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1612 2016

  2. [2]

    Joint 2D-3D-Semantic Data for Indoor Scene Understanding

    I. Armeni, S. Sax, A. R. Zamir, S. Savarese, Joint 2D-3D-Semantic Data for Indoor Scene Understanding, arXiv:1702.01105 [cs] (Apr. 2017).doi:10.48550/arXiv.1702.01105. URLhttp://arxiv.org/abs/1702.01105

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1702.01105 2017

  3. [3]

    Armeni, O

    I. Armeni, O. Sener, A. R. Zamir, H. Jiang, I. Brilakis, M. Fischer, S. Savarese, 3D Seman- tic Parsing of Large-Scale Indoor Spaces, in: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, Las Vegas, NV, USA, 2016, pp. 1534–1543. doi:10.1109/CVPR.2016.170. URLhttp://ieeexplore.ieee.org/ document/7780539/

    work page doi:10.1109/cvpr.2016.170 2016

  4. [4]

    G. Qian, Y. Li, H. Peng, J. Mai, H. A. A. K. Hammoud, M. Elhoseiny, B. Ghanem, PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies, arXiv:2206.04670 [cs] (Oct. 2022).doi:10. 48550/arXiv.2206.04670. URLhttp://arxiv.org/abs/2206.04670

    work page arXiv 2022

  5. [5]

    Thomas, Y.-H

    H. Thomas, Y.-H. H. Tsai, T. D. Bar- foot, J. Zhang, KPConvX: Modernizing Ker- nel Point Convolution with Kernel Attention, arXiv:2405.13194 [cs] (May 2024).doi:10. 48550/arXiv.2405.13194. URLhttp://arxiv.org/abs/2405.13194

    work page arXiv 2024

  6. [6]

    Y. Guo, H. Wang, Q. Hu, H. Liu, L. Liu, M. Bennamoun, Deep Learning for 3D Point Clouds: A Survey, IEEE Transac- tions on Pattern Analysis and Machine Intelligence 43 (12) (2021) 4338–4364. doi:10.1109/TPAMI.2020.3005434. URLhttps://ieeexplore.ieee.org/ document/9127813/

    work page doi:10.1109/tpami.2020.3005434 2021

  7. [7]

    Zhang, Y

    R. Zhang, Y. Wu, W. Jin, X. Meng, Deep-Learning-Based Point Cloud Se- mantic Segmentation: A Survey, Elec- tronics 12 (17) (2023) 3642.doi: 10.3390/electronics12173642. URLhttps://www.mdpi.com/2079-9292/ 12/17/3642

    work page doi:10.3390/electronics12173642 2023

  8. [8]

    ALTRO: A Fast Solver for Constrained Trajectory Optimization,

    A. Milioto, I. Vizzo, J. Behley, C. Stach- niss, RangeNet ++: Fast and Accurate LiDAR Semantic Segmentation, in: 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, Macau, China, 2019, pp. 4213–4220. doi:10.1109/IROS40897.2019.8967762. URLhttps://ieeexplore.ieee.org/ document/8967762/

    work page doi:10.1109/iros40897.2019.8967762 2019

  9. [9]

    Cortinhal, G

    T. Cortinhal, G. Tzelepis, E. Erdal Aksoy, SalsaNext: Fast, Uncertainty-Aware Seman- tic Segmentation of LiDAR Point Clouds, in: G. Bebis, Z. Yin, E. Kim, J. Bender, K. Subr, B. C. Kwon, J. Zhao, D. Kalkofen, G. Baciu (Eds.), Advances in Visual Com- puting, Vol. 12510, Springer International Publishing, Cham, 2020, pp. 207–222, series Title: Lecture Notes...

    work page doi:10.1007/978-3-030-64559-5_16 2020

  10. [10]

    B. Wu, A. Wan, X. Yue, K. Keutzer, Squeeze- Seg: Convolutional Neural Nets with Recur- rent CRF for Real-Time Road-Object Seg- mentation from 3D LiDAR Point Cloud, arXiv:1710.07368 [cs] (Oct. 2017).doi:10. 48550/arXiv.1710.07368. URLhttp://arxiv.org/abs/1710.07368

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  11. [11]

    B. Wu, X. Zhou, S. Zhao, X. Yue, K. Keutzer, SqueezeSegV2: Improved Model Structure and Unsupervised Domain Adaptation for Road- Object Segmentation from a LiDAR Point Cloud, in: 2019 International Conference on Robotics and Automation (ICRA), IEEE, Montreal, QC, Canada, 2019, pp. 4376–4382. doi:10.1109/ICRA.2019.8793495. URLhttps://ieeexplore.ieee.org/ d...

    work page doi:10.1109/icra.2019.8793495 2019

  12. [12]

    E. E. Aksoy, S. Baci, S. Cavdar, SalsaNet: Fast Road and Vehicle Segmentation in Li- DAR Point Clouds for Autonomous Driving, arXiv:1909.08291 [cs] (Sep. 2019).doi:10. 48550/arXiv.1909.08291. URLhttp://arxiv.org/abs/1909.08291

    work page arXiv 1909

  13. [13]

    In: Proceedings of the IEEE/CVF Conference on Computer 25 Vision and Pattern Recognition, pp

    Y. Zhang, Z. Zhou, P. David, X. Yue, Z. Xi, B. Gong, H. Foroosh, PolarNet: An Improved Grid Representation for Online LiDAR Point Clouds Semantic Segmentation, in: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, Seattle, WA, USA, 2020, pp. 9598–9607. doi:10.1109/CVPR42600.2020.00962. URLhttps://ieeexplore.ieee.org/ docume...

    work page doi:10.1109/cvpr42600.2020.00962 2020

  14. [14]

    Avidan, G

    G.Shi, R.Li, C.Ma, PillarNet: Real-Timeand High-Performance Pillar-Based 3D Object De- tection, in: S. Avidan, G. Brostow, M. Cissé, G. M. Farinella, T. Hassner (Eds.), Computer Vision – ECCV 2022, Vol. 13670, Springer Nature Switzerland, Cham, 2022, pp. 35–52, series Title: Lecture Notes in Computer Sci- ence.doi:10.1007/978-3-031-20080-9_3. URLhttps://l...

    work page doi:10.1007/978-3-031-20080-9_3 2022

  15. [15]

    Lee, Matthew Tan, Yuke Zhu, and Jeannette Bohg

    M. Gerdzhev, R. Razani, E. Taghavi, L. Bingbing, TORNADO-Net: mulTi- view tOtal vaRiatioN semAntic segmen- tation with Diamond inceptiOn module, in: 2021 IEEE International Conference on Robotics and Automation (ICRA), IEEE, Xi’an, China, 2021, pp. 9543–9549. doi:10.1109/ICRA48506.2021.9562041. URLhttps://ieeexplore.ieee.org/ document/9562041/

    work page doi:10.1109/icra48506.2021.9562041 2021

  16. [16]

    K. Chen, R. Oldja, N. Smolyanskiy, S. Birch- field, A. Popov, D. Wehr, I. Eden, J. Pehserl, MVLidarNet: Real-Time Multi-Class Scene Understanding for Autonomous Driving Using Multiple Views, arXiv:2006.05518 [cs] (Aug. 2020).doi:10.48550/arXiv.2006.05518. URLhttp://arxiv.org/abs/2006.05518

    work page doi:10.48550/arxiv.2006.05518 2006

  17. [17]

    Y. A. Alnaggar, M. Afifi, K. Amer, M. Elhelw, Multi Projection Fusion for Real-time Seman- tic Segmentation of 3D LiDAR Point Clouds, arXiv:2011.01974 [cs] (Nov. 2020).doi:10. 48550/arXiv.2011.01974. URLhttp://arxiv.org/abs/2011.01974

    work page arXiv 2011

  18. [18]

    Çiçek, A

    Ö. Çiçek, A. Abdulkadir, S. S. Lienkamp, T. Brox, O. Ronneberger, 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation, in: S. Ourselin, L. Joskowicz, M. R. Sabuncu, G. Unal, W. Wells (Eds.), Medical Image Computing and Computer- Assisted Intervention – MICCAI 2016, Springer International Publishing, Cham, 2016, pp. 424–432

    work page 2016

  19. [19]

    SEGCloud: Semantic Segmentation of 3D Point Clouds

    L.P.Tchapmi, C.B.Choy, I.Armeni, J.Gwak, S. Savarese, SEGCloud: Semantic Segmen- tation of 3D Point Clouds, arXiv:1710.07563 [cs] (Oct. 2017).doi:10.48550/arXiv.1710. 07563. URLhttp://arxiv.org/abs/1710.07563

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1710 2017

  20. [20]

    Graham, M

    B. Graham, M. Engelcke, L. V. D. Maaten, 3D Semantic Segmentation with Submanifold Sparse Convolutional Networks, in: 2018 IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, IEEE, Salt Lake City, UT, USA, 2018, pp. 9224–9232. doi:10.1109/CVPR.2018.00961. URLhttps://ieeexplore.ieee.org/ document/8579059/

    work page doi:10.1109/cvpr.2018.00961 2018

  21. [21]

    C. Choy, J. Gwak, S. Savarese, 4D Spatio- Temporal ConvNets: Minkowski Convolu- tional Neural Networks, in: 2019 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), IEEE, Long Beach, CA, USA, 2019, pp. 3070–3079. doi:10.1109/CVPR.2019.00319. URLhttps://ieeexplore.ieee.org/ document/8953494/

    work page doi:10.1109/cvpr.2019.00319 2019

  22. [22]

    Contributors, Spconv: Spatially Sparse Convolution Library (2022)

    S. Contributors, Spconv: Spatially Sparse Convolution Library (2022). URLhttps://github.com/traveller59/ spconv

    work page 2022

  23. [23]

    X. Ding, X. Zhang, J. Han, G. Ding, Scal- ing Up Your Kernels to 31×31: Revisiting Large Kernel Design in CNNs, in: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, New Orleans, LA, USA, 2022, pp. 11953–11965. doi:10.1109/CVPR52688.2022.01166. URLhttps://ieeexplore.ieee.org/ document/9880273/

    work page doi:10.1109/cvpr52688.2022.01166 2022

  24. [24]

    Y. Chen, J. Liu, X. Zhang, X. Qi, J. Jia, LargeKernel3D: Scaling up Kernels in 3D 15 Sparse CNNs, in: 2023 IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR), IEEE, Vancou- ver, BC, Canada, 2023, pp. 13488–13498. doi:10.1109/CVPR52729.2023.01296. URLhttps://ieeexplore.ieee.org/ document/10203060/

    work page doi:10.1109/cvpr52729.2023.01296 2023

  25. [25]

    T. Feng, W. Wang, F. Ma, Y. Yang, LSK3DNet: Towards Effective and Efficient 3D Perception with Large Sparse Kernels, in: 2024 IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition (CVPR), IEEE, Seattle, WA, USA, 2024, pp. 14916–14927. doi:10.1109/CVPR52733.2024.01413. URLhttps://ieeexplore.ieee.org/ document/10656196/

    work page doi:10.1109/cvpr52733.2024.01413 2024

  26. [26]

    B. Peng, X. Wu, L. Jiang, Y. Chen, H. Zhao, Z. Tian, J. Jia, OA-CNNs: Omni-Adaptive Sparse CNNs for 3D Semantic Segmentation, in: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, Seattle, WA, USA, 2024, pp. 21305– 21315.doi:10.1109/CVPR52733.2024.02013. URLhttps://ieeexplore.ieee.org/ document/10655421/

    work page doi:10.1109/cvpr52733.2024.02013 2024

  27. [27]

    C. R. Qi, L. Yi, H. Su, L. J. Guibas, Point- Net++: Deep Hierarchical Feature Learn- ing on Point Sets in a Metric Space, arXiv:1706.02413 [cs] (Jun. 2017).doi:10. 48550/arXiv.1706.02413. URLhttp://arxiv.org/abs/1706.02413

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  28. [28]

    Q. Hu, B. Yang, L. Xie, S. Rosa, Y. Guo, Z. Wang, N. Trigoni, A. Markham, RandLA- Net: Efficient Semantic Segmentation of Large-Scale Point Clouds, in: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, Seattle, WA, USA, 2020, pp. 11105–11114. doi:10.1109/CVPR42600.2020.01112. URLhttps://ieeexplore.ieee.org/ document/9156466/

    work page doi:10.1109/cvpr42600.2020.01112 2020

  29. [29]

    Large-scale Point Cloud Semantic Segmentation with Superpoint Graphs

    L. Landrieu, M. Simonovsky, Large-scale Point CloudSemanticSegmentationwithSuperpoint Graphs, arXiv:1711.09869 [cs] (Mar. 2018). doi:10.48550/arXiv.1711.09869. URLhttp://arxiv.org/abs/1711.09869

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1711.09869 2018

  30. [30]

    Y. Wang, Y. Sun, Z. Liu, S. E. Sarma, M. M. Bronstein, J. M. Solomon, Dynamic Graph CNN for Learning on Point Clouds, ACM Transactions on Graphics 38 (5) (2019) 1–12. doi:10.1145/3326362. URLhttps://dl.acm.org/doi/10.1145/ 3326362

    work page doi:10.1145/3326362 2019

  31. [31]

    H. Lei, N. Akhtar, A. Mian, SegGCN: Efficient 3D Point Cloud Segmentation With Fuzzy Spherical Kernel, in: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, Seattle, WA, USA, 2020, pp. 11608–11617. doi:10.1109/CVPR42600.2020.01163. URLhttps://ieeexplore.ieee.org/ document/9157177/

    work page doi:10.1109/cvpr42600.2020.01163 2020

  32. [32]

    Tatarchenko, J

    M. Tatarchenko, J. Park, V. Koltun, Q.-Y. Zhou, Tangent Convolutions for Dense Predic- tion in 3D, in: 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, IEEE, Salt Lake City, UT, USA, 2018, pp. 3887–3896.doi:10.1109/CVPR.2018.00409. URLhttps://ieeexplore.ieee.org/ document/8578507/

    work page doi:10.1109/cvpr.2018.00409 2018

  33. [33]

    W. Wu, Z. Qi, L. Fuxin, PointConv: Deep Convolutional Networks on 3D Point Clouds, arXiv:1811.07246 [cs] (Nov. 2020).doi:10. 48550/arXiv.1811.07246. URLhttp://arxiv.org/abs/1811.07246

    work page arXiv 2020

  34. [34]

    Kpconv: Flexible and deformable convolution for point clouds,

    H. Thomas, C. R. Qi, J.-E. Deschaud, B. Mar- cotegui, F. Goulette, L. J. Guibas, KPConv: Flexible and Deformable Convolution for Point Clouds, arXiv:1904.08889 [cs] (Aug. 2019). doi:10.48550/arXiv.1904.08889. URLhttp://arxiv.org/abs/1904.08889

    work page doi:10.48550/arxiv.1904.08889 1904

  35. [35]

    X. Li, Z. Zhang, Y. Li, M. Huang, J. Zhang, SFL-NET: Slight Filter Learn- ing Network for Point Cloud Semantic Segmentation, IEEE Transactions on Geo- science and Remote Sensing 61 (2023) 1–14. doi:10.1109/TGRS.2023.3313876. URLhttps://ieeexplore.ieee.org/ document/10250869/

    work page doi:10.1109/tgrs.2023.3313876 2023

  36. [36]

    H. Zhao, L. Jiang, J. Jia, P. Torr, V. Koltun, Point Transformer, in: 2021 IEEE/CVF International Conference on Computer Vision (ICCV), IEEE, Mon- treal, QC, Canada, 2021, pp. 16239–16248. doi:10.1109/ICCV48922.2021.01595. 16 URLhttps://ieeexplore.ieee.org/ document/9710703/

    work page doi:10.1109/iccv48922.2021.01595 2021

  37. [37]

    Guo, J.-X

    M.-H. Guo, J.-X. Cai, Z.-N. Liu, T.-J. Mu, R. R. Martin, S.-M. Hu, PCT: Point cloud transformer, Computational Visual Media 7 (2) (2021) 187–199. doi:10.1007/s41095-021-0229-5. URLhttps://ieeexplore.ieee.org/ document/10897555/

    work page doi:10.1007/s41095-021-0229-5 2021

  38. [38]

    X. Wu, Y. Lao, L. Jiang, X. Liu, H. Zhao, Point Transformer V2: Grouped Vector Attention and Partition-based Pooling, in: S. Koyejo, S. Mohamed, A. Agarwal, D. Bel- grave, K. Cho, A. Oh (Eds.), Advances in Neural Information Processing Systems, Vol. 35, Curran Associates, Inc., 2022, pp. 33330–33342. URLhttps://proceedings.neurips. cc/paper_files/paper/20...

    work page 2022

  39. [39]

    X. Wu, L. Jiang, P.-S. Wang, Z. Liu, X. Liu, Y. Qiao, W. Ouyang, T. He, H. Zhao, Point Transformer V3: Simpler, Faster, Stronger, in: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, Seattle, WA, USA, 2024, pp. 4840– 4851.doi:10.1109/CVPR52733.2024.00463. URLhttps://ieeexplore.ieee.org/ document/10658198/

    work page doi:10.1109/cvpr52733.2024.00463 2024

  40. [40]

    Y. Yue, D. Robert, J. Wang, S. Hong, J. D. Wegner, C. Rupprecht, K. Schindler, LitePT: Lighter Yet Stronger Point Trans- former, arXiv:2512.13689 (Mar. 2026).doi: 10.48550/arXiv.2512.13689. URLhttp://arxiv.org/abs/2512.13689

    work page doi:10.48550/arxiv.2512.13689 2026

  41. [41]

    Z. Liu, H. Hu, Y. Lin, Z. Yao, Z. Xie, Y. Wei, J. Ning, Y. Cao, Z. Zhang, L. Dong, F. Wei, B. Guo, Swin Transformer V2: Scal- ing Up Capacity and Resolution, in: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, New Orleans, LA, USA, 2022, pp. 11999–12009. doi:10.1109/CVPR52688.2022.01170. URLhttps://ieeexplore.ieee.org/ do...

    work page doi:10.1109/cvpr52688.2022.01170 2022

  42. [42]

    Yang, Y.-X

    Y.-Q. Yang, Y.-X. Guo, J.-Y. Xiong, Y. Liu, H. Pan, P.-S. Wang, X. Tong, B. Guo, Swin3D: A Pretrained Transformer Backbone for 3D Indoor Scene Understanding, Compu- tational Visual Media 11 (1) (2025) 83–101. doi:10.26599/CVM.2025.9450383. URLhttps://ieeexplore.ieee.org/ document/10901941/

    work page doi:10.26599/cvm.2025.9450383 2025

  43. [43]

    Kellner, B

    M. Kellner, B. Stahl, A. Reiterer, Fused Projection-Based Point Cloud Segmen- tation, Sensors 22 (3) (2022) 1139. doi:10.3390/s22031139. URLhttps://www.mdpi.com/1424-8220/ 22/3/1139

    work page doi:10.3390/s22031139 2022

  44. [44]

    Limit results for distribu ted estimation of invariant subspaces in multiple networks inference and pca

    Y.Hou, X.Zhu, Y.Ma, C.C.Loy, Y.Li, Point- to-Voxel Knowledge Distillation for LiDAR Semantic Segmentation, arXiv:2206.02099 [cs] (Jun. 2022).doi:10.48550/arXiv.2206. 02099. URLhttp://arxiv.org/abs/2206.02099

    work page doi:10.48550/arxiv.2206 2022

  45. [45]

    J. Xu, R. Zhang, J. Dou, Y. Zhu, J. Sun, S. Pu, RPVNet: A Deep and Efficient Range-Point-Voxel Fusion Network for Li- DAR Point Cloud Segmentation, in: 2021 IEEE/CVF International Conference on Computer Vision (ICCV), IEEE, Mon- treal, QC, Canada, 2021, pp. 16004–16013. doi:10.1109/ICCV48922.2021.01572. URLhttps://ieeexplore.ieee.org/ document/9709941/

    work page doi:10.1109/iccv48922.2021.01572 2021

  46. [46]

    Walk in the cloud: Learning curves for point clouds shape analysis, pp

    M. Caron, H. Touvron, I. Misra, H. Jegou, J. Mairal, P. Bojanowski, A. Joulin, Emerging Properties in Self-Supervised Vision Trans- formers, in: 2021 IEEE/CVF International Conference on Computer Vision (ICCV), IEEE, Montreal, QC, Canada, 2021, pp. 9630– 9640.doi:10.1109/ICCV48922.2021.00951. URLhttps://ieeexplore.ieee.org/ document/9709990/

    work page doi:10.1109/iccv48922.2021.00951 2021

  47. [47]

    K. He, X. Chen, S. Xie, Y. Li, P. Dollar, R. Girshick, Masked Autoencoders Are Scal- able Vision Learners, in: 2022 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), IEEE, New Orleans, LA, USA, 2022, pp. 15979–15988. doi:10.1109/CVPR52688.2022.01553. URLhttps://ieeexplore.ieee.org/ document/9879206/

    work page doi:10.1109/cvpr52688.2022.01553 2022

  48. [48]

    DINOv2: Learning Robust Visual Features without Supervision

    M. Oquab, T. Darcet, T. Moutakanni, H. Vo, M. Szafraniec, V. Khalidov, P. Fernandez, 17 D. Haziza, F. Massa, A. El-Nouby, M. Assran, N. Ballas, W. Galuba, R. Howes, P.-Y. Huang, S.-W. Li, I. Misra, M. Rabbat, V. Sharma, G. Synnaeve, H. Xu, H. Jegou, J. Mairal, P. Labatut, A. Joulin, P. Bojanowski, DI- NOv2: Learning Robust Visual Features with- out Superv...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2304.07193 2024

  49. [49]

    DINOv3

    O. Siméoni, H. V. Vo, M. Seitzer, F. Bal- dassarre, M. Oquab, C. Jose, V. Khali- dov, M. Szafraniec, S. Yi, M. Ramamon- jisoa, F. Massa, D. Haziza, L. Wehrstedt, J. Wang, T. Darcet, T. Moutakanni, L. Sen- tana, C. Roberts, A. Vedaldi, J. Tolan, J. Brandt, C. Couprie, J. Mairal, H. Jé- gou, P. Labatut, P. Bojanowski, DINOv3, arXiv:2508.10104 [cs] (Aug. 202...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  50. [50]

    S. Xie, J. Gu, D. Guo, C. R. Qi, L. Guibas, O. Litany, PointContrast: Unsupervised Pre- training for 3D Point Cloud Understand- ing, in: A. Vedaldi, H. Bischof, T. Brox, J.-M. Frahm (Eds.), Computer Vision – ECCV 2020, Springer International Publish- ing, Cham, 2020, pp. 574–591

    work page 2020

  51. [51]

    Y. Pang, W. Wang, F. E. H. Tay, W. Liu, Y. Tian, L. Yuan, Masked Autoencoders for Point Cloud Self-supervised Learning, in: S. Avidan, G. Brostow, M. Cissé, G. M. Farinella, T. Hassner (Eds.), Computer Vision – ECCV 2022, Vol. 13662, Springer Nature Switzerland, Cham, 2022, pp. 604–621, series Title: Lecture Notes in Computer Science. doi:10.1007/978-3-03...

    work page doi:10.1007/978-3-031-20086-1_35 2022

  52. [52]

    X. Wu, X. Wen, X. Liu, H. Zhao, Masked Scene Contrast: A Scalable Framework for Unsupervised 3D Representation Learning, in: 2023 IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition (CVPR), IEEE, Vancouver, BC, Canada, 2023, pp. 9415–9424. doi:10.1109/CVPR52729.2023.00908. URLhttps://ieeexplore.ieee.org/ document/10203752/

    work page doi:10.1109/cvpr52729.2023.00908 2023

  53. [53]

    X. Wu, D. DeTone, D. Frost, T. Shen, C. Xie, N. Yang, J. Engel, R. Newcombe, H. Zhao, J. Straub, Sonata: Self-Supervised Learning of Reliable Point Representations, arXiv:2503.16429 [cs] (Mar. 2025).doi:10. 48550/arXiv.2503.16429. URLhttp://arxiv.org/abs/2503.16429

    work page arXiv 2025

  54. [54]

    Zhang, X

    Y. Zhang, X. Wu, Y. Lao, C. Wang, Z. Tian, N. Wang, H. Zhao, Concerto: Joint 2D- 3D Self-Supervised Learning Emerges Spatial Representations, arXiv:2510.23607 [cs] (Oct. 2025).doi:10.48550/arXiv.2510.23607. URLhttp://arxiv.org/abs/2510.23607

    work page doi:10.48550/arxiv.2510.23607 2025

  55. [55]

    H. Zhu, H. Yang, X. Wu, D. Huang, S. Zhang, X. He, H. Zhao, C. Shen, Y. Qiao, T. He, W. Ouyang, PonderV2: Pave the Way for 3D Foundation Model with A Universal Pre- trainingParadigm, arXiv:2310.08586[cs](Apr. 2025).doi:10.48550/arXiv.2310.08586. URLhttp://arxiv.org/abs/2310.08586

    work page doi:10.48550/arxiv.2310.08586 2025

  56. [56]

    Behley, M

    J. Behley, M. Garbade, A. Milioto, J. Quen- zel, S. Behnke, C. Stachniss, J. Gall, SemanticKITTI: A Dataset for Semantic Scene Understanding of LiDAR Sequences, arXiv:1904.01416 [cs] (Aug. 2019).doi:10. 48550/arXiv.1904.01416. URLhttp://arxiv.org/abs/1904.01416

    work page arXiv 1904

  57. [57]

    P. Sun, H. Kretzschmar, X. Dotiwalla, A. Chouard, V. Patnaik, P. Tsui, J. Guo, Y. Zhou, Y. Chai, B. Caine, V. Vasudevan, W. Han, J. Ngiam, H. Zhao, A. Timofeev, S. Ettinger, M. Krivokon, A. Gao, A. Joshi, S. Zhao, S. Cheng, Y. Zhang, J. Shlens, Z. Chen, D. Anguelov, Scalability in Percep- tion for Autonomous Driving: Waymo Open Dataset, arXiv:1912.04838 [...

    work page doi:10.48550/arxiv.1912.04838 1912

  58. [58]

    nuScenes: A multimodal dataset for autonomous driving

    H. Caesar, V. Bankiti, A. H. Lang, S. Vora, V. E. Liong, Q. Xu, A. Krishnan, Y. Pan, G. Baldan, O. Beijbom, nuScenes: A mul- timodal dataset for autonomous driving, arXiv:1903.11027 [cs] (May 2020).doi:10. 48550/arXiv.1903.11027. URLhttp://arxiv.org/abs/1903.11027

    work page internal anchor Pith review Pith/arXiv arXiv 1903

  59. [59]

    Roynard, J.-E

    X. Roynard, J.-E. Deschaud, F. Goulette, Paris-Lille-3D: A large and high-quality ground-truth urban point cloud dataset for automatic segmentation and classification, 18 The International Journal of Robotics Re- search 37 (6) (2018) 545–557.doi:10.1177/ 0278364918767506. URLhttps://journals.sagepub.com/doi/ 10.1177/0278364918767506

    work page doi:10.1177/0278364918767506 2018

  60. [60]

    W. Tan, N. Qin, L. Ma, Y. Li, J. Du, G. Cai, K. Yang, J. Li, Toronto-3D: A Large-scale Mo- bile LiDAR Dataset for Semantic Segmenta- tion of Urban Roadways, in: 2020 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition Workshops (CVPRW), IEEE, Seattle, WA, USA, 2020, pp. 797–806. doi:10.1109/CVPRW50498.2020.00109. URLhttps://ieeexplore.ieee.or...

    work page doi:10.1109/cvprw50498.2020.00109 2020

  61. [62]

    Thomas, F

    H. Thomas, F. Goulette, J.-E. Deschaud, B. Marcotegui, Y. LeGall, Semantic Clas- sification of 3D Point Clouds with Multi- scale Spherical Neighborhoods, in: 2018 International Conference on 3D Vision (3DV), IEEE, Verona, 2018, pp. 390–398. doi:10.1109/3DV.2018.00052. URLhttps://ieeexplore.ieee.org/ document/8490990/

    work page doi:10.1109/3dv.2018.00052 2018

  62. [63]

    Varney, V

    N. Varney, V. K. Asari, Q. Graehling, Pyra- mid Point: A Multi-Level Focusing Network for Revisiting Feature Layers (2020).doi: 10.48550/ARXIV.2011.08692. URLhttps://arxiv.org/abs/2011.08692

    work page doi:10.48550/arxiv.2011.08692 2020

  63. [64]

    S. Yoo, Y. Jeong, M. Jameela, G. Sohn, Human Vision Based 3D Point Cloud Se- mantic Segmentation of Large-Scale Outdoor Scenes, in: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), IEEE, Vancou- ver, BC, Canada, 2023, pp. 6577–6586. doi:10.1109/CVPRW59228.2023.00699. URLhttps://ieeexplore.ieee.org/ document/10208664/

    work page doi:10.1109/cvprw59228.2023.00699 2023

  64. [65]

    S. Yoo, Y. Jeong, M. M. Sheikholeslami, G. Sohn, EyeNet++: A Multiscale and Multidensity Approach for Outdoor 3-D Semantic Segmentation Inspired by the Hu- man Visual Field, IEEE Transactions on Geoscience and Remote Sensing 63 (2025) 1–19.doi:10.1109/TGRS.2025.3589287. URLhttps://ieeexplore.ieee.org/ document/11080501/

    work page doi:10.1109/tgrs.2025.3589287 2025

  65. [66]

    Contributors, Spconv: Spatially sparse convolution library,https://github.com/ traveller59/spconv(2022)

    S. Contributors, Spconv: Spatially sparse convolution library,https://github.com/ traveller59/spconv(2022)

    work page 2022

  66. [67]

    M. Fey, J. E. Lenssen, Fast Graph Repre- sentation Learning with PyTorch Geometric, arXiv:1903.02428 [cs] (Apr. 2019).doi:10. 48550/arXiv.1903.02428. URLhttp://arxiv.org/abs/1903.02428

    work page internal anchor Pith review Pith/arXiv arXiv 1903

  67. [68]

    T. Dao, D. Y. Fu, S. Ermon, A. Rudra, C. Ré, FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness, version Number: 2 (2022).doi:10.48550/ARXIV. 2205.14135. URLhttps://arxiv.org/abs/2205.14135

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv 2022

  68. [69]

    FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning

    T. Dao, FlashAttention-2: Faster Attention with Better Parallelism and Work Partition- ing, version Number: 1 (2023).doi:10. 48550/ARXIV.2307.08691. URLhttps://arxiv.org/abs/2307.08691

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  69. [70]

    Ioffe, C

    S. Ioffe, C. Szegedy, Batch Normalization: Accelerating Deep Network Training by Re- ducing Internal Covariate Shift, in: F. Bach, D. Blei (Eds.), Proceedings of the 32nd In- ternational Conference on Machine Learning, Vol. 37 of Proceedings of Machine Learning Research, PMLR, Lille, France, 2015, pp. 448–456. URLhttps://proceedings.mlr.press/ v37/ioffe15.html

    work page 2015

  70. [71]

    J. L. Ba, J. R. Kiros, G. E. Hinton, Layer Nor- malization, version Number: 1 (2016).doi: 10.48550/ARXIV.1607.06450. URLhttps://arxiv.org/abs/1607.06450

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1607.06450 2016

  71. [72]

    C. R. Harris, K. J. Millman, S. J. Van Der Walt, R. Gommers, P. Virtanen, D. Cour- napeau, E. Wieser, J. Taylor, S. Berg, N. J. Smith, R. Kern, M. Picus, S. Hoyer, M. H. Van Kerkwijk, M. Brett, A. Hal- dane, J. F. Del Río, M. Wiebe, P. Peterson, P. Gérard-Marchant, K. Sheppard, T. Reddy, 19 W. Weckesser, H. Abbasi, C. Gohlke, T. E. Oliphant, Array program...

    work page doi:10.1038/s41586-020-2649-2 2020

  72. [73]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Kopf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, S. Chintala, PyTorch: An Imperative Style, High-Performance Deep Learning Library, in: H. Wallach, H. Larochelle, A. Beygelzimer, F. d....

    work page 2019

  73. [74]

    S. K. Lam, A. Pitrou, S. Seibert, Numba: a LLVM-based Python JIT compiler, in: Proceedings of the Second Workshop on the LLVM Compiler Infrastructure in HPC, ACM, Austin Texas, 2015, pp. 1– 6.doi:10.1145/2833157.2833162. URLhttps://dl.acm.org/doi/10.1145/ 2833157.2833162

    work page doi:10.1145/2833157.2833162 2015

  74. [75]

    Decoupled Weight Decay Regularization

    I. Loshchilov, F. Hutter, Decoupled Weight Decay Regularization, version Number: 3 (2017).doi:10.48550/ARXIV.1711.05101. URLhttps://arxiv.org/abs/1711.05101

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1711.05101 2017

  75. [76]

    L. N. Smith, N. Topin, Super-Convergence: Very Fast Training of Neural Networks Us- ing Large Learning Rates, version Number: 3 (2017).doi:10.48550/ARXIV.1708.07120. URLhttps://arxiv.org/abs/1708.07120

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1708.07120 2017

  76. [77]

    Rethinking the Inception Architecture for Computer Vision

    C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, Z. Wojna, Rethinking the Inception Architec- ture for Computer Vision, version Number: 3 (2015).doi:10.48550/ARXIV.1512.00567. URLhttps://arxiv.org/abs/1512.00567

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1512.00567 2015

  77. [78]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, N. Houlsby, An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale (2020).doi:10.48550/ ARXIV.2010.11929. URLhttps://arxiv.org/abs/2010.11929

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  78. [79]

    X. Jiao, C. Lv, J. Zhao, R. Yi, Y.-H. Wen, Z. Pan, Z. Wu, Y.-J. Liu, Weighted Poisson- disk Resampling on Large-Scale Point Clouds, Proceedings of the AAAI Conference on Artificial Intelligence 39 (4) (2025) 4084–4092. doi:10.1609/aaai.v39i4.32428. URLhttps://ojs.aaai.org/index.php/ AAAI/article/view/32428

    work page doi:10.1609/aaai.v39i4.32428 2025

  79. [80]

    Kellner, A

    M. Kellner, A. Schmitt, A. Reiterer, Automatic Generation of 3D Bridge Models from 3D Point Clouds, Re- sults in Engineering (2026) 109532doi: 10.1016/j.rineng.2026.109532. URLhttps://linkinghub.elsevier.com/ retrieve/pii/S2590123026005724 20

    work page doi:10.1016/j.rineng.2026.109532 2026