pith. sign in

arxiv: 2605.15475 · v1 · pith:M2JUNQY3new · submitted 2026-05-14 · 💻 cs.CV · cs.MM

A Unified Non-Parametric and Interpretable Point Cloud Analysis via t-FCW Graph Representation

Haijian Lai , Bowen Liu , Man Xu , Chan-Tong Lam , Jo\~ao Macedo , Benjamin Ng , Sio-Kei Im This is my paper

Pith reviewed 2026-05-19 14:33 UTC · model grok-4.3

classification 💻 cs.CV cs.MM
keywords point cloudgraph representationnon-parametricinterpretabilityclassificationsegmentationModelNet40feature extraction
0
0 comments X

The pith

An empowered t-FCW graph representation embeds point clouds non-parametrically into a metric space while inheriting surface robustness and supplying dimension-wise interpretability.

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

This paper introduces an empowered version of the transposed Fully Connected Weighted graph to represent point clouds. The authors examine the properties behind its effectiveness and construct a network that relies on this representation alone as its feature extractor. They further build memory banks from the same representation to handle classification, part segmentation, and semantic segmentation. The approach yields an efficient pipeline that completes ModelNet40 classification in roughly seven seconds on an NVIDIA RTX A5000 GPU and works either by itself or as an addition to other models.

Core claim

The central claim is that the empowered t-FCW graph representation embeds point clouds into a metric space, inherits robustness from surface descriptors, and supplies interpretability through dimension-wise relations. These properties support a network that uses the representation exclusively as a feature extractor together with memory banks for classification, part segmentation, and semantic segmentation, while delivering high computational efficiency.

What carries the argument

The empowered transposed Fully Connected Weighted (t-FCW) graph representation, which embeds point clouds into a metric space and serves as a non-parametric feature extractor.

If this is right

  • Classification, part segmentation, and semantic segmentation become feasible using only t-FCW features stored in memory banks.
  • ModelNet40 classification completes in approximately 7 seconds on an NVIDIA RTX A5000 GPU.
  • The same representation functions as a lightweight standalone baseline.
  • It also serves as a plug-in that complements existing deep point-cloud models.

Where Pith is reading between the lines

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

  • The non-parametric nature could reduce the need for large training sets or GPU memory in 3D vision pipelines.
  • Dimension-wise relations might allow direct inspection of which geometric features drive a given prediction.
  • The same representation could be tested on other 3D datasets to check whether the reported robustness generalizes.

Load-bearing premise

The properties of the empowered t-FCW graph representation are sufficient to support accurate classification, part segmentation, and semantic segmentation when used exclusively as feature extractors without additional learned components or post-processing.

What would settle it

Running the t-FCW memory-bank pipeline on the ModelNet40 test set and checking whether classification accuracy remains competitive with standard deep models; a large drop would falsify the sufficiency claim.

Figures

Figures reproduced from arXiv: 2605.15475 by Benjamin Ng, Bowen Liu, Chan-Tong Lam, Haijian Lai, Jo\~ao Macedo, Man Xu, Sio-Kei Im.

Figure 1
Figure 1. Figure 1: Feature extraction of the t-FCW networks. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Structures of united and empowered t-FCW blocks [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Wings and engines of airplanes with their dimension distributions and t-FCW representations, where the cosine similarity [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Highlight the features of t-FCW (left) and empowered t [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Stability analysis across batch sizes and data shuffling. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: , we recommend selecting K-nearest neighbors across different datasets by highlighting star marks. 30 40 50 60 70 80 90 100 110 120 K Neighbors 0.0 0.2 0.4 0.6 0.8 1.0 Normalized Performance (0-1) K=60 K=70 K=90 K=100 ModelNet40 ScanObjectNN S3DIS ShapeNet-Part [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Max processing points’ volume of various methods [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
read the original abstract

We introduce an empowered transposed Fully Connected Weighted (t-FCW) graph representation to embed point clouds into a metric space. While original t-FCW has shown promising results for point cloud classification, the reasons behind its effectiveness and its broader applicability remained unclear. In this work, we analyze the properties that make the empowered and original t-FCW effective and design a network that uses the empowered t-FCW exclusively as feature extractors. From an interpretability perspective, we build memory banks for classification, part segmentation, and semantic segmentation using the empowered t-FCW. Our analysis reveals that the empowered t-FCW inherits robustness from surface descriptors, provides interpretability through dimension-wise relations. These properties enable a highly efficient and interpretable network, which processes the ModelNet40 classification problem in approximately 7 seconds on an NVIDIA RTX A5000 GPU. Importantly, empowered t-FCW can function both as a lightweight standalone baseline and as a complementary plug-in to existing deep models.

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 manuscript introduces an empowered transposed Fully Connected Weighted (t-FCW) graph representation for embedding point clouds into a metric space. It analyzes the properties (robustness inherited from surface descriptors and interpretability via dimension-wise relations) that make the representation effective, designs a network that uses the empowered t-FCW exclusively as feature extractors, and constructs memory banks to support classification, part segmentation, and semantic segmentation. The work emphasizes efficiency, reporting that the approach processes ModelNet40 classification in approximately 7 seconds on an NVIDIA RTX A5000 GPU, and positions the method as both a lightweight standalone baseline and a plug-in for existing deep models.

Significance. If the central claims are supported by rigorous quantitative evidence, the work would offer a notable contribution to point cloud analysis by providing a non-parametric, interpretable alternative that unifies multiple tasks through memory-bank lookup. The reported computational efficiency and potential for plug-in use would be valuable strengths, particularly if the dimension-wise relations yield competitive accuracy without learned refinement.

major comments (2)
  1. [Experiments] Experiments section (segmentation results): The claim that memory banks built from empowered t-FCW enable accurate part segmentation and semantic segmentation when used exclusively as feature extractors rests on the unshown sufficiency of nearest-neighbor or prototype matching. No ablation is reported that removes all learned components on ShapeNet or S3DIS while retaining only the t-FCW memory bank; the 7-second ModelNet40 timing is given only for classification.
  2. [Abstract] Abstract and introduction: Effectiveness, robustness, and interpretability are asserted, yet the manuscript supplies no quantitative results, derivations, ablation studies, or error analysis in the provided summary to ground these claims. This undermines the load-bearing assertion that the properties of the empowered t-FCW are sufficient for the reported tasks.
minor comments (1)
  1. [Abstract] The abstract would be strengthened by including at least one key quantitative result (e.g., accuracy or IoU) to support the effectiveness claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment below and outline the revisions we will make to strengthen the presentation of our results.

read point-by-point responses

  1. Referee: [Experiments] Experiments section (segmentation results): The claim that memory banks built from empowered t-FCW enable accurate part segmentation and semantic segmentation when used exclusively as feature extractors rests on the unshown sufficiency of nearest-neighbor or prototype matching. No ablation is reported that removes all learned components on ShapeNet or S3DIS while retaining only the t-FCW memory bank; the 7-second ModelNet40 timing is given only for classification.

    Authors: We agree that the current version does not include an explicit ablation isolating the t-FCW memory bank (with nearest-neighbor matching) for part segmentation on ShapeNet and semantic segmentation on S3DIS, nor does it report timing for those tasks. The manuscript positions empowered t-FCW as a non-parametric extractor, but to directly demonstrate sufficiency we will add the requested ablation studies and corresponding runtime measurements in the revised experiments section. revision: yes

  2. Referee: [Abstract] Abstract and introduction: Effectiveness, robustness, and interpretability are asserted, yet the manuscript supplies no quantitative results, derivations, ablation studies, or error analysis in the provided summary to ground these claims. This undermines the load-bearing assertion that the properties of the empowered t-FCW are sufficient for the reported tasks.

    Authors: The full manuscript provides quantitative classification accuracy on ModelNet40, analysis showing robustness inherited from surface descriptors, and interpretability through dimension-wise relations, along with supporting derivations. However, the abstract and introduction summarize these without embedding the key numbers or references. We will revise both sections to include concise quantitative anchors and explicit pointers to the supporting analyses. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper introduces the empowered t-FCW as a new graph representation for point clouds, analyzes its properties (robustness from surface descriptors, interpretability via dimension-wise relations), and constructs memory banks for downstream tasks. These steps are presented as independent design choices and empirical observations rather than reductions to prior fitted parameters or self-referential definitions. No equations or claims reduce by construction to inputs; the 7-second timing is a direct runtime measurement, not a prediction. Self-citations to original t-FCW are not load-bearing for uniqueness theorems or ansatzes. The derivation remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no mathematical details, so no free parameters, axioms, or invented entities can be identified.

pith-pipeline@v0.9.0 · 5720 in / 1147 out tokens · 66112 ms · 2026-05-19T14:33:47.636516+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

40 extracted references · 40 canonical work pages

  1. [1]

    A comprehensive overview of deep learning techniques for 3d point cloud classification and semantic segmentation,

    S. Sarker, P. Sarker, G. Stone, R. Gorman, A. Tavakkoli, G. Bebis, and J. Sattarvand, “A comprehensive overview of deep learning techniques for 3d point cloud classification and semantic segmentation,”Machine Vision and Applications, vol. 35, no. 4, p. 67, 2024

    work page 2024

  2. [2]

    Deep learning-based 3d point cloud classification: A systematic survey and outlook,

    H. Zhang, C. Wang, S. Tian, B. Lu, L. Zhang, X. Ning, and X. Bai, “Deep learning-based 3d point cloud classification: A systematic survey and outlook,”Displays, vol. 79, p. 102456, 2023

    work page 2023

  3. [3]

    Advancements in point cloud-based 3d defect classification and segmentation for industrial systems: A comprehensive survey,

    A. Rani, D. Ortiz-Arroyo, and P. Durdevic, “Advancements in point cloud-based 3d defect classification and segmentation for industrial systems: A comprehensive survey,”Information Fusion, p. 102575, 2024

    work page 2024

  4. [4]

    Fast graph representation learning with PyTorch Geometric,

    M. Fey and J. E. Lenssen, “Fast graph representation learning with PyTorch Geometric,” inICLR Workshop on Representation Learning on Graphs and Manifolds, 2019

    work page 2019

  5. [5]

    Torch-points3d: A modular multi-task framework for reproducible deep learning on 3d point clouds,

    T. Chaton, N. Chaulet, S. Horache, and L. Landrieu, “Torch-points3d: A modular multi-task framework for reproducible deep learning on 3d point clouds,”2020 International Conference on 3D Vision (3DV), pp. 1– 10, 2020. [Online]. Available: https://api.semanticscholar.org/CorpusID: 222272273

    work page 2020

  6. [6]

    Pointnet++ pytorch,

    E. Wijmans, “Pointnet++ pytorch,”https://github.com/erikwijmans/Poin tnet2 PyTorch, 2018

    work page 2018

  7. [7]

    Pointnext: Revisiting pointnet++ with improved training and scaling strategies,

    G. Qian, Y . Li, H. Peng, J. Mai, H. Hammoud, M. Elhoseiny, and B. Ghanem, “Pointnext: Revisiting pointnet++ with improved training and scaling strategies,”Advances in neural information processing systems, vol. 35, pp. 23 192–23 204, 2022

    work page 2022

  8. [8]

    Starting from non-parametric networks for 3d point cloud analysis,

    R. Zhang, L. Wang, Y . Wang, P. Gao, H. Li, and J. Shi, “Starting from non-parametric networks for 3d point cloud analysis,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 5344–5353

    work page 2023

  9. [9]

    No time to train: Empowering non-parametric networks for few-shot 3d scene segmentation,

    X. Zhu, R. Zhang, B. He, Z. Guo, J. Liu, H. Xiao, C. Fu, H. Dong, and P. Gao, “No time to train: Empowering non-parametric networks for few-shot 3d scene segmentation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 3838–3847

    work page 2024

  10. [10]

    Point-fcw: Transposed- fcw graph representation for point cloud classification using tda,

    H. Lai, B. Liu, C.-T. Lam, B. Ng, and S.-K. Im, “Point-fcw: Transposed- fcw graph representation for point cloud classification using tda,”IEEE Signal Processing Letters, 2025

    work page 2025

  11. [11]

    Pointnet: Deep learning on point sets for 3d classification and segmentation,

    C. R. Qi, H. Su, K. Mo, and L. J. Guibas, “Pointnet: Deep learning on point sets for 3d classification and segmentation,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 652–660

    work page 2017

  12. [12]

    Pointnet++: Deep hierarchical feature learning on point sets in a metric space,

    C. R. Qi, L. Yi, H. Su, and L. J. Guibas, “Pointnet++: Deep hierarchical feature learning on point sets in a metric space,” inAdvances in Neural Information Processing Systems, 2017, pp. 5099–5108

    work page 2017

  13. [13]

    Rethinking network design and local geometry in point cloud: A simple residual MLP framework,

    X. Ma, C. Qin, H. You, H. Ran, and Y . Fu, “Rethinking network design and local geometry in point cloud: A simple residual MLP framework,” inInternational Conference on Learning Representations, 2022. [Online]. Available: https://openreview.net/forum?id=3Pbra- u76D

    work page 2022

  14. [14]

    Geometric back-projection net- work for point cloud classification,

    S. Qiu, S. Anwar, and N. Barnes, “Geometric back-projection net- work for point cloud classification,”IEEE Transactions on Multimedia, vol. 24, pp. 1943–1955, 2022

    work page 1943

  15. [15]

    Surface representation for point clouds,

    H. Ran, J. Liu, and C. Wang, “Surface representation for point clouds,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 18 942–18 952

    work page 2022

  16. [16]

    Risurconv: Rotation invariant sur- face attention-augmented convolutions for 3d point cloud classification and segmentation,

    Z. Zhang, L. Yang, and Z. Xiang, “Risurconv: Rotation invariant sur- face attention-augmented convolutions for 3d point cloud classification and segmentation,” in2024 European Conference on Computer Vision (ECCV), 2024, pp. 1–14

    work page 2024

  17. [17]

    V oxnet: A 3d convolutional neural net- work for real-time object recognition,

    D. Maturana and S. Scherer, “V oxnet: A 3d convolutional neural net- work for real-time object recognition,” in2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2015, pp. 922– 928. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 13

    work page 2015

  18. [18]

    Dynamic graph cnn for learning on point clouds,

    Y . Wang, Y . Sun, Z. Liu, S. E. Sarma, M. M. Bronstein, and J. M. Solomon, “Dynamic graph cnn for learning on point clouds,”ACM Transactions on Graphics (tog), vol. 38, no. 5, pp. 1–12, 2019

    work page 2019

  19. [19]

    Multi-view convolutional neural networks for 3d shape recognition,

    H. Su, S. Maji, E. Kalogerakis, and E. Learned-Miller, “Multi-view convolutional neural networks for 3d shape recognition,” inProceedings of the IEEE International Conference on Computer Vision (ICCV), December 2015

    work page 2015

  20. [20]

    Nonparametric variational auto-encoders for hierarchical representation learning,

    P. Goyal, Z. Hu, X. Liang, C. Wang, and E. P. Xing, “Nonparametric variational auto-encoders for hierarchical representation learning,” in Proceedings of the IEEE International Conference on Computer Vision, 2017, pp. 5094–5102

    work page 2017

  21. [21]

    Global optimization of lipschitz func- tions,

    C. Malherbe and N. Vayatis, “Global optimization of lipschitz func- tions,” inInternational Conference on Machine Learning. PMLR, 2017, pp. 2314–2323

    work page 2017

  22. [22]

    Fourier features let networks learn high frequency functions in low dimensional domains,

    M. Tancik, P. Srinivasan, B. Mildenhall, S. Fridovich-Keil, N. Raghavan, U. Singhal, R. Ramamoorthi, J. Barron, and R. Ng, “Fourier features let networks learn high frequency functions in low dimensional domains,” Advances in neural information processing systems, vol. 33, pp. 7537– 7547, 2020

    work page 2020

  23. [23]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” inProceedings of the 31st International Conference on Neural Information Processing Systems, ser. NIPS’17. Red Hook, NY , USA: Curran Associates Inc., 2017, p. 6000–6010

    work page 2017

  24. [24]

    3d shape classification by registration: Neural-network-free and training-free,

    C. Gou, Y . Mou, W. Li, N. Purohit, S. Yadav, H. Bai, X. Zhang, and L. Chen, “3d shape classification by registration: Neural-network-free and training-free,” inICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2025, pp. 1–5

    work page 2025

  25. [25]

    Non-parametric represen- tation learning with kernels,

    P. Esser, M. Fleissner, and D. Ghoshdastidar, “Non-parametric represen- tation learning with kernels,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 11, 2024, pp. 11 910–11 918

    work page 2024

  26. [26]

    Interpretable3d: An ad-hoc interpretable classifier for 3d point clouds,

    T. Feng, R. Quan, X. Wang, W. Wang, and Y . Yang, “Interpretable3d: An ad-hoc interpretable classifier for 3d point clouds,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 2, 2024, pp. 1761–1769

    work page 2024

  27. [27]

    giotto-tda:: A topological data analysis toolkit for machine learning and data exploration,

    G. Tauzin, U. Lupo, L. Tunstall, J. B. P ´erez, M. Caorsi, A. M. Medina- Mardones, A. Dassatti, and K. Hess, “giotto-tda:: A topological data analysis toolkit for machine learning and data exploration,”Journal of Machine Learning Research, vol. 22, no. 39, pp. 1–6, 2021

    work page 2021

  28. [28]

    Tip-adapter: Training-free adaption of clip for few-shot classification,

    R. Zhang, W. Zhang, R. Fang, P. Gao, K. Li, J. Dai, Y . Qiao, and H. Li, “Tip-adapter: Training-free adaption of clip for few-shot classification,” inComputer Vision – ECCV 2022, S. Avidan, G. Brostow, M. Ciss ´e, G. M. Farinella, and T. Hassner, Eds. Cham: Springer Nature Switzerland, 2022, pp. 493–510

    work page 2022

  29. [29]

    Neural style transfer: A review,

    Y . Jing, Y . Yang, Z. Feng, J. Ye, Y . Yu, and M. Song, “Neural style transfer: A review,”IEEE transactions on visualization and computer graphics, vol. 26, no. 11, pp. 3365–3385, 2019

    work page 2019

  30. [30]

    Learning what not to segment: A new perspective on few-shot segmentation,

    C. Lang, G. Cheng, B. Tu, and J. Han, “Learning what not to segment: A new perspective on few-shot segmentation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 8057–8067

    work page 2022

  31. [31]

    3d shapenets: A deep representation for volumetric shapes,

    Z. Wu, S. Song, A. Khosla, F. Yu, L. Zhang, X. Tang, and J. Xiao, “3d shapenets: A deep representation for volumetric shapes,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2015, pp. 1912–1920

    work page 2015

  32. [32]

    Re- visiting point cloud classification: A new benchmark dataset and clas- sification model on real-world data,

    M. A. Uy, Q.-H. Pham, B.-S. Hua, T. Nguyen, and S.-K. Yeung, “Re- visiting point cloud classification: A new benchmark dataset and clas- sification model on real-world data,” inProceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 1588–1597

    work page 2019

  33. [33]

    A scalable active framework for region annotation in 3d shape collections,

    L. Yi, V . G. Kim, D. Ceylan, I.-C. Shen, M. Yan, H. Su, C. Lu, Q. Huang, A. Sheffer, and L. Guibas, “A scalable active framework for region annotation in 3d shape collections,”ACM Transactions on Graphics (ToG), vol. 35, no. 6, pp. 1–12, 2016

    work page 2016

  34. [34]

    3d semantic parsing of large-scale indoor spaces,

    I. Armeni, O. Sener, A. R. Zamir, H. Jiang, I. Brilakis, M. Fischer, and S. Savarese, “3d semantic parsing of large-scale indoor spaces,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 1534–1543

    work page 2016

  35. [35]

    Few-shot 3d point cloud semantic segmentation,

    N. Zhao, T.-S. Chua, and G. H. Lee, “Few-shot 3d point cloud semantic segmentation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 8873–8882

    work page 2021

  36. [36]

    Point-gn: A non-parametric network using gaussian positional encoding for point cloud classification,

    M. Mohammadi and A. Salarpour, “Point-gn: A non-parametric network using gaussian positional encoding for point cloud classification,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2025, pp. 3487–3496

    work page 2025

  37. [37]

    Few-shot learning with graph neural networks,

    V . G. Satorras and J. B. Estrach, “Few-shot learning with graph neural networks,” inInternational conference on learning representations, 2018

    work page 2018

  38. [38]

    Prototype adaption and projection for few-and zero-shot 3d point cloud semantic segmentation,

    S. He, X. Jiang, W. Jiang, and H. Ding, “Prototype adaption and projection for few-and zero-shot 3d point cloud semantic segmentation,” IEEE Transactions on Image Processing, vol. 32, pp. 3199–3211, 2023

    work page 2023

  39. [39]

    Generating 3d adversarial point clouds,

    C. Xiang, C. R. Qi, and B. Li, “Generating 3d adversarial point clouds,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 9136–9144

    work page 2019

  40. [40]

    Benchmarking and analyzing point cloud classification under corruptions,

    J. Ren, L. Pan, and Z. Liu, “Benchmarking and analyzing point cloud classification under corruptions,” inInternational Conference on Ma- chine Learning (ICML), 2022

    work page 2022