Pith. sign in

REVIEW 3 major objections 4 minor 41 references

Enhancing 3D Point Cloud Classification with ModelNet-R and Point-SkipNet

T0 review · 3 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Cleaning up ModelNet's labels, removing flat and mislabeled samples, and sharpening five confusable classes raises reported accuracy for every model tested, and the proposed lightweight Point-SkipNet reaches 94.33% overall accuracy with onl

desk verdict ModelNet-R is a plausible dataset-cleaning starting point and Point-SkipNet is a fine lightweight baseline, but the paper's central performance claims are confounded by a changed test set and a circular refinement process. read the letter →

arxiv 2509.05198 v1 pith:NDWWDDW6 submitted 2025-09-05 cs.CV cs.AIcs.LGcs.RO

Mohammad Saeid , Amir Salarpour , Pedram MohajerAnsari This is my paper · ORCID
classification cs.CVcs.AIcs.LGcs.RO
keywords 3DpointcloudclassificationModelNet-RdatasetrefinementbenchmarkcleaninglightweightgraphnetworkskipconnectionsNet40
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Point cloud classifiers are usually compared on ModelNet40, but this paper claims the benchmark itself is part of the problem: it contains mislabeled objects, flat 2D stand-ins, and classes that look alike. The authors construct ModelNet-R by relabeling and removing samples in five confusable classes, and report that every model they test gains accuracy when trained and evaluated on this cleaned version. They also introduce Point-SkipNet, a graph-based network with 1.47M parameters, and report it reaches 94.33% overall accuracy and 92.93% mean class accuracy on ModelNet-R. If the dataset-quality claim holds, future point cloud comparisons need a cleaner standard, and lightweight architectures may be closer to heavier models than their parameter counts suggest.

What carries the argument

The key machinery is twofold. On the data side, ModelNet-R is constructed by explicit class definitions—plant contains only plants, flower_pot includes plant plus pot, vase is an empty pot, cup has a handle, bowl is wide and shallow—and by guided relabeling and removal of samples in those five classes. On the model side, Point-SkipNet's Sample and Group Module applies farthest point sampling followed by ball-query neighborhood grouping; each neighborhood is passed through an MLP, max-pooled, and concatenated with original coordinates at every stage via skip connections, preserving spatial information while keeping the parameter count at 1.47M.

What would settle it

Train a model on ModelNet-R and test it on the original ModelNet test set (and vice versa); if the accuracy gains mostly vanish in this cross-evaluation, the reported improvement is an artifact of a changed evaluation set. A second check: randomly remove and relabel the same number of samples from the same five classes with no quality rationale; if accuracy rises as much as ModelNet-R's reported gains, the cleanup narrative is not the active ingredient.

Watch

Extended reading notes

Core claim

The central claim is that dataset quality, not just architecture, is a first-order factor in 3D point cloud classification. The authors refine five confusing classes of ModelNet40—flower_pot, plant, vase, cup, bowl—removing 19 samples, relabeling hundreds, and tightening class definitions. They report that eight models all improve on ModelNet-R compared with the original benchmark, with overall accuracy gains from +0.32 to +3.95 points. They then present Point-SkipNet, which uses farthest point sampling, ball-query grouping, MLP feature extraction, max pooling, and concatenated skip connections; with 1.47M parameters it reaches 94.33% overall accuracy and 92.93% mean class accuracy on ModelN

Load-bearing premise

The claim that better data quality causes the accuracy gains rests on comparing ModelNet-R against the original ModelNet, but the two sets differ in composition, labels, and class definitions, and the relabeling was guided by the very models being evaluated.

Editorial extensions

If this is right

  • On ModelNet-R, all eight reported models improve in overall accuracy, with gains from +0.32 to +3.95 points, so dataset refinement is presented as a transferable intervention rather than a quirk of one architecture.
  • Point-SkipNet's 1.47M parameters versus PointMLP's 12.60M parameters, at competitive accuracy on ModelNet-R, implies that high accuracy does not require transformer-scale models.
  • Concatenation skip connections and rotation-only augmentation are the reported best design choices for Point-SkipNet on ModelNet-R.
  • Models benchmarked on original ModelNet should be re-evaluated on ModelNet-R to reflect the corrected labels and removed samples.
  • Because only 5 of 40 classes were refined, the reported improvements are framed as a lower bound on what full refinement might yield.

Reading between the lines

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

  • The gains are measured by training and evaluating on a changed benchmark, so part of the improvement likely reflects a smaller and more self-consistent test set rather than a general property of cleaner data; a cross-evaluation on the original test split would separate the two effects.
  • The relabeling was informed by confusion-matrix analyses of the same models being evaluated, so ModelNet-R may encode those models' preferred boundaries; an independent annotation protocol or a held-out class would test whether the refinements generalize beyond the five modified classes.
  • The same refinement recipe could be applied to the remaining 35 ModelNet classes or to other CAD benchmarks; if gains saturate or fail to appear there, the mechanism behind the improvement would be clarified.
  • Point-SkipNet's skip connections and lightweight grouping are not specific to classification; the architecture could plausibly be adapted to segmentation or detection, but the paper provides no evidence for that.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper makes two contributions: a refined version of ModelNet40 called ModelNet-R, obtained by relabeling/removing samples in five classes and redefining class boundaries (Section III-A, III-B), and a lightweight graph-based point cloud classifier called Point-SkipNet that uses farthest-point sampling, ball queries, MLPs, max-pooling, and skip connections (Section III-C). The central claims are that models trained and evaluated on ModelNet-R achieve higher accuracy than on the original ModelNet (Section IV-E), and that Point-SkipNet reaches state-of-the-art accuracy on ModelNet-R with a much smaller parameter count. Experiments compare several baselines on ModelNet and ModelNet-R, and ablation studies examine augmentation and skip-connection modes (Section IV-F).

Significance. If the dataset refinement were evaluated in a controlled way, ModelNet-R could be a useful resource for the point-cloud community, as it directly addresses known label noise and class-overlap issues. The proposed Point-SkipNet architecture is simple, parameter-efficient (1.47M), and the paper provides enough architectural detail to reproduce it. However, the current evidence for the main claims is not reliable: the ModelNet-R evaluation set differs from ModelNet in sample composition and class definitions, and the refinement was guided by the very models being evaluated. The state-of-the-art claim is also contradicted by the paper's own table. These are load-bearing issues, not presentation problems.

major comments (3)
  1. [Section IV-E, Table IV] The central comparison is uncontrolled. ModelNet-R differs from ModelNet in three confounded ways: 19 samples were removed, many samples were relabeled (Table III), and class definitions were changed (Section III-A3). An accuracy increase on ModelNet-R cannot therefore be attributed to improved data quality; the test set itself is different. To support the claim that refinement improves generalization, the authors should report accuracy on the intersection of unchanged samples, or use the original ModelNet test set with only label corrections applied, or otherwise control for the changed test composition. Additionally, PointNet++ (SSG) has no original ModelNet values ('-'), so its reported +1.91% OA and +4.16% mAcc deltas are unverifiable.
  2. [Section III-B] The refinement process was guided by 'visual observations and confusion matrix analyzes' of classifiers. Since the models later evaluated on ModelNet-R are of the same type, the relabeling/removal decisions are partially tuned to the models' existing errors. Measuring accuracy on the resulting ModelNet-R therefore partly measures the effect of the authors' relabeling choices rather than a general property of data quality. An independent validation is needed: e.g., pre-specify refinement rules before examining model errors, or validate on a held-out subset of classes/samples that were not used to guide the refinement.
  3. [Abstract, Section IV-E, Table IV] The claim that 'Point-SkipNet achieves state-of-the-art accuracy on ModelNet-R' is contradicted by Table IV. PointMLP reaches 95.33% OA versus 94.33% for Point-SkipNet, and PointNet++ (MSG) reaches 94.06% with 1.74M parameters. Without a stated parameter budget or an explicit Pareto definition (e.g., 'best accuracy among models with fewer than 2M parameters'), the state-of-the-art claim is not falsifiable. The authors should either remove the claim or define and apply a precise efficiency--accuracy criterion.
minor comments (4)
  1. [Section IV-F, Tables V and VI] The ablation tables do not state which dataset (ModelNet or ModelNet-R) was used. The text also does not define the 'Main' augmentation mode or explain why rotation augmentation is compared against 'All augmentations'. Please clarify the experimental protocol and report variance across runs.
  2. [Table III] The table header is confusing: the row labeled 'Total' is a column sum, while the column labeled 'Total' is a row sum. Please rename these to 'New class count' and 'Original class count' and clarify in the caption that rows are original classes and columns are new classes.
  3. [Equations (4)-(5), Figure 6] Notation is inconsistent: Eq. (4) uses N_i and N_j with N_j < N_i, while Figure 6 uses N0, N1, N2 without connecting these to the equation variables. The ball query radius r and neighbor counts k1, k2 are listed in the figure but not defined in the text. Please align the notation.
  4. [Throughout] There are minor typos and formatting issues: the GitHub URL in the abstract reads 'ModeNetR' instead of 'ModelNetR'; reference [39] lacks publication details; and Figures 2, 6, and 8 present the same module with slightly different annotations, which is distracting.

Circularity Check

1 steps flagged · score 6.0 of 10

Dataset-quality improvement is partly circular: ModelNet-R labels were edited using classifier confusion, so the reported accuracy gains are an artifact of the relabeling, not an independent validation.

  1. fitted input called prediction [Section III-B and Section IV-E, Table IV]
    "We used visual observations and confusion matrix analyzes to target classes with frequent misclassifications. For these classes, we meticulously moved or removed data samples based on the refined criteria. ... All models exhibit improved performance on ModelNet-R, demonstrating the importance of high-quality datasets in enhancing classification accuracy."

    The refinement is guided by the confusion behavior of classifiers, and the resulting dataset is then used as the evaluation benchmark. The 'improvement' on ModelNet-R relative to ModelNet is therefore not an independent test of data quality: it is the measured consequence of changing test labels and removing/relabeling samples (e.g., 171 plant->flower_pot, 72 flower_pot->vase, 19 removed) to match the refined class definitions that were motivated by those confusions. The claim that models 'exhibit significant performance improvements' is thus a prediction fitted to the input labels, not a first-principles result.

full rationale

The central empirical claim of the paper is that refining ModelNet into ModelNet-R improves classification accuracy across models. This claim is partially circular because the refinement process used confusion-matrix analyses of classifiers to decide which samples to relabel or remove (Section III-B), and the same type of classifiers are then evaluated on the resulting benchmark (Section IV-E, Table IV). The improved accuracy is therefore confounded with the authors' relabeling decisions and cannot be cleanly attributed to dataset quality. The Point-SkipNet architecture itself is not circular; it is a standard FPS + ball-query + MLP + max-pool + skip-connection pipeline. The 'state-of-the-art' claim is internally contradicted by Table IV (PointMLP 95.33% vs Point-SkipNet 94.33% OA), but that is a correctness/support issue rather than a circularity. Self-citations ([38],[39]) are contextual and not load-bearing. Because the dataset-quality conclusion is central and partly reduces to the benchmark construction, the circularity score is 6.

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

The paper introduces no new theoretical entities. The free parameters are the unspecified ball query radius and the hand-chosen architecture dimensions. The key assumptions are the reliability of subjective dataset curation and the comparability of accuracy across different test sets.

free parameters (4)
  • Ball query radius r
    Introduced in Eq. 5 as a predefined radius but never given a value; the architecture's behavior depends on it.
  • Neighbor counts k1, k2 = k1=32, k2=64
    Hand-chosen in Figure 6; part of the model design.
  • Layer widths d1, d2, d3 = 128, 256, 1024
    Chosen by hand; the capacity of Point-SkipNet depends on them.
  • Sampled point counts N1, N2 = 512, 128
    Set in Figure 6; these reduce the point cloud size at each stage.
assumptions (3)
  • domain assumption Mislabeled samples can be reliably identified by visual inspection and expert cross-referencing.
    Used in Section III-A1 to justify label corrections; no inter-annotator agreement or objective ground truth is provided.
  • ad hoc to paper Confusion matrix misclassifications indicate dataset label or class-definition problems rather than model capacity limits.
    Section III-B uses confusion matrix analysis to decide which samples to move or remove; this assumes model errors are data errors.
  • domain assumption Accuracy on the refined ModelNet-R test set is comparable to accuracy on the original ModelNet test set.
    The paper's headline improvement comparisons rely on this, but the two test sets differ in composition and size, so the comparison is not like-for-like.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing 3D Point Cloud Classification with ModelNet-R and Point-SkipNet." pith.science (2026). https://pith.science/paper/NDWWDDW6

@misc{pith2026250905198,
  author       = {Pith},
  title        = {Pith review of: Enhancing 3D Point Cloud Classification with ModelNet-R and Point-SkipNet},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NDWWDDW6}},
  note         = {Machine review of arXiv:2509.05198}
}
read the original abstract

The classification of 3D point clouds is crucial for applications such as autonomous driving, robotics, and augmented reality. However, the commonly used ModelNet40 dataset suffers from limitations such as inconsistent labeling, 2D data, size mismatches, and inadequate class differentiation, which hinder model performance. This paper introduces ModelNet-R, a meticulously refined version of ModelNet40 designed to address these issues and serve as a more reliable benchmark. Additionally, this paper proposes Point-SkipNet, a lightweight graph-based neural network that leverages efficient sampling, neighborhood grouping, and skip connections to achieve high classification accuracy with reduced computational overhead. Extensive experiments demonstrate that models trained in ModelNet-R exhibit significant performance improvements. Notably, Point-SkipNet achieves state-of-the-art accuracy on ModelNet-R with a substantially lower parameter count compared to contemporary models. This research highlights the crucial role of dataset quality in optimizing model efficiency for 3D point cloud classification. For more details, see the code at: https://github.com/m-saeid/ModeNetR_PointSkipNet.

Figures

Figures reproduced from arXiv: 2509.05198 by the authors.

Figure 1
Figure 1. shows some of the problems, including instances of misclassified objects and datasets that cannot be reliably classified. Existing point-based models like PointMLP [10] and APES [11] have set impressive benchmarks for point cloud classifica- (a) 2D data (b) Multi object (c) Wrong label [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. General architecture of Point-SkipNet in resource-limited environments. • Performance Evaluation: Evaluating the impact of dataset quality and model efficiency on 3D classification accuracy, demonstrating their combined potential for im￾proved performance. II. RELATED WORKS A. 3D Point Cloud Datasets Over the years, numerous 3D datasets have been introduced to drive research in areas such as autonomous driving [13]–… view at source ↗
Figure 3
Figure 3. Flattened 2D structures in ModelNet lacking volumetric [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (5 more)
Figure 5
Figure 5. Figure 5: Examples of similar geometries in ModelNet classes. R 3 . Formally, a point cloud P with N points can be written as P = {pi} N i=1 ⊂ R 3 , where each point pi is: pi = (xi , yi , zi). (1) Point cloud classification involves learning a function cls that maps P to a clas…
Figure 4
Figure 4. Figure 4: Size inconsistencies: normalized objects appear similar [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 6
Figure 6. Figure 6: Point-SkipNet architecture. (a) Flower_pot (b) Vase [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Comparison of vase and flower pot with high surface [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]
Figure 8
Figure 8. Figure 8: Sample and Group Module in Point-SkipNet. [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 35 canonical work pages

  1. [1]

    Discovering new shadow patterns for black- box attacks on lane detection of autonomous vehicles

    Pedram MohajerAnsari, Alkim Domeke, Jan de V oor, Arkajyoti Mi- tra, Grace Johnson, Amir Salarpour, Habeeb Olufowobi, Mohammad Hamad, and Mert D Pesé. Discovering new shadow patterns for black- box attacks on lane detection of autonomous vehicles. arXiv preprint arXiv:2409.18248, 2024

    arXiv 2024

  2. [2]

    Avoiding the crash: A vision language model evaluation in critical traffic scenarios

    David Fernandez, Pedram MohajerAnsari, Amir Salarpour, and Mert D Pesé. Avoiding the crash: A vision language model evaluation in critical traffic scenarios. Technical report, SAE Technical Paper, 2025

    work page 2025

  3. [3]

    Grasping pose estimation for scara robot based on deep learning of point cloud

    Zhengtuo Wang, Yuetong Xu, Quan He, Zehua Fang, Guanhua Xu, and Jianzhong Fu. Grasping pose estimation for scara robot based on deep learning of point cloud. The International Journal of Advanced Manufacturing Technology, 108:1217–1231, 2020

    work page 2020

  4. [4]

    Long-term estimation of human spatial interactions through multiple laser ranging sensors

    Amir Salarpour, Hassan Khotanlou, and Nikolaos Mavridis. Long-term estimation of human spatial interactions through multiple laser ranging sensors. In 2014 International Conference on Robotics and Emerging Allied Technologies in Engineering (iCREATE) , pages 109–114. IEEE, 2014

    work page 2014

  5. [5]

    Bim-based registration and localization of 3d point clouds of indoor scenes using geometric features for augmented reality

    Bilawal Mahmood, SangUk Han, and Dong-Eun Lee. Bim-based registration and localization of 3d point clouds of indoor scenes using geometric features for augmented reality. Remote Sensing, 12(14):2302, 2020

    work page 2020

  6. [6]

    A mobile augmented reality system for the real-time visualization of pipes in point cloud data with a depth sensor

    Young-Hoon Jin, In-Tae Hwang, and Won-Hyung Lee. A mobile augmented reality system for the real-time visualization of pipes in point cloud data with a depth sensor. Electronics, 9(5):836, 2020

    work page 2020

  7. [7]

    Point cloud generation using deep adversarial local features for augmented and mixed reality contents

    Sohee Lim, Minwoo Shin, and Joonki Paik. Point cloud generation using deep adversarial local features for augmented and mixed reality contents. IEEE Transactions on Consumer Electronics , 68(1):69–76, 2022

    work page 2022

  8. [8]

    Graph neural networks in point clouds: A survey

    Dilong Li, Chenghui Lu, Ziyi Chen, Jianlong Guan, Jing Zhao, and Jixiang Du. Graph neural networks in point clouds: A survey. Remote Sensing, 16(14):2518, 2024

    work page 2024

Show all 41 references
  1. [9]

    3d shapenets: A deep representation for volumetric shapes

    Zhirong Wu, Shuran Song, Aditya Khosla, Fisher Yu, Linguang Zhang, Xiaoou Tang, and Jianxiong Xiao. 3d shapenets: A deep representation for volumetric shapes. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , June 2015

    2015

  2. [10]

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

    Xu Ma, Can Qin, Haoxuan You, Haoxi Ran, and Yun Fu. Rethinking network design and local geometry in point cloud: A simple residual mlp framework. arXiv preprint arXiv:2202.07123 , 2022

    2022 arXiv

  3. [11]

    Attention-based point cloud edge sampling

    Chengzhi Wu, Junwei Zheng, Julius Pfrommer, and Jürgen Beyerer. Attention-based point cloud edge sampling. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 5333–5343, 2023

    2023

  4. [12]

    Mohammed Aldeen, Pedram MohajerAnsari, Jin Ma, Mashrur Chowd- hury, Long Cheng, and Mert D. Pesé. An initial exploration of employing large multimodal models in defending against autonomous vehicles attacks. In 2024 IEEE Intelligent Vehicles Symposium (IV), pages 3334– 3341, 2024

    2024

  5. [13]

    Deep learning for lidar point clouds in autonomous driving: A review

    Ying Li, Lingfei Ma, Zilong Zhong, Fei Liu, Michael A Chapman, Dongpu Cao, and Jonathan Li. Deep learning for lidar point clouds in autonomous driving: A review. IEEE Transactions on Neural Networks and Learning Systems , 32(8):3412–3432, 2020

    2020

  6. [14]

    Deep learning for image and point cloud fusion in autonomous driving: A review

    Yaodong Cui, Ren Chen, Wenbo Chu, Long Chen, Daxin Tian, Ying Li, and Dongpu Cao. Deep learning for image and point cloud fusion in autonomous driving: A review. IEEE Transactions on Intelligent Transportation Systems, 23(2):722–739, 2021

    2021

  7. [15]

    Lidar point cloud compression, processing and learning for autonomous driving

    Rashid Abbasi, Ali Kashif Bashir, Hasan J Alyamani, Farhan Amin, Jaehyeok Doh, and Jianwen Chen. Lidar point cloud compression, processing and learning for autonomous driving. IEEE Transactions on Intelligent Transportation Systems , 24(1):962–979, 2022

    2022

  8. [16]

    Robotics dexterous grasping: The methods based on point cloud and deep learning

    Haonan Duan, Peng Wang, Yayu Huang, Guangyun Xu, Wei Wei, and Xiaofei Shen. Robotics dexterous grasping: The methods based on point cloud and deep learning. Frontiers in Neurorobotics, 15:658280, 2021

    2021

  9. [17]

    Trajectory planning and optimization for robotic machining based on measured point cloud

    Gang Wang, Wenlong Li, Cheng Jiang, Dahu Zhu, Zhongwei Li, Wei Xu, Huan Zhao, and Han Ding. Trajectory planning and optimization for robotic machining based on measured point cloud. IEEE transactions on robotics, 38(3):1621–1637, 2021

    2021

  10. [18]

    Shapenet: An information-rich 3d model repository

    Angel X Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, et al. Shapenet: An information-rich 3d model repository. arXiv preprint arXiv:1512.03012, 2015

    2015 arXiv

  11. [19]

    Revisiting point cloud classification: A new benchmark dataset and classification model on real-world data

    Mikaela Angelina Uy, Quang-Hieu Pham, Binh-Son Hua, Thanh Nguyen, and Sai-Kit Yeung. Revisiting point cloud classification: A new benchmark dataset and classification model on real-world data. In Proceedings of the IEEE/CVF international conference on computer vision, pages 15...

    2019

  12. [20]

    3d semantic parsing of large-scale indoor spaces

    Iro Armeni, Ozan Sener, Amir R Zamir, Helen Jiang, Ioannis Brilakis, Martin Fischer, and Silvio Savarese. 3d semantic parsing of large-scale indoor spaces. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 1534–1543, 2016

    2016

  13. [21]

    Sun rgb-d: A rgb-d scene understanding benchmark suite

    Shuran Song, Samuel P Lichtenberg, and Jianxiong Xiao. Sun rgb-d: A rgb-d scene understanding benchmark suite. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 567–576, 2015

    2015

  14. [22]

    nuscenes: A multimodal dataset for autonomous driving

    Holger Caesar, Varun Bankiti, Alex H Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Giancarlo Baldan, and Oscar Beijbom. nuscenes: A multimodal dataset for autonomous driving. In Proceedings of the IEEE/CVF conference on computer vision and pattern r...

    2020

  15. [23]

    Vision meets robotics: The kitti dataset

    Andreas Geiger, Philip Lenz, Christoph Stiller, and Raquel Urtasun. Vision meets robotics: The kitti dataset. The International Journal of Robotics Research, 32(11):1231–1237, 2013

    2013

  16. [24]

    V oxelnet: End-to-end learning for point cloud based 3d object detection

    Yin Zhou and Oncel Tuzel. V oxelnet: End-to-end learning for point cloud based 3d object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 4490–4499, 2018

    2018

  17. [25]

    Projection-based point convolution for efficient point cloud segmentation

    Pyunghwan Ahn, Juyoung Yang, Eojindl Yi, Chanho Lee, and Junmo Kim. Projection-based point convolution for efficient point cloud segmentation. IEEE Access, 10:15348–15358, 2022

    2022

  18. [26]

    Semantic segmentation of 3d lidar data using deep learning: a review of projection-based methods

    Alok Jhaldiyal and Navendu Chaudhary. Semantic segmentation of 3d lidar data using deep learning: a review of projection-based methods. Applied Intelligence, 53(6):6844–6855, 2023

    2023

  19. [27]

    A fast and efficient projection-based approach for surface reconstruction

    M Gopi and Shankar Krishnan. A fast and efficient projection-based approach for surface reconstruction. In Proceedings. XV Brazilian Symposium on Computer Graphics and Image Processing , pages 179–

  20. [28]

    Back- propagation applied to handwritten zip code recognition

    Yann LeCun, Bernhard Boser, John S Denker, Donnie Henderson, Richard E Howard, Wayne Hubbard, and Lawrence D Jackel. Back- propagation applied to handwritten zip code recognition. Neural com- putation, 1(4):541–551, 1989

    1989

  21. [29]

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

    Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets for 3d classification and segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 652–660, 2017

    2017

  22. [30]

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

    Charles Ruizhongtai Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointnet++: Deep hierarchical feature learning on point sets in a metric space. Advances in neural information processing systems , 30, 2017

    2017

  23. [31]

    Dynamic graph cnn for learning on point clouds

    Yue Wang, Yongbin Sun, Ziwei Liu, Sanjay E Sarma, Michael M Bronstein, and Justin M Solomon. Dynamic graph cnn for learning on point clouds. ACM Transactions on Graphics (tog) , 38(5):1–12, 2019

    2019

  24. [32]

    Kpconv: Flexible and deformable convolution for point clouds

    Hugues Thomas, Charles R Qi, Jean-Emmanuel Deschaud, Beatriz Marcotegui, François Goulette, and Leonidas J Guibas. Kpconv: Flexible and deformable convolution for point clouds. In Proceedings of the IEEE/CVF international conference on computer vision , pages 6411– 6420, 2019

    2019

  25. [33]

    Point transformer

    Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip HS Torr, and Vladlen Koltun. Point transformer. In Proceedings of the IEEE/CVF interna- tional conference on computer vision , pages 16259–16268, 2021

    2021

  26. [34]

    Point transformer v2: Grouped vector attention and partition-based pooling

    Xiaoyang Wu, Yixing Lao, Li Jiang, Xihui Liu, and Hengshuang Zhao. Point transformer v2: Grouped vector attention and partition-based pooling. Advances in Neural Information Processing Systems, 35:33330– 33342, 2022

    2022

  27. [35]

    Point transformer v3: Simpler faster stronger

    Xiaoyang Wu, Li Jiang, Peng-Shuai Wang, Zhijian Liu, Xihui Liu, Yu Qiao, Wanli Ouyang, Tong He, and Hengshuang Zhao. Point transformer v3: Simpler faster stronger. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 4840– 4851, 2024

    2024

  28. [36]

    Pct: Point cloud transformer

    Meng-Hao Guo, Jun-Xiong Cai, Zheng-Ning Liu, Tai-Jiang Mu, Ralph R Martin, and Shi-Min Hu. Pct: Point cloud transformer. Computational Visual Media, 7:187–199, 2021

    2021

  29. [37]

    Parameter is not all you need: Starting from non-parametric networks for 3d point cloud analysis

    Renrui Zhang, Liuhui Wang, Ziyu Guo, Yali Wang, Peng Gao, Hong- sheng Li, and Jianbo Shi. Parameter is not all you need: Starting from non-parametric networks for 3d point cloud analysis. arXiv preprint arXiv:2303.08134, 2023

    2023 arXiv

  30. [38]

    Point-gn: A non-parametric network using gaussian positional encoding for point cloud classifica- tion

    Marzieh Mohammadi and Amir Salarpour. Point-gn: A non-parametric network using gaussian positional encoding for point cloud classifica- tion. arXiv preprint arXiv:2412.03056 , 2024

    2024 arXiv

  31. [39]

    Point-ln: A lightweight framework for efficient point cloud classification using non-parametric positional encoding

    Marzieh Mohammadi, Amir Salarpour, and Pedram MohajerAnsari. Point-ln: A lightweight framework for efficient point cloud classification using non-parametric positional encoding. pages 1–7, 2025

    2025

  32. [40]

    Point-planenet: Plane kernel based con- volutional neural network for point clouds analysis

    SM Moein Peyghambarzadeh, Fatemeh Azizmalayeri, Hassan Khotan- lou, and Amir Salarpour. Point-planenet: Plane kernel based con- volutional neural network for point clouds analysis. Digital Signal Processing, 98:102633, 2020

    2020

  33. [41]

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

    Tiange Xiang, Chaoyi Zhang, Yang Song, Jianhui Yu, and Weidong Cai. Walk in the cloud: Learning curves for point clouds shape analysis. In Proceedings of the IEEE/CVF international conference on computer vision, pages 915–924, 2021

    2021

Pith tools

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