Pith. sign in

REVIEW 3 major objections 5 minor 57 references

PointGAC: Geometric-Aware Codebook for Masked Point Cloud Modeling

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read PointGAC replaces regression-based masked point cloud modeling with a clustering-based codebook alignment: the student learns which cluster center a masked patch belongs to, avoiding over-constrained feature regression and reporting gains…

desk verdict PointGAC reports solid, consistent gains across four point cloud tasks with a clustering-based MPM objective, but the ablation doesn't isolate the 'avoid over-constraining' claim and a few equations are under-specified. read the letter →

arxiv 2507.04801 v2 pith:CMWXWB6S submitted 2025-07-07 cs.CV

Abiao Li , Chenlei Lv , Yuming Fang , Yifan Zuo , Jian Zhang , Guofeng Mei This is my paper
classification cs.CV
keywords pointcloudmaskedmodelingself-supervisedrepresentationlearningonlinecodebookk-meansclusteringteacher-studentframeworkgeometry-awarepartitioningKLdivergencealignment
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

This paper claims that masked point cloud modeling should be framed as a clustering problem instead of a regression problem. It introduces a teacher-student setup in which the teacher's patch features update an online codebook by k-means, and the student learns to predict which codebook cluster center each masked patch belongs to, rather than reproducing the teacher's exact features. The paper argues that this avoids over-constraining the student on fine details and yields more generalizable representations. It reports improved downstream accuracy on classification, few-shot classification, segmentation, and completion benchmarks, with the main results including 94.6% accuracy on ModelNet40 with voting and 61.8 mIoU on S3DIS.

What carries the argument

The load-bearing object is an online codebook $C = [c_1,\dots,c_K]$ of $D$-dimensional vectors that serve as cluster centers for patch features. It is updated by exponential-moving-average online k-means: each teacher feature is assigned to its nearest codebook vector, and accumulated counts and sums update the vector toward the cluster center with momentum $\gamma=0.99$. Teacher and student soft assignments $Q_t$ and $Q_s$ over the codebook are generated by softmax over codebook dot products with temperatures $\tau_t$ (cosine-annealed) and $\tau_s$ (fixed, larger), and the student is trained to match the teacher's assignment via KL divergence. A maintenance mechanism perturbs dead vectors with a sigmoid-weighted blend of the vector and the most similar teacher feature, and a geometry-aware partitioning strategy uses Potts energy segmentation plus masked optimal transport, solved by Sinkhorn, to produce non-overlapping geometrically homogeneous patches.

What would settle it

Use the same pretraining pipeline but replace the teacher-derived cluster assignments for masked patches with random cluster labels; if downstream accuracy stays at the reported level, then alignment to true cluster membership is not what drives the gains. A complementary check is to visualize patches that share a codebook center: if they are geometrically or semantically unrelated, the central premise fails.

Watch

Extended reading notes

Core claim

The central claim is that aligning feature distributions over a learned codebook is a better pretext objective than aligning exact point-wise features. Concretely, the teacher encodes complete patches, and its features update codebook vectors through exponential-moving-average online k-means so each vector becomes a cluster center. Teacher features are converted to soft assignment distributions over the codebook with a cosine-annealed temperature; the student encoder-decoder sees only visible patches and produces reconstructed masked features, which are converted to softer assignments; the two distributions are matched with a KL loss. Two auxiliary components support the claim: a geometry-aware partitioning strategy that cuts the point cloud into non-overlapping, geometrically homogeneous patches via Potts energy segmentation and masked optimal transport, and a codebook maintenance mechanism that perturbs rarely updated vectors toward a similar teacher feature so dead codebook entries stay active. On this basis the paper reports the best results among its comparisons: 94.6% overall accuracy on ModelNet40 with voting, 89.4% on ScanObjectNN PB-T50-RS, 61.8 mIoU on S3DIS, and 1.63 Chamfer distance on ShapeNet55 hard completion.

Load-bearing premise

The method assumes that online k-means over the teacher's patch features produces cluster centers that correspond to meaningful geometric or semantic categories, so that training the student to predict cluster membership teaches useful representations rather than arbitrary labels.

Editorial extensions

If this is right

  • If the clustering paradigm is right, masked-patch pretext tasks can drop point-wise feature regression without losing representational quality, simplifying the optimization target.
  • The reported numbers imply that a codebook objective transfers across classification, segmentation, and completion with a single pretrained encoder.
  • The ablation suggests that reviving dead codebook vectors is necessary for the clustering objective to outperform the regression baseline.
  • The geometry-aware partitioning result implies that non-overlapping, geometrically consistent patches make the pretext task harder and the learned features better.
  • The mask-ratio experiments indicate that with such partitioning, higher masking ratios can strengthen the learned representations.

Reading between the lines

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

  • Beyond the paper's reported benchmarks, the codebook centers could be inspected directly: if centers correspond to geometric shapes or semantic parts, the method provides interpretable tokens for 3D scenes, which the paper does not demonstrate.
  • The same cluster-center alignment could transfer to other 3D inputs such as depth maps or LiDAR scans, where exact feature regression is even more sensitive to sensor noise; this extension is not tested in the paper.
  • A reader could isolate whether the gain comes from clustering per se or from the additional regularization of the soft assignments by comparing against a regression baseline with the same patch partitioning and maintenance mechanism.
  • The cosine-annealed teacher temperature is a potential sensitivity point: training dynamics may depend on how quickly the teacher assignments sharpen, which the paper does not analyze.
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 / 5 minor

Summary. The paper proposes PointGAC, a masked point cloud modeling method that replaces the common regression objective (predicting the teacher's exact masked features) with a clustering objective: the teacher updates an online k-means codebook on complete-patch features, and the student is trained to match the teacher's soft assignment of masked features to codebook centers via a KL divergence loss. The method also introduces a geometry-aware partitioning strategy (GAP) based on Potts energy segmentation and masked optimal transport, and a codebook maintenance mechanism that perturbs dead codebook vectors with the most similar teacher features. The authors report consistent improvements over prior self-supervised point cloud methods on ModelNet40 classification, ScanObjectNN classification, few-shot classification, S3DIS semantic segmentation, ShapeNetPart part segmentation, and ShapeNet55 completion, with code publicly available.

Significance. If the results hold, PointGAC is a useful contribution to self-supervised point cloud representation learning: it achieves state-of-the-art or competitive numbers on four downstream task families with a single pretraining recipe, and the public code plus detailed ablations (Tables 5-7) make the empirical claims largely reproducible. The paper also ships a falsifiable design hypothesis (clustering-based alignment generalizes better than feature regression), and the downstream evaluations provide independent support that the pretrained representations are useful. The main weakness is that the ablation evidence does not cleanly isolate the proposed 'clustering avoids over-constraining' mechanism, and two formal parts of the method description (the KL objective and the optimal-transport constraint in Eq. 2) are incomplete or incorrect as written.

major comments (3)
  1. [Sec. 3.3, Table 5]
  2. [Sec. 3.2]
  3. [Sec. 3.1, Eq. (2)]
minor comments (5)
  1. [Abstract and Sec. 1]
  2. [Sec. 3.2, Eq. (7)]
  3. [Sec. 4.1, part segmentation]
  4. [Sec. 3.1]
  5. [References]

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: PointGAC is an empirical self-supervised pretext objective evaluated on external benchmarks; the cited prior works are contextual and not load-bearing.

full rationale

PointGAC does not present a first-principles derivation or a fitted prediction; it introduces a self-supervised training objective in which teacher features update a codebook, teacher and student soft assignments over that codebook are aligned with KL divergence, and the method is evaluated by downstream transfer accuracy. The only self-referential loop is the normal self-supervised one: codebook vectors are defined by the teacher features, and the codebook in turn defines the assignment targets. That loop is not a circularity in the paper's argument because the claimed validation is external: ModelNet40 classification, ScanObjectNN classification, S3DIS and ShapeNetPart segmentation, and ShapeNet55 completion. The codebook and assignment machinery are internal training tools, not quantities fitted to those benchmarks. Self-citations (e.g., refs. 23, 24, 32, 33) appear only in the introduction as examples of prior point-cloud understanding work and are not used to justify the method's design choices or to forbid alternatives. No uniqueness theorem, ansatz, or fitted value is imported from author-only work. Ablation results in Table 5 that show online k-means alone (Model B) underperforming regression are an empirical attribution concern about which component drives the gains, but they do not make any prediction equivalent to its input by construction, so they do not constitute circularity.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The method rests on several hand-chosen hyperparameters (annealed temperatures, codebook size, mask ratios, an unreported PES mu) and on the domain assumption that patch features cluster into semantically meaningful codebook centers. No new physical or model-level entities are introduced.

free parameters (6)
  • codebook_size_K = 8192
    Set following OBoW [12], not tuned on a validation set.
  • EMA_momentum_gamma = 0.99
    Chosen by hand to keep codebook updates stable (Sec 3.2).
  • teacher_temperature_tau_t = cosine 0.07 to 0.04
    Annealed during training to sharpen assignment distributions (Sec 3.2).
  • student_temperature_tau_s = 0.1
    Fixed to keep student assignments smoother than teacher assignments (Sec 3.2).
  • PES_mu = not reported
    Regularization weight in Eq 1 that controls partition coarseness; its value is not given.
  • mask_ratio_r = 0.7, 0.8, or 0.9
    Chosen per experiment; Table 6 shows sensitivity to this value.
assumptions (4)
  • domain assumption PES produces geometrically meaningful segments that are stable across point clouds.
    Used to construct non-overlapping patches in Sec 3.1; if segments are arbitrary, patch embeddings lose geometric meaning.
  • domain assumption Online k-means codebook vectors converge to cluster centers that capture useful semantic categories.
    Core of Sec 3.2; the teacher assignments serve as targets for the student.
  • standard math EMA teacher update yields stable and improving targets.
    Common practice from BYOL/MAE-based methods; the paper relies on it without proof (Sec 3.2).
  • ad hoc to paper Aligning cluster assignments instead of raw features reduces over-constraining and improves generalization.
    This is the paper's central hypothesis (Sec 1, Sec 3.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of PointGAC: Geometric-Aware Codebook for Masked Point Cloud Modeling." pith.science (2026). https://pith.science/paper/CMWXWB6S

@misc{pith2026250704801,
  author       = {Pith},
  title        = {Pith review of: PointGAC: Geometric-Aware Codebook for Masked Point Cloud Modeling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CMWXWB6S}},
  note         = {Machine review of arXiv:2507.04801}
}
read the original abstract

Most masked point cloud modeling (MPM) methods follow a regression paradigm to reconstruct the coordinate or feature of masked regions. However, they tend to over-constrain the model to learn the details of the masked region, resulting in failure to capture generalized features. To address this limitation, we propose \textbf{\textit{PointGAC}}, a novel clustering-based MPM method that aims to align the feature distribution of masked regions. Specially, it features an online codebook-guided teacher-student framework. Firstly, it presents a geometry-aware partitioning strategy to extract initial patches. Then, the teacher model updates a codebook via online k-means based on features extracted from the complete patches. This procedure facilitates codebook vectors to become cluster centers. Afterward, we assigns the unmasked features to their corresponding cluster centers, and the student model aligns the assignment for the reconstructed masked features. This strategy focuses on identifying the cluster centers to which the masked features belong, enabling the model to learn more generalized feature representations. Benefiting from a proposed codebook maintenance mechanism, codebook vectors are actively updated, which further increases the efficiency of semantic feature learning. Experiments validate the effectiveness of the proposed method on various downstream tasks. Code is available at https://github.com/LAB123-tech/PointGAC

Figures

Figures reproduced from arXiv: 2507.04801 by the authors.

Figure 1
Figure 1. Fundamental concept of PointGAC. Both the unmasked teacher features and the reconstructed masked features at corre￾sponding positions are assigned to the same cluster center over the codebook. This clustering paradigm avoids forcing the student model to overfit the exact feature details of the teacher model. ing research interest. This is achieved by designing pretext tasks that encourage the model to capture meanin… view at source ↗
Figure 2
Figure 2. The framework of PointGAC. This framework utilizes a online codebook to achieve feature alignment. It aligns the reconstructed features on the student side with the encoded features on the teacher side at corresponding positions. the teacher and student assignments are aligned using a KL divergence loss. During training, gradients are propagated only through the student branch, while the teacher encoder is updated u… view at source ↗
Figure 3
Figure 3. Visualization comparison of codebook update frequency [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison on S3DIS semantic segmentation. The first column shows the original point cloud input, followed by [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Qualitative comparison of different algorithms on [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

57 extracted references · 49 canonical work pages

  1. [1]

    Semi-supervised learning of visual features by non- parametrically predicting view assignments with support samples

    Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bo- janowski, Armand Joulin, Nicolas Ballas, and Michael Rab- bat. Semi-supervised learning of visual features by non- parametrically predicting view assignments with support samples. In ICCV, pages 8443–8452, 2021. 2

    work page 2021

  2. [2]

    Masked siamese net- works for label-efficient learning

    Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bo- janowski, Florian Bordes, Pascal Vincent, Armand Joulin, Mike Rabbat, and Nicolas Ballas. Masked siamese net- works for label-efficient learning. In ECCV, pages 456–473. Springer, 2022. 2

    work page 2022

  3. [4]

    Beit: Bert pre-training of image transformers

    Hangbo Bao, Li Dong, Songhao Piao, and Furu Wei. Beit: Bert pre-training of image transformers. arXiv preprint arXiv:2106.08254, 2021. 3

    arXiv 2021

  4. [5]

    Unsupervised learn- ing of visual features by contrasting cluster assignments

    Mathilde Caron, Ishan Misra, Julien Mairal, Priya Goyal, Pi- otr Bojanowski, and Armand Joulin. Unsupervised learn- ing of visual features by contrasting cluster assignments. NeurIPS, 33:9912–9924, 2020. 2, 8

    work page 2020

  5. [6]

    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. 5

    arXiv 2015

  6. [7]

    Pointgpt: Auto-regressively generative pre- training from point clouds

    Guangyan Chen, Meiling Wang, Yi Yang, Kai Yu, Li Yuan, and Yufeng Yue. Pointgpt: Auto-regressively generative pre- training from point clouds. NeurIPS, 36, 2024. 1, 6

    work page 2024

  7. [8]

    Lightspeed computation of optimal transport

    Cuturi M Sinkhorn Distances. Lightspeed computation of optimal transport. NeurIPS, 26:2292–2300, 2013. 2, 3, 8

    work page 2013

  8. [9]

    Point-mpp: Point cloud self-supervised learning from masked position prediction

    Songlin Fan, Wei Gao, and Ge Li. Point-mpp: Point cloud self-supervised learning from masked position prediction. IEEE TNNLS, pages 1–13, 2024. 6

    work page 2024

Show all 57 references
  1. [10]

    Pos-bert: Point cloud one-stage bert pre-training

    Kexue Fu, Peng Gao, ShaoLei Liu, Linhao Qu, Longxiang Gao, and Manning Wang. Pos-bert: Point cloud one-stage bert pre-training. ESWA, 240:122563, 2024. 1, 2, 6

    2024

  2. [11]

    Human pose as compositional tokens

    Zigang Geng, Chunyu Wang, Yixuan Wei, Ze Liu, Houqiang Li, and Han Hu. Human pose as compositional tokens. In CVPR, pages 660–671, 2023. 2

    2023

  3. [12]

    Obow: Online bag-of-visual-words generation for self-supervised learning

    Spyros Gidaris, Andrei Bursuc, Gilles Puy, Nikos Ko- modakis, Matthieu Cord, and Patrick P ´erez. Obow: Online bag-of-visual-words generation for self-supervised learning. In CVPR, pages 6830–6840, 2021. 2, 5

    2021

  4. [13]

    Weakly supervised segmentation-aided classification of urban scenes from 3d li- dar point clouds

    St ´ephane Guinard and Loic Landrieu. Weakly supervised segmentation-aided classification of urban scenes from 3d li- dar point clouds. ISPRS Archives, 42:151–157, 2017. 3

    2017

  5. [14]

    Momentum contrast for unsupervised visual rep- resentation learning

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual rep- resentation learning. In CVPR, pages 9729–9738, 2020. 8

    2020

  6. [15]

    Masked autoencoders are scalable vision learners

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Doll´ar, and Ross Girshick. Masked autoencoders are scalable vision learners. In CVPR, pages 16000–16009, 2022. 3

    2022

  7. [16]

    Masked autoencoders are scalable vision learners

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Doll´ar, and Ross Girshick. Masked autoencoders are scalable vision learners. In CVPR, pages 16000–16009, 2022. 1

    2022

  8. [17]

    3d-jepa: A joint embedding predictive architecture for 3d self-supervised representation learning

    Naiwen Hu, Haozhe Cheng, Yifan Xie, Shiqi Li, and Jihua Zhu. 3d-jepa: A joint embedding predictive architecture for 3d self-supervised representation learning. arXiv preprint arXiv:2409.15803, 2024. 1, 2, 6

    2024 arXiv

  9. [18]

    Masked autoencoders in 3d point cloud repre- sentation learning

    Jincen Jiang, Xuequan Lu, Lizhi Zhao, Richard Dazaley, and Meili Wang. Masked autoencoders in 3d point cloud repre- sentation learning. IEEE TMM, 2023. 2, 6

    2023

  10. [19]

    A convex relax- ation for weakly supervised classifiers

    Armand Joulin and Francis Bach. A convex relax- ation for weakly supervised classifiers. arXiv preprint arXiv:1206.6413, 2012. 2

    2012 arXiv

  11. [20]

    Adam: A method for stochastic opti- mization

    Diederik P Kingma. Adam: A method for stochastic opti- mization. arXiv preprint arXiv:1412.6980, 2014. 5

    2014 arXiv

  12. [21]

    Stratified trans- former for 3d point cloud segmentation

    Xin Lai, Jianhui Liu, Li Jiang, Liwei Wang, Hengshuang Zhao, Shu Liu, Xiaojuan Qi, and Jiaya Jia. Stratified trans- former for 3d point cloud segmentation. In CVPR, pages 8500–8509, 2022. 1

    2022

  13. [22]

    Large-scale point cloud semantic segmentation with superpoint graphs

    Loic Landrieu and Martin Simonovsky. Large-scale point cloud semantic segmentation with superpoint graphs. In CVPR, pages 4558–4567, 2018. 3

    2018

  14. [23]

    Laptran: Transformer embedding graph laplacian for point cloud part segmentation

    Abiao Li, Chenlei Lv, Yuming Fang, and Yifan Zuo. Laptran: Transformer embedding graph laplacian for point cloud part segmentation. In ICIP, pages 3070–3074. IEEE, 2023. 1

    2023

  15. [24]

    Gstran: Joint geometric and semantic co- herence for point cloud segmentation

    Abiao Li, Chenlei Lv, Guofeng Mei, Yifan Zuo, Jian Zhang, and Yuming Fang. Gstran: Joint geometric and semantic co- herence for point cloud segmentation. In ICPR, pages 453–

  16. [25]

    Cross-modal and uncertainty-aware agglomeration for open- vocabulary 3d scene understanding

    Jinlong Li, Cristiano Saltori, Fabio Poiesi, and Nicu Sebe. Cross-modal and uncertainty-aware agglomeration for open- vocabulary 3d scene understanding. In CVPR, pages 19390– 19400, 2025. 1

    2025

  17. [26]

    Patchmixing masked autoencoders for 3d point cloud self-supervised learning

    Chengxing Lin, Wenju Xu, Jian Zhu, Yongwei Nie, Ruichu Cai, and Xuemiao Xu. Patchmixing masked autoencoders for 3d point cloud self-supervised learning. IEEE TCSVT,

  18. [27]

    Masked discrimina- tion for self-supervised learning on point clouds

    Haotian Liu, Mu Cai, and Yong Jae Lee. Masked discrimina- tion for self-supervised learning on point clouds. In ECCV, pages 657–675. Springer, 2022. 6

    2022

  19. [28]

    Pointgame: Geometrically and adap- tively masked autoencoder on point clouds

    Yun Liu, Xuefeng Yan, Zhiqi Li, Zhilei Chen, Zeyong Wei, and Mingqiang Wei. Pointgame: Geometrically and adap- tively masked autoencoder on point clouds. IEEE TGRS, 61: 1–12, 2023. 1, 2, 6

    2023

  20. [29]

    Sgdr: Stochas- tic gradient descent with warm restarts

    Ilya Loshchilov and Frank Hutter. Sgdr: Stochas- tic gradient descent with warm restarts. arXiv preprint arXiv:1608.03983, 2016. 5

    2016 arXiv

  21. [30]

    3dctn: 3d convolution-transformer network for point cloud classification

    Dening Lu, Qian Xie, Kyle Gao, Linlin Xu, and Jonathan Li. 3dctn: 3d convolution-transformer network for point cloud classification. IEEE T-ITS, 23(12):24854–24865, 2022. 1

    2022

  22. [31]

    Mot: Masked opti- mal transport for partial domain adaptation

    You-Wei Luo and Chuan-Xian Ren. Mot: Masked opti- mal transport for partial domain adaptation. In CVPR, pages 3531–3540. IEEE, 2023. 3

    2023

  23. [32]

    Data augmentation- free unsupervised learning for 3d point cloud understanding

    Guofeng Mei, Cristiano Saltori, Fabio Poiesi, Jian Zhang, Elisa Ricci, Nicu Sebe, and Qiang Wu. Data augmentation- free unsupervised learning for 3d point cloud understanding. arXiv preprint arXiv:2210.02798, 2022. 1

    2022 arXiv

  24. [33]

    Unsupervised point cloud representation learning by clustering and neural ren- dering

    Guofeng Mei, Cristiano Saltori, Elisa Ricci, Nicu Sebe, Qiang Wu, Jian Zhang, and Fabio Poiesi. Unsupervised point cloud representation learning by clustering and neural ren- dering. IJCV, 132(8):3251–3269, 2024. 1

    2024

  25. [34]

    Masked autoencoders for point cloud self-supervised learning

    Yatian Pang, Wenxiao Wang, Francis EH Tay, Wei Liu, Yonghong Tian, and Li Yuan. Masked autoencoders for point cloud self-supervised learning. In ECCV, pages 604–621. Springer, 2022. 1, 6

    2022

  26. [35]

    Clusterformer: Cluster-based transformer for 3d object detection in point clouds

    Yu Pei, Xian Zhao, Hao Li, Jingyuan Ma, Jingwei Zhang, and Shiliang Pu. Clusterformer: Cluster-based transformer for 3d object detection in point clouds. InICCV, pages 6664– 6673, 2023. 1

    2023

  27. [36]

    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 CVPR, pages 652–660, 2017. 4

    2017

  28. [37]

    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. NeruIPS, 30, 2017. 6

    2017

  29. [38]

    A conditional denoising diffusion proba- bilistic model for point cloud upsampling

    Wentao Qu, Yuantian Shao, Lingwu Meng, Xiaoshui Huang, and Liang Xiao. A conditional denoising diffusion proba- bilistic model for point cloud upsampling. In CVPR, pages 20786–20795, 2024. 1

    2024

  30. [39]

    An end-to-end robust point cloud semantic segmentation network with single-step conditional diffusion models

    Wentao Qu, Jing Wang, YongShun Gong, Xiaoshui Huang, and Liang Xiao. An end-to-end robust point cloud semantic segmentation network with single-step conditional diffusion models. In CVPR, pages 27325–27335, 2025. 1

    2025

  31. [40]

    Zero-shot text-to-image generation

    Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea V oss, Alec Radford, Mark Chen, and Ilya Sutskever. Zero-shot text-to-image generation. In ICML, pages 8821–

  32. [41]

    Point-jepa: A joint embedding predictive architecture for self-supervised learning on point cloud

    Ayumu Saito, Prachi Kudeshia, and Jiju Poovvancheri. Point-jepa: A joint embedding predictive architecture for self-supervised learning on point cloud. arXiv preprint arXiv:2404.16432, 2024. 6

    2024 arXiv

  33. [42]

    Ri-mae: Rotation-invariant masked autoencoders for self-supervised point cloud representation learning

    Kunming Su, Qiuxia Wu, Panpan Cai, Xiaogang Zhu, Xuequan Lu, Zhiyong Wang, and Kun Hu. Ri-mae: Rotation-invariant masked autoencoders for self-supervised point cloud representation learning. arXiv preprint arXiv:2409.00353, 2024. 1, 5, 6

    2024 arXiv

  34. [43]

    Point-msd: Jointly mamba self- supervised self-distilling point cloud representation learning

    Linzhi Su, Mengna Yang, Jie Liu, Xingxing Hao, Chenyang Zhang, and Xin Cao. Point-msd: Jointly mamba self- supervised self-distilling point cloud representation learning. In ICSP, pages 310–316. IEEE, 2024. 1, 2, 6

    2024

  35. [44]

    Point-lgmask: Local and global contexts embedding for point cloud pre-training with multi- ratio masking

    Yuan Tang, Xianzhi Li, Jinfeng Xu, Qiao Yu, Long Hu, Yixue Hao, and Min Chen. Point-lgmask: Local and global contexts embedding for point cloud pre-training with multi- ratio masking. IEEE TMM, 2023. 1, 2, 6

    2023

  36. [45]

    What do single-view 3d reconstruction networks learn? In CVPR, pages 3405–3414, 2019

    Maxim Tatarchenko, Stephan R Richter, Ren ´e Ranftl, Zhuwen Li, Vladlen Koltun, and Thomas Brox. What do single-view 3d reconstruction networks learn? In CVPR, pages 3405–3414, 2019. 7

    2019

  37. [46]

    Neural discrete representation learning

    Aaron Van Den Oord, Oriol Vinyals, et al. Neural discrete representation learning. NeurIPS, 30, 2017. 2, 4

    2017

  38. [47]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. NeurIPS, 30, 2017. 6

    2017

  39. [48]

    Rethinking masked representation learning for 3d point cloud understanding

    Chuxin Wang, Yixin Zha, Jianfeng He, Wenfei Yang, and Tianzhu Zhang. Rethinking masked representation learning for 3d point cloud understanding. IEEE TIP, 2024. 1, 2, 3, 6

    2024

  40. [49]

    3d feature prediction for masked-autoencoder-based point cloud pre- training

    Siming Yan, Yuqi Yang, Yuxiao Guo, Hao Pan, Peng-shuai Wang, Xin Tong, Yang Liu, and Qixing Huang. 3d feature prediction for masked-autoencoder-based point cloud pre- training. arXiv preprint arXiv:2304.06911, 2023. 6

    2023 arXiv

  41. [50]

    Pointr: Diverse point cloud completion with geometry-aware transformers

    Xumin Yu, Yongming Rao, Ziyi Wang, Zuyan Liu, Jiwen Lu, and Jie Zhou. Pointr: Diverse point cloud completion with geometry-aware transformers. In ICCV, pages 12498– 12507, 2021. 1

    2021

  42. [51]

    Pointr: Diverse point cloud completion with geometry-aware transformers

    Xumin Yu, Yongming Rao, Ziyi Wang, Zuyan Liu, Jiwen Lu, and Jie Zhou. Pointr: Diverse point cloud completion with geometry-aware transformers. In ICCV, pages 12498– 12507, 2021. 7

    2021

  43. [52]

    Point-BERT: Pre-training 3d point cloud transformers with masked point modeling

    Xumin Yu, Lulu Tang, Yongming Rao, Tiejun Huang, Jie Zhou, and Jiwen Lu. Point-BERT: Pre-training 3d point cloud transformers with masked point modeling. In CVPR, pages 19313–19322, 2022. 2, 6

    2022

  44. [53]

    Point2vec for self-supervised representa- tion learning on point clouds

    Karim Abou Zeid, Jonas Schult, Alexander Hermans, and Bastian Leibe. Point2vec for self-supervised representa- tion learning on point clouds. In GCPR, pages 131–146. Springer, 2023. 1, 2, 5, 6, 7

    2023

  45. [54]

    Towards compact 3d representations via point feature enhancement masked au- toencoders

    Yaohua Zha, Huizhen Ji, Jinmin Li, Rongsheng Li, Tao Dai, Bin Chen, Zhi Wang, and Shu-Tao Xia. Towards compact 3d representations via point feature enhancement masked au- toencoders. In AAAI, pages 6962–6970, 2024. 2

    2024

  46. [55]

    Point-m2ae: multi-scale masked autoencoders for hierarchical point cloud pre-training

    Renrui Zhang, Ziyu Guo, Peng Gao, Rongyao Fang, Bin Zhao, Dong Wang, Yu Qiao, and Hongsheng Li. Point-m2ae: multi-scale masked autoencoders for hierarchical point cloud pre-training. NeurIPS, 35:27061–27074, 2022. 2, 6

    2022

  47. [56]

    Pcp- mae: Learning to predict centers for point masked autoen- coders

    Xiangdong Zhang, Shaofeng Zhang, and Junchi Yan. Pcp- mae: Learning to predict centers for point masked autoen- coders. arXiv preprint arXiv:2408.08753, 2024. 1, 6

    2024 arXiv

  48. [57]

    ibot: Image bert pre-training with online tokenizer

    Jinghao Zhou, Chen Wei, Huiyu Wang, Wei Shen, Cihang Xie, Alan Yuille, and Tao Kong. ibot: Image bert pre-training with online tokenizer. arXiv preprint arXiv:2111.07832 ,

    arXiv

  49. [58]

    3d-oae: Occlusion auto- encoders for self-supervised learning on point clouds

    Junsheng Zhou, Xin Wen, Baorui Ma, Yu-Shen Liu, Yue Gao, Yi Fang, and Zhizhong Han. 3d-oae: Occlusion auto- encoders for self-supervised learning on point clouds. In ICRA, pages 15416–15423, 2024. 1

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.