Pith. sign in

REVIEW 3 major objections 4 minor 85 references

RWKV can be made to generalize across point cloud domains by replacing its grid-style token shift with a geometry-aware aggregation and aligning key-feature distributions among source domains.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

PointDGRWKV applies RWKV-like attention to domain-generalized point cloud classification, adding a geometric token shift and key-distribution alignment, and reports state-of-the-art accuracy on PointDA-10 and PointDG-3to1.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection Real empirical gains from an RWKV-style point cloud DG model, but the central mechanistic story for CD-KDA is undercut by softmax shift-invariance; still worth a serious referee. the 3 major comments →

arxiv 2508.20835 v2 pith:RKEPFJMZ submitted 2025-08-28 cs.CV

PointDGRWKV: Generalizing RWKV-like Architecture to Unseen Domains for Point Cloud Classification

Hao Yang , Qianyu Zhou , Haijia Sun , Xiangtai Li , Xuequan Lu , Lizhuang Ma , Shuicheng Yan This is my paper
classification cs.CV
keywords Domain generalizationPoint cloud classificationRWKVLinear complexityAdaptive geometric token shiftKey distribution alignmentBi-WKV attentionUnseen domains
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

The paper sets out to show that RWKV—a sequence architecture prized for global receptive fields and linear attention cost—can be made to generalize across point cloud domains, and that the two standard RWKV mechanisms actually hurt on unstructured 3D data. Fixed directional token shift (Q-Shift) assumes a grid-like token order and therefore distorts local neighborhoods of points; Bi-WKV attention places key vectors inside an exponential, so small shifts in key distributions between domains become large shifts in attention. The proposed PointDGRWKV counters the first with Adaptive Geometric Token Shift (AGT-Shift), a parameter-free spatial-hashing neighborhood aggregation, and the second with Cross-Domain Key Distribution Alignment (CD-KDA), a loss that matches key-vector means and covariances across source domains. The reported results are the best averages so far on the two DG benchmarks used—75.66% on PointDA-10 and 81.92% on PointDG-3to1—edging out the previous Mamba-based method while keeping linear complexity. If correct, this is evidence that linear-attention architectures can be competitive for domain-generalized 3D recognition once their position-biased mechanisms are re-derived for geometry.

Core claim

The paper's central claim is that RWKV can be made state-of-the-art for domain-generalized point cloud classification by fixing two specific architectural mismatches. First, Q-Shift's fixed grid-direction token shift is replaced by AGT-Shift, which uses spatial hashing to find local cells and aggregates neighbor features by distance to the cell's geometric center; this restores local geometry modeling without KNN or learned parameters. Second, CD-KDA aligns the first and second moments of the key vectors across source domains, because those keys enter the Bi-WKV attention weight through an exponential and their cross-domain drift is thereby amplified. The network is trained with classificati

What carries the argument

Bi-WKV attention is the exponential-in-k mechanism whose sensitivity motivates the alignment loss; AGT-Shift is the geometry-aware replacement for Q-Shift; CD-KDA is the alignment loss. AGT-Shift partitions the point cloud into spatial grid cells via hashing, then for each point computes a shifted feature as a weighted average of the features in its cell, with weights decaying by distance to the geometric center; it is O(N), parameter-free, and avoids pairwise distance computation. CD-KDA adds a term L_CD-KDA = pairwise sum over source domains of L2 distance between key means plus Frobenius distance between key covariance matrices, so the exponential 'keys' no longer differ systematically ac

Load-bearing premise

The load-bearing premise is that matching the mean and covariance of key vectors across source domains (Eq. 5) will also stabilize attention on an unseen target domain whose key statistics were never aligned, rather than merely overfitting to source-domain statistics.

What would settle it

Train PointDGRWKV with CD-KDA disabled but with a fixed affine transform per source domain that equalizes key means and covariances by construction; if the accuracy gain persists without optimizing the alignment loss, the proposed mechanism is not the cause. Alternatively, measure attention-weight KL divergence between source and target layers before and after CD-KDA: if target attention distributions do not move closer to source distributions while accuracy rises, the alignment story fails.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • RWKV-like linear-attention models can be the best reported DG PCC method, so domain generalization for 3D is not limited to CNNs, Transformers, or Mamba backbones.
  • AGT-Shift offers a parameter-free, O(N) neighborhood aggregation that could substitute KNN or graph construction in other point-cloud sequence models, removing a common scalability bottleneck.
  • The success of key-only alignment suggests attention-weight stability, not feature-content consistency, is the main lever for cross-domain robustness in exponential attention.
  • Since the architecture remains linear in sequence length and is smaller than several baselines (e.g., 2.13M parameters and 3.22 GFLOPs for the base variant), it is practical for deployment on larger or denser point clouds.
  • Larger variants of PointDGRWKV improve DG accuracy further, indicating that additional capacity and denser sampling help even when no target-domain data is available.

Where Pith is reading between the lines

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

  • A testable extension of the paper's diagnosis: any attention weight that exponentiates raw key scores should suffer the same cross-domain amplification, so CD-KDA-style key alignment may transfer to Transformer or hybrid attention blocks.
  • The paper leaves implicit that AGT-Shift's fixed grid hash may need adaptive cell sizes at varying point densities; testing density-aware grids could extend the method to sparse LiDAR-style scans.
  • The benchmarks use 5-10 shared classes with synthetic/real source domains; a more demanding setting (larger class sets, sensor-level LiDAR vs RGB-D shifts) would test whether second-order key alignment is enough.
  • Editorial note: aligning only mean and covariance assumes key shifts are roughly Gaussian; if target keys differ in higher-order structure, whitening or adversarial alignment would be a sterner test and might further improve transfer.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes PointDGRWKV, the first RWKV-based framework for domain generalization in point cloud classification. It introduces two modules: AGT-Shift, a parameter-free spatial-hashing-based token shift for local geometric modeling, and CD-KDA, a loss that aligns the mean and covariance of key features across source domains to reduce attention drift. Experiments on PointDA-10 and PointDG-3to1 report state-of-the-art average accuracies of 75.66% and 81.92%, respectively, with linear complexity and lower computational cost than Transformer- and Mamba-based baselines.

Significance. If the results hold, the paper opens a new architectural direction for DG PCC, demonstrating that RWKV-like models can outperform prior backbones on two standard benchmarks while retaining linear complexity. The comparisons in Table 1 and the ablations in Tables 2–4 provide a useful empirical base, and the public code is a valuable asset. However, the mechanistic justification for CD-KDA's mean-alignment term is theoretically incorrect, and the experimental evidence does not currently isolate whether the observed gain comes from the mean term, the covariance term, or indirect regularization. This needs to be addressed before the central contribution can be fully accepted.

major comments (3)
  1. [Section 3.3, Eq. (2), Eq. (5)] The claim that 'high or low mean values of k features will cause significant bias at e^k level' is not correct. In the Bi-WKV formulation, for each channel, adding a constant δ_c to all keys in a domain multiplies every term in the numerator and denominator of Eq. (2) by e^{δ_c}, which cancels. Thus the attention output wkvt is invariant to per-channel additive shifts of k. Consequently, the mean-alignment term in Eq. (5) cannot directly influence attention. The observed ~2-point gain from CD-KDA in Table 2 must therefore be attributed to the covariance term or to indirect training effects. Please provide an ablation (or a controlled experiment) isolating the mean-only and covariance-only contributions, or revise the stated mechanism accordingly.
  2. [Section 4.2, Table 1] All accuracy numbers are single runs with no standard deviation or significance tests. The claimed improvements over PointDGMamba are 0.81 percentage points on PointDA-10 and 1.39 points on PointDG-3to1; these margins are small enough that they may be within run-to-run variation. Given that the central claim is state-of-the-art performance, please report the mean and standard deviation over at least three seeds, or provide a statistical test, to establish reliability.
  3. [Section 4.3, Tables 2–4] The ablations for AGT-Shift and CD-KDA are conducted only on the PointDA-10 benchmark. The main claims of generalization to unseen domains are also based on PointDG-3to1, and the paper does not show whether the modules contribute similarly on that benchmark. Please report the module ablations on at least one PointDG-3to1 setting, or explicitly justify why PointDA-10 alone is representative.
minor comments (4)
  1. [Section 4.4, Table 4] The text states that alignment on k alone achieves the best performance, but Table 4 reports 75.68 for 'k and v' versus 75.66 for 'Only k'. This discrepancy is small but should be acknowledged or corrected, as it weakens the interpretation that aligning k alone is optimal.
  2. [Section 4.4, paragraph 'Effect of Model Scale'] The sentence 'we design three variants of our PointDGMamba' should read 'PointDGRWKV'.
  3. [References] Several reference entries appear twice (e.g., entries [26] and [77] show duplicate listings). Please clean up the bibliography.
  4. [Figure 3] The illustrative example (k1=-0.3, k2=1.0) shows the effect of absolute differences on e^k, but in a softmax normalization the relevant quantity is relative differences. The figure should be tied more explicitly to the invariance property of Eq. (2) to avoid misleading readers.

Circularity Check

0 steps flagged

No significant circularity: results are held-out benchmark accuracies and CD-KDA is a source-domain regularizer; score reflects only minor non-load-bearing self-citation.

full rationale

The reported accuracies (75.66% on PointDA-10, 81.92% on PointDG-3to1) are measured on held-out target domains after training only on source domains; no target labels or target statistics enter the training objective. CD-KDA (Eq. 5) is a regularizer applied to source-domain key features, not a fitted estimator of the target metric, so the accuracy numbers are not forced by construction. AGT-Shift is defined independently via spatial hashing and weighted aggregation (Eqs. 3-4), and its contribution is evaluated by ablations on held-out accuracy. The paper does rely on the authors' own PointDG-3to1 benchmark and PointDGMamba baseline from prior work [68], but this is a provenance/self-preference issue rather than load-bearing circularity: the benchmark is a public protocol, the comparison is external, and the independent PointDA-10 benchmark corroborates the ranking. One genuine concern is that the paper's mechanistic explanation for the mean-alignment term in CD-KDA is questionable: in Eq. 2, a per-channel additive shift applied to all keys multiplies both numerator and denominator by the same exponential factor and cancels, so mean differences of k do not by themselves shift the attention weights. That is a correctness/interpretability risk, not a circularity, because the method's empirical evaluation does not depend on the validity of that mechanism. Overall, no derivation step reduces the paper's predictions to its inputs; the only deduction from the score is the minor self-citation in benchmarking.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The central empirical result rests on four unproved modeling assumptions: source-domain key alignment transfers to unseen targets, fixed grid hashing captures local geometry, Q-Shift distortion harms DG, and only key alignment matters. The ablations support these for PointDA-10, but no formal guarantees or error bars are provided. No new physical or symbolic entities are introduced, so the invented-entity ledger is empty.

free parameters (4)
  • lambda_2 (CD-KDA loss weight) = 0.3 (default)
    Weight of the alignment loss in Eq. 6; fixed without sensitivity analysis and likely tuned on the benchmarks.
  • lambda (AGT-Shift residual fusion) = not reported
    Balance between original and shifted features in Eq. 4; no value or search range is given.
  • spatial hash grid step size = not reported
    Determines neighborhood partitioning for AGT-Shift in Eq. 3; directly controls local geometry aggregation and is not specified.
  • C' (channel subset size for AGT-Shift) = not reported
    The number of channels perturbed by token shift in Eq. 4 is not stated.
axioms (4)
  • domain assumption Aligning source-domain key mean and covariance transfers to unseen target domains.
    CD-KDA objective in Eq. 5 assumes cross-source alignment improves unseen-target attention; this is not proven, only empirically supported by ablations.
  • domain assumption Fixed-step spatial hashing captures local geometry as well as KNN for arbitrary point clouds.
    AGT-Shift in Eq. 3 assumes grid cells defined by fixed step sizes are valid local neighborhoods without pairwise distances; grid size is not specified.
  • domain assumption RWKV's fixed-direction Q-Shift distorts point cloud spatial structure.
    Motivation in Section 3.1 and Figure 2; qualitative claim not formally quantified.
  • domain assumption Value vectors need not be aligned because they do not enter attention weights directly.
    Justifies aligning only keys in Eq. 5; supported by Table 4 ablation but not by a theoretical argument.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of PointDGRWKV: Generalizing RWKV-like Architecture to Unseen Domains for Point Cloud Classification." pith.science (2026). https://pith.science/paper/RKEPFJMZ

@misc{pith2026250820835,
  author       = {Pith},
  title        = {Pith review of: PointDGRWKV: Generalizing RWKV-like Architecture to Unseen Domains for Point Cloud Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RKEPFJMZ}},
  note         = {Machine review of arXiv:2508.20835}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Domain Generalization (DG) has been recently explored to enhance the generalizability of Point Cloud Classification (PCC) models toward unseen domains. Prior works are based on convolutional networks, Transformer or Mamba architectures, either suffering from limited receptive fields or high computational cost, or insufficient long-range dependency modeling. RWKV, as an emerging architecture, possesses superior linear complexity, global receptive fields, and long-range dependency. In this paper, we present the first work that studies the generalizability of RWKV models in DG PCC. We find that directly applying RWKV to DG PCC encounters two significant challenges: RWKV's fixed direction token shift methods, like Q-Shift, introduce spatial distortions when applied to unstructured point clouds, weakening local geometric modeling and reducing robustness. In addition, the Bi-WKV attention in RWKV amplifies slight cross-domain differences in key distributions through exponential weighting, leading to attention shifts and degraded generalization. To this end, we propose PointDGRWKV, the first RWKV-based framework tailored for DG PCC. It introduces two key modules to enhance spatial modeling and cross-domain robustness, while maintaining RWKV's linear efficiency. In particular, we present Adaptive Geometric Token Shift to model local neighborhood structures to improve geometric context awareness. In addition, Cross-Domain key feature Distribution Alignment is designed to mitigate attention drift by aligning key feature distributions across domains. Extensive experiments on multiple benchmarks demonstrate that PointDGRWKV achieves state-of-the-art performance on DG PCC.

Figures

Figures reproduced from arXiv: 2508.20835 by Haijia Sun, Hao Yang, Lizhuang Ma, Qianyu Zhou, Shuicheng Yan, Xiangtai Li, Xuequan Lu.

Figure 1
Figure 1. Figure 1: Accuracy-speed tradeoff in DG PCC. (Left) Overall ac [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the exponential function’s amplification [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The architecture of PointDGRWKV consists of two key components: (a) [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: T-SNE visualization results of target domain feature [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗

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

85 extracted references · 72 canonical work pages · 2 internal anchors

  1. [1]

    Self- supervised learning for domain adaptation on point clouds

    Idan Achituve, Haggai Maron, and Gal Chechik. Self- supervised learning for domain adaptation on point clouds. In Proceedings of Winter Conference on Applications of Computer Vision, pages 123–133, 2021. 6

    2021

  2. [2]

    3dmfv: Three-dimensional point cloud classification in real-time using convolutional neural networks

    Yizhak Ben-Shabat, Michael Lindenbaum, and Anath Fis- cher. 3dmfv: Three-dimensional point cloud classification in real-time using convolutional neural networks. IEEE Robotics and Automation Letters, 3(4):3145–3152, 2018. 1

    2018

  3. [3]

    A survey of augmented reality

    Mark Billinghurst, Adrian Clark, Gun Lee, et al. A survey of augmented reality. Foundations and Trends® in Human– Computer Interaction, 8(2-3):73–272, 2015. 1

    2015

  4. [4]

    nuscenes: A multi- modal dataset for autonomous driving

    Holger Caesar, Varun Bankiti, Alex H Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Gi- ancarlo Baldan, and Oscar Beijbom. nuscenes: A multi- modal dataset for autonomous driving. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11621–11631, 2020. 1

    2020

  5. [5]

    Zig-rir: Zigzag rwkv-in-rwkv for efficient medical image segmentation

    Tianxiang Chen, Xudong Zhou, Zhentao Tan, Yue Wu, Ziyang Wang, Zi Ye, Tao Gong, Qi Chu, Nenghai Yu, and Le Lu. Zig-rir: Zigzag rwkv-in-rwkv for efficient medical image segmentation. IEEE Transactions on Medical Imag- ing, 2025. 2, 3

    2025

  6. [6]

    Scannet: Richly-annotated 3d reconstructions of indoor scenes

    Angela Dai, Angel X Chang, Manolis Savva, Maciej Hal- ber, Thomas Funkhouser, and Matthias Nießner. Scannet: Richly-annotated 3d reconstructions of indoor scenes. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 5828–5839, 2017. 6

    work page 2017

  7. [7]

    Stylerwkv: High-quality and high-efficiency style transfer with rwkv- like architecture

    Miaomiao Dai, Qianyu Zhou, and Lizhuang Ma. Stylerwkv: High-quality and high-efficiency style transfer with rwkv- like architecture. In IEEE International Conference on Mul- timedia and Expo, pages 01–06, 2025. 2, 3

    work page 2025

  8. [8]

    VG4D: Vision-Language Model Goes 4D Video Recognition

    Zhichao Deng, Xiangtai Li, Xia Li, Yunhai Tong, Shen Zhao, and Mengyuan Liu. Vg4d: Vision-language model goes 4d video recognition. arXiv preprint arXiv:2404.11605, 2024. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  9. [9]

    Vision-rwkv: Efficient and scalable vi- sual perception with rwkv-like architectures

    Yuchen Duan, Weiyun Wang, Zhe Chen, Xizhou Zhu, Lewei Lu, Tong Lu, Yu Qiao, Hongsheng Li, Jifeng Dai, and Wenhai Wang. Vision-rwkv: Efficient and scalable vi- sual perception with rwkv-like architectures. arXiv preprint arXiv:2403.02308, 2024. 2, 3, 5, 6

    Pith/arXiv arXiv 2024

  10. [10]

    Self-supervised global- local structure modeling for point cloud domain adapta- tion with reliable voted pseudo labels

    Hehe Fan, Xiaojun Chang, Wanyue Zhang, Yi Cheng, Ying Sun, and Mohan Kankanhalli. Self-supervised global- local structure modeling for point cloud domain adapta- tion with reliable voted pseudo labels. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6377–6386, 2022. 2

    work page 2022

  11. [11]

    Explore in-context learning for 3d point cloud understanding

    Zhongbin Fang, Xiangtai Li, Xia Li, Joachim M Buhmann, Chen Change Loy, and Mengyuan Liu. Explore in-context learning for 3d point cloud understanding. Advances in Neu- ral Information Processing Systems, 36, 2024. 2

    work page 2024

  12. [12]

    Dmt: Dynamic mutual training for semi-supervised learning

    Zhengyang Feng, Qianyu Zhou, Qiqi Gu, Xin Tan, Guan- gliang Cheng, Xuequan Lu, Jianping Shi, and Lizhuang Ma. Dmt: Dynamic mutual training for semi-supervised learning. Patter Recognition, 130:108777, 2022. 2

    work page 2022

  13. [13]

    3d-future: 3d fur- niture shape with texture

    Huan Fu, Rongfei Jia, Lin Gao, Mingming Gong, Binqiang Zhao, Steve Maybank, and Dacheng Tao. 3d-future: 3d fur- niture shape with texture. International Journal of Computer Vision, 129:3313–3337, 2021. 6

    work page 2021

  14. [14]

    Unsupervised domain adaptation by backpropagation

    Yaroslav Ganin and Victor Lempitsky. Unsupervised domain adaptation by backpropagation. In International Conference on Machine Learning, pages 1180–1189, 2015. 2

    work page 2015

  15. [15]

    Pit: Position-invariant transform for cross-fov domain adaptation

    Qiqi Gu, Qianyu Zhou, Minghao Xu, Zhengyang Feng, Guangliang Cheng, Xuequan Lu, Jianping Shi, and Lizhuang Ma. Pit: Position-invariant transform for cross-fov domain adaptation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 8761–8770, 2021. 2

    work page 2021

  16. [16]

    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. 2, 6, 8

    work page 2021

  17. [17]

    Label-free re- gional consistency for image-to-image translation

    Shaohua Guo, Qianyu Zhou, Ye Zhou, Qiqi Gu, Junshu Tang, Zhengyang Feng, and Lizhuang Ma. Label-free re- gional consistency for image-to-image translation. In IEEE International Conference on Multimedia and Expo, pages 1– 6, 2021. 2

    work page 2021

  18. [18]

    Semantic3d

    Timo Hackel, Nikolay Savinov, Lubor Ladicky, Jan D Weg- ner, Konrad Schindler, and Marc Pollefeys. Semantic3d. net: A new large-scale point cloud classification benchmark. arXiv preprint arXiv:1704.03847, 2017. 1

    Pith/arXiv arXiv 2017

  19. [19]

    A survey on vision transformer

    Kai Han, Yunhe Wang, Hanting Chen, Xinghao Chen, Jianyuan Guo, Zhenhua Liu, Yehui Tang, An Xiao, Chunjing Xu, Yixing Xu, et al. A survey on vision transformer. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(1):87–110, 2022. 2

    work page 2022

  20. [20]

    Pointrwkv: Efficient rwkv-like model for hierarchical point cloud learn- ing

    Qingdong He, Jiangning Zhang, Jinlong Peng, Haoyang He, Xiangtai Li, Yabiao Wang, and Chengjie Wang. Pointrwkv: Efficient rwkv-like model for hierarchical point cloud learn- ing. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 3410–3418, 2025. 3, 6

    work page 2025

  21. [21]

    Metasets: Meta-learning on point sets for generalizable representations

    Chao Huang, Zhangjie Cao, Yunbo Wang, Jianmin Wang, and Mingsheng Long. Metasets: Meta-learning on point sets for generalizable representations. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion, pages 8863–8872, 2021. 3, 6

    work page 2021

  22. [22]

    Gen- eration for unsupervised domain adaptation: A gan-based approach for object classification with 3d point cloud data

    Junxuan Huang, Junsong Yuan, and Chunming Qiao. Gen- eration for unsupervised domain adaptation: A gan-based approach for object classification with 3d point cloud data. In IEEE International Conference on Acoustics, Speech and Signal Processing, pages 3753–3757, 2022. 2

    work page 2022

  23. [23]

    Sug: Single-dataset unified generalization for 3d point cloud classification

    Siyuan Huang, Bo Zhang, Botian Shi, Hongsheng Li, Yikang Li, and Peng Gao. Sug: Single-dataset unified generalization for 3d point cloud classification. In Proceedings of the ACM International Conference on Multimedia, pages 8644–8652,

    work page

  24. [24]

    Dg- pic: Domain generalized point-in-context learning for point cloud understanding

    Jincen Jiang, Qianyu Zhou, Yuhang Li, Xuequan Lu, Meili Wang, Lizhuang Ma, Jian Chang, and Jian Jun Zhang. Dg- pic: Domain generalized point-in-context learning for point cloud understanding. In European Conference on Computer Vision. Springer, 2024. 3

    work page 2024

  25. [25]

    Pcotta: Continual test-time adaptation for multi-task point cloud understanding

    Jincen Jiang, Qianyu Zhou, Yuhang Li, Xinkui Zhao, Meili Wang, Lizhuang Ma, Jian Chang, Jian Zhang, Xuequan Lu, et al. Pcotta: Continual test-time adaptation for multi-task point cloud understanding. Advances in Neural Information Processing Systems, 37:96229–96253, 2024. 2

    work page 2024

  26. [26]

    Synergiz- ing contrastive learning and optimal transport for 3d point cloud domain adaptation

    Siddharth Katageri, Arkadipta De, Chaitanya Devaguptapu, VSSV Prasad, Charu Sharma, and Manohar Kaul. Synergiz- ing contrastive learning and optimal transport for 3d point cloud domain adaptation. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages 2942–2951, 2024. 2

    work page 2024

  27. [27]

    Single domain generalization for lidar seman- tic segmentation

    Hyeonseong Kim, Yoonsu Kang, Changgyoon Oh, and Kuk- Jin Yoon. Single domain generalization for lidar seman- tic segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 17587– 17598, 2023. 2, 3

    work page 2023

  28. [28]

    3d- vfield: Adversarial augmentation of point clouds for domain generalization in 3d object detection

    Alexander Lehner, Stefano Gasperini, Alvaro Marcos- Ramiro, Michael Schmidt, Mohammad-Ali Nikouei Mahani, Nassir Navab, Benjamin Busam, and Federico Tombari. 3d- vfield: Adversarial augmentation of point clouds for domain generalization in 3d object detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion, pages 17295–...

    work page 2022

  29. [29]

    Pointaugment: an auto-augmentation framework for point cloud classification

    Ruihui Li, Xianzhi Li, Pheng-Ann Heng, and Chi-Wing Fu. Pointaugment: an auto-augmentation framework for point cloud classification. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 6378–6387, 2020. 1

    work page 2020

  30. [30]

    Pointcnn: Convolution on x-transformed points

    Yangyan Li, Rui Bu, Mingchao Sun, Wei Wu, Xinhan Di, and Baoquan Chen. Pointcnn: Convolution on x-transformed points. Advances in Neural Information Processing Systems, 31, 2018. 1, 2

    work page 2018

  31. [31]

    Point- mamba: A simple state space model for point cloud analysis

    Dingkang Liang, Xin Zhou, Xinyu Wang, Xingkui Zhu, Wei Xu, Zhikang Zou, Xiaoqing Ye, and Xiang Bai. Point- mamba: A simple state space model for point cloud analysis. arXiv preprint arXiv:2402.10739, 2024. 2

    Pith/arXiv arXiv 2024

  32. [32]

    Point cloud domain adaptation via masked local 3d structure prediction

    Hanxue Liang, Hehe Fan, Zhiwen Fan, Yi Wang, Tianlong Chen, Yu Cheng, and Zhangyang Wang. Point cloud domain adaptation via masked local 3d structure prediction. InEuro- pean Conference on Computer Vision, pages 156–172, 2022. 2

    work page 2022

  33. [33]

    Cloudmix: Dual mixup consistency for unpaired point cloud completion

    Fengqi Liu, Jingyu Gong, Qianyu Zhou, Xuequan Lu, Ran Yi, Yuan Xie, and Lizhuang Ma. Cloudmix: Dual mixup consistency for unpaired point cloud completion. IEEE Transactions on Visualization and Computer Graphics , 31 (4):2182–2195, 2024. 2, 6

    work page 2024

  34. [34]

    Dgmamba: Domain generalization via generalized state space model

    Shaocong Long, Qianyu Zhou, Xiangtai Li, Xuequan Lu, Chenhao Ying, Yuan Luo, Lizhuang Ma, and Shuicheng Yan. Dgmamba: Domain generalization via generalized state space model. In Proceedings of the 30th ACM International Conference on Multimedia), pages 3607–3616, 2024. 3

    work page 2024

  35. [35]

    Rethinking domain generalization: Dis- criminability and generalizability

    Shaocong Long, Qianyu Zhou, Chenhao Ying, Lizhuang Ma, and Yuan Luo. Rethinking domain generalization: Dis- criminability and generalizability. IEEE Transactions on Circuits and Systems for Video Technology , 34(11):11783– 11797, 2024

    work page 2024

  36. [36]

    Domain Generalization via Discrete Codebook Learning

    Shaocong Long, Qianyu Zhou, Xikun Jiang, Chenhao Ying, Lizhuang Ma, and Yuan Luo. Domain generalization via dis- crete codebook learning. arXiv preprint arXiv:2504.06572, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  37. [37]

    Diverse target and contribution scheduling for domain generalization

    Shaocong Long, Qianyu Zhou, Chenhao Ying, Lizhuang Ma, and Yuan Luo. Diverse target and contribution scheduling for domain generalization. IEEE Transactions on Image Pro- cessing, 34:4242–4257, 2025. 3

    work page 2025

  38. [38]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. 5

    Pith/arXiv arXiv 2017

  39. [39]

    Rwkv: Reinventing rnns for the transformer era

    Bo Peng, Eric Alcaide, Quentin Anthony, Alon Albalak, Samuel Arcadinho, Stella Biderman, Huanqi Cao, Xin Cheng, Michael Chung, Matteo Grella, et al. Rwkv: Reinventing rnns for the transformer era. arXiv preprint arXiv:2305.13048, 2023. 2

    Pith/arXiv arXiv 2023

  40. [40]

    Dgcnn: A convolutional neural network over large-scale labeled graphs

    Anh Viet Phan, Minh Le Nguyen, Yen Lam Hoang Nguyen, and Lam Thu Bui. Dgcnn: A convolutional neural network over large-scale labeled graphs. Neural Networks, 108:533– 543, 2018. 1

    work page 2018

  41. [41]

    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. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 652–660,

    work page

  42. [42]

    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. 1, 2

    work page 2017

  43. [43]

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

    Guocheng Qian, Yuchen Li, Houwen Peng, Jinjie Mai, Hasan Hammoud, Mohamed Elhoseiny, and Bernard Ghanem. Pointnext: Revisiting pointnet++ with improved training and scaling strategies. Advances in Neural Informa- tion Processing Systems, 35:23192–23204, 2022. 6

    work page 2022

  44. [44]

    Pointdan: A multi-scale 3d domain adaption net- work for point cloud representation

    Can Qin, Haoxuan You, Lichen Wang, C-C Jay Kuo, and Yun Fu. Pointdan: A multi-scale 3d domain adaption net- work for point cloud representation. Advances in Neural In- formation Processing Systems, 32, 2019. 2, 6

    work page 2019

  45. [45]

    Dense-resolution network for point cloud classification and segmentation

    Shi Qiu, Saeed Anwar, and Nick Barnes. Dense-resolution network for point cloud classification and segmentation. In Proceedings of the IEEE/CVF Winter Conference on Appli- cations of Computer Vision, pages 3813–3822, 2021. 1

    work page 2021

  46. [46]

    Geometric back- projection network for point cloud classification

    Shi Qiu, Saeed Anwar, and Nick Barnes. Geometric back- projection network for point cloud classification. IEEE Transactions on Multimedia, 24:1943–1955, 2021. 6, 8

    work page 1943

  47. [47]

    Benchmarking and analyzing point cloud classification under corruptions

    Jiawei Ren, Liang Pan, and Ziwei Liu. Benchmarking and analyzing point cloud classification under corruptions. In In- ternational Conference on Machine Learning, pages 18559– 18575, 2022. 1

    work page 2022

  48. [48]

    Domain adaptation on point clouds via geometry-aware implicits

    Yuefan Shen, Yanchao Yang, Mi Yan, He Wang, Youyi Zheng, and Leonidas J Guibas. Domain adaptation on point clouds via geometry-aware implicits. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7223–7232, 2022. 2

    work page 2022

  49. [49]

    Ba-sam: Scalable bias-mode at- tention mask for segment anything model

    Yiran Song, Qianyu Zhou, Xiangtai Li, Deng-Ping Fan, Xue- quan Lu, and Lizhuang Ma. Ba-sam: Scalable bias-mode at- tention mask for segment anything model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3162–3173, 2024. 3

    work page 2024

  50. [50]

    Su-sam: A simple unified framework for adapting segment anything model in underperformed scenes

    Yiran Song, Qianyu Zhou, Xuequan Lu, Zhiwen Shao, and Lizhuang Ma. Su-sam: A simple unified framework for adapting segment anything model in underperformed scenes. arXiv preprint arXiv:2401.17803, 2024. 3

    Pith/arXiv arXiv 2024

  51. [51]

    X-3d: Explicit 3d structure modeling for point cloud recog- nition

    Shuofeng Sun, Yongming Rao, Jiwen Lu, and Haibin Yan. X-3d: Explicit 3d structure modeling for point cloud recog- nition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 5074–5083, 2024. 6

    work page 2024

  52. [52]

    Soft robot perception using em- bedded soft sensors and recurrent neural networks

    Thomas George Thuruthel, Benjamin Shih, Cecilia Laschi, and Michael Thomas Tolley. Soft robot perception using em- bedded soft sensors and recurrent neural networks. Science Robotics, 4(26):eaav1488, 2019. 1

    work page 2019

  53. [53]

    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 1588– 1597, 2019. 1

    work page 2019

  54. [54]

    Cross-dataset point cloud recognition using deep-shallow domain adaptation net- work

    Feiyu Wang, Wen Li, and Dong Xu. Cross-dataset point cloud recognition using deep-shallow domain adaptation net- work. IEEE Transactions on Image Processing , 30:7364– 7377, 2021. 2

    work page 2021

  55. [55]

    Generalizing to unseen domains: A survey on domain generalization

    Jindong Wang, Cuiling Lan, Chang Liu, Yidong Ouyang, Tao Qin, Wang Lu, Yiqiang Chen, Wenjun Zeng, and S Yu Philip. Generalizing to unseen domains: A survey on domain generalization. IEEE Transactions on Knowledge and Data Engineering, 35(8):8052–8072, 2022. 3

    work page 2022

  56. [56]

    Deep visual domain adapta- tion: A survey

    Mei Wang and Weihong Deng. Deep visual domain adapta- tion: A survey. Neurocomputing, 312:135–153, 2018. 2

    work page 2018

  57. [57]

    Unsupervised domain adap- tation for cross-scene multispectral point cloud classifica- tion

    Qingwang Wang, Mingye Wang, Jiangbo Huang, Tianzhu Liu, Tao Shen, and Yanfeng Gu. Unsupervised domain adap- tation for cross-scene multispectral point cloud classifica- tion. IEEE Transactions on Geoscience and Remote Sensing,

    work page

  58. [58]

    Tf-fas: Twofold-element fine-grained semantic guidance for gen- eralizable face anti-spoofing

    Xudong Wang, Ke-Yue Zhang, Taiping Yao, Qianyu Zhou, Shouhong Ding, Pingyang Dai, and Rongrong Ji. Tf-fas: Twofold-element fine-grained semantic guidance for gen- eralizable face anti-spoofing. In European Conference on Computer Vision. Springer, 2024. 3

    work page 2024

  59. [59]

    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, 38(5):1–12, 2019. 1, 2

    work page 2019

  60. [60]

    Learning generaliz- able part-based feature representation for 3d point clouds

    Xin Wei, Xiang Gu, and Jian Sun. Learning generaliz- able part-based feature representation for 3d point clouds. Advances in Neural Information Processing Systems , 35: 29305–29318, 2022. 3, 6

    work page 2022

  61. [61]

    Squeezesegv2: Improved model structure and unsupervised domain adaptation for road-object segmenta- tion from a lidar point cloud

    Bichen Wu, Xuanyu Zhou, Sicheng Zhao, Xiangyu Yue, and Kurt Keutzer. Squeezesegv2: Improved model structure and unsupervised domain adaptation for road-object segmenta- tion from a lidar point cloud. In 2019 International Confer- ence on Robotics and Automation , pages 4376–4382, 2019. 2

    work page 2019

  62. [62]

    3d shapenets: A deep representation for volumetric shapes

    Zhirong Wu, Shuran Song, Aditya Khosla, Fisher Yu, Lin- guang 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, pages 1912–1920, 2015. 6

    work page 1912

  63. [63]

    3d semantic segmentation in the wild: Learning generalized models for adverse-condition point clouds

    Aoran Xiao, Jiaxing Huang, Weihao Xuan, Ruijie Ren, Kangcheng Liu, Dayan Guan, Abdulmotaleb El Saddik, Shi- jian Lu, and Eric P Xing. 3d semantic segmentation in the wild: Learning generalized models for adverse-condition point clouds. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 9382– 9392, 2023. 2, 3

    work page 2023

  64. [64]

    Learning cross- domain features for domain generalization on point clouds

    Hang Xiao, Ming Cheng, and Liangwei Shi. Learning cross- domain features for domain generalization on point clouds. In Chinese Conference on Pattern Recognition and Com- puter Vision, pages 68–81, 2022. 3

    work page 2022

  65. [65]

    Semi-supervised 3d object detection via adaptive pseudo-labeling

    Hongyi Xu, Fengqi Liu, Qianyu Zhou, Jinkun Hao, Zhijie Cao, Zhengyang Feng, and Lizhuang Ma. Semi-supervised 3d object detection via adaptive pseudo-labeling. In IEEE International Conference on Image Processing, pages 3183– 3187, 2021. 2

    work page 2021

  66. [66]

    Push-and-pull: A general training framework with differen- tial augmentor for domain generalized point cloud classifica- tion

    Jiahao Xu, Xinzhu Ma, Lin Zhang, Bo Zhang, and Tao Chen. Push-and-pull: A general training framework with differen- tial augmentor for domain generalized point cloud classifica- tion. IEEE Transactions on Circuits and Systems for Video Technology, 2024. 3

    work page 2024

  67. [67]

    Geometry sharing network for 3d point cloud classification and segmentation

    Mingye Xu, Zhipeng Zhou, and Yu Qiao. Geometry sharing network for 3d point cloud classification and segmentation. In Proceedings of the AAAI Conference on Artificial Intelli- gence, pages 12500–12507, 2020. 1

    work page 2020

  68. [68]

    Point- dgmamba: Domain generalization of point cloud classifica- tion via generalized state space model

    Hao Yang, Qianyu Zhou, Haijia Sun, Xiangtai Li, Fengqi Liu, Xuequan Lu, Lizhuang Ma, and Shuicheng Yan. Point- dgmamba: Domain generalization of point cloud classifica- tion via generalized state space model. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 9193– 9201, 2025. 2, 3, 6, 8

    work page 2025

  69. [69]

    Video rwkv: Video action recognition based rwkv

    Zhuowen Yin, Chengru Li, and Xingbo Dong. Video rwkv: Video action recognition based rwkv. arXiv preprint arXiv:2411.05636, 2024. 3

    Pith/arXiv arXiv 2024

  70. [70]

    Mamba or rwkv: Exploring high-quality and high-efficiency segment anything model

    Haobo Yuan, Xiangtai Li, Lu Qi, Tao Zhang, Ming-Hsuan Yang, Shuicheng Yan, and Chen Change Loy. Mamba or rwkv: Exploring high-quality and high-efficiency segment anything model. arXiv preprint arXiv:2406.19369, 2024. 3

    Pith/arXiv arXiv 2024

  71. [71]

    Deep learning-based 3d point cloud classification: A systematic survey and out- look

    Huang Zhang, Changshuo Wang, Shengwei Tian, Baoli Lu, Liping Zhang, Xin Ning, and Xiao Bai. Deep learning-based 3d point cloud classification: A systematic survey and out- look. Displays, 79:102456, 2023. 1

    work page 2023

  72. [72]

    Pointhop: An explainable machine learning method for point cloud classification

    Min Zhang, Haoxuan You, Pranav Kadam, Shan Liu, and C-C Jay Kuo. Pointhop: An explainable machine learning method for point cloud classification. IEEE Transactions on Multimedia, 22(7):1744–1755, 2020. 1

    work page 2020

  73. [73]

    Point cloud mamba: Point cloud learning via state space model

    Tao Zhang, Haobo Yuan, Lu Qi, Jiangning Zhang, Qianyu Zhou, Shunping Ji, Shuicheng Yan, and Xiangtai Li. Point cloud mamba: Point cloud learning via state space model. In Proceedings of the AAAI Conference on Artificial Intelli- gence, pages 10121–10130, 2025. 2, 6

    work page 2025

  74. [74]

    A graph-cnn for 3d point cloud classification

    Yingxue Zhang and Michael Rabbat. A graph-cnn for 3d point cloud classification. In IEEE International Conference on Acoustics, Speech and Signal Processing , pages 6279– 6283, 2018. 1

    work page 2018

  75. [75]

    Point transformer

    Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip HS Torr, and Vladlen Koltun. Point transformer. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 16259–16268, 2021. 2

    work page 2021

  76. [76]

    Domain generalization: A survey

    Kaiyang Zhou, Ziwei Liu, Yu Qiao, Tao Xiang, and Chen Change Loy. Domain generalization: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(4):4396–4415, 2022. 3

    work page 2022

  77. [77]

    Uncertainty-aware consistency regularization for cross- domain semantic segmentation

    Qianyu Zhou, Zhengyang Feng, Qiqi Gu, Guangliang Cheng, Xuequan Lu, Jianping Shi, and Lizhuang Ma. Uncertainty-aware consistency regularization for cross- domain semantic segmentation. Computer Vision and Image Understanding, 221:103448, 2022. 2

    work page 2022

  78. [78]

    Adaptive mixture of ex- perts learning for generalizable face anti-spoofing

    Qianyu Zhou, Ke-Yue Zhang, Taiping Yao, Ran Yi, Shouhong Ding, and Lizhuang Ma. Adaptive mixture of ex- perts learning for generalizable face anti-spoofing. In Pro- ceedings of the 30th ACM International Conference on Mul- timedia, pages 6009–6018, 2022. 3

    work page 2022

  79. [79]

    Generative do- main adaptation for face anti-spoofing

    Qianyu Zhou, Ke-Yue Zhang, Taiping Yao, Ran Yi, Kekai Sheng, Shouhong Ding, and Lizhuang Ma. Generative do- main adaptation for face anti-spoofing. In European Con- ference on Computer Vision, pages 335–356. Springer, 2022. 2

    work page 2022

  80. [80]

    Domain adaptive semantic segmentation via regional contrastive consistency regularization

    Qianyu Zhou, Chuyun Zhuang, Ran Yi, Xuequan Lu, and Lizhuang Ma. Domain adaptive semantic segmentation via regional contrastive consistency regularization. In IEEE In- ternational Conference on Multimedia and Expo, pages 01– 06, 2022

    work page 2022

Showing first 80 references.

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.