pith. sign in

arxiv: 2606.31570 · v1 · pith:FTAWTWTQnew · submitted 2026-06-30 · 💻 cs.CV · cs.AI

Mitigating Positional Leakage in 3D Masked Autoencoders for Robust Representation Learning

Xu Yan , Huiqun Wang , Chen Wang , Lei Ren , Di Huang This is my paper

Pith reviewed 2026-07-01 05:45 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords 3D point cloudsmasked autoencodersself-supervised learningpositional leakagerepresentation learningpoint cloud reconstructionsemantic features
0
0 comments X

The pith

Decoder over-reliance on position weakens semantic features in 3D masked autoencoders; MPL-MAE counters this with recalibrated embeddings and gated interfaces.

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

In 3D masked autoencoders the decoder reconstructs point coordinates directly and therefore tends to lean on positional signals at the expense of semantic content from the encoder. The paper introduces MPL-MAE to reduce that leakage by adding a recalibrated positional embedding module that keeps geometric topology while suppressing raw metric coordinates, plus a gated positional interface that adjusts how much position information reaches the decoder at each step. These changes aim to create a more even exchange between spatial priors and encoder-derived semantics. If the approach holds, the resulting encoder features should support stronger transfer to downstream point-cloud tasks without extra labeled data.

Core claim

The decoder in existing 3D MAE frameworks tends to over-rely on positional information, which weakens semantic representation learning and leads to suboptimal feature quality. MPL-MAE mitigates this by introducing a recalibrated positional embedding module that suppresses metric-dominant coordinate signals while preserving geometric topology, together with a gated positional interface module that dynamically regulates positional injection during reconstruction; the two modules together promote a more balanced interaction between spatial priors and semantic features.

What carries the argument

Recalibrated positional embedding module and gated positional interface module, which together suppress over-reliance on coordinates while keeping geometric structure during decoder reconstruction.

If this is right

  • The encoder produces features that carry more semantic information usable across classification, segmentation, and detection tasks.
  • Reconstruction quality remains high while positional signals are dynamically limited rather than removed outright.
  • The same encoder can be frozen and transferred to new 3D tasks with less performance drop than in prior MAE setups.
  • Overall training yields representations that interact more evenly with both spatial and semantic signals.

Where Pith is reading between the lines

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

  • The gated interface could be tested on other coordinate-reconstruction objectives, such as surface reconstruction or scene completion, to check whether leakage is a broader issue.
  • If the recalibration preserves topology, the method might also reduce sensitivity to coordinate noise or scaling differences in real-world scans.
  • Similar positional over-reliance may appear in non-MAE 3D self-supervised methods that decode coordinates, suggesting a general design pattern rather than an MAE-specific fix.

Load-bearing premise

Suppressing metric-dominant coordinate signals while preserving geometric topology will shift the decoder toward semantic encoder features rather than simply replacing one form of leakage with another.

What would settle it

Training both a standard 3D MAE and MPL-MAE on identical data and benchmarks, then measuring downstream task accuracy; if MPL-MAE shows no gain or a drop, the claim that the modules improve semantic representation quality would be contradicted.

Figures

Figures reproduced from arXiv: 2606.31570 by Chen Wang, Di Huang, Huiqun Wang, Lei Ren, Xu Yan.

Figure 1
Figure 1. Figure 1: (a) Pipeline of MAE for 3D point cloud representation learning. (b) Empirical analysis of representative 3D MAE methods. Left: gradient ratio between positional embeddings (π) and encoder outputs (Z) during training. Right: sensitivity analysis of reconstruction error under perturbations applied to positional embeddings. Among various self-supervised paradigms, masked autoencoders (MAE), orig￾inally propos… view at source ↗
Figure 2
Figure 2. Figure 2: Pipeline of MPL-MAE. The RPE module generates recalibrated positional embeddings for the decoder. A GPI module is inserted before each Transformer layer in the decoder to regulate positional injection, promoting balanced optimization and robust geometric feature learning. 3.4 Framework By integrating the Recalibrated Positional Embedding (RPE) module and the Gated Positional Interface (GPI), we present MPL… view at source ↗
read the original abstract

Masked autoencoding has emerged as a prominent paradigm for self-supervised learning on 3D point clouds, achieving competitive performance across downstream tasks. Unlike its 2D counterpart, 3D masked autoencoding directly reconstructs spatial coordinates, making it inherently susceptible to positional leakage. In this work, we identify that the decoder in existing 3D MAE frameworks tends to over-rely on positional information, which weakens semantic representation learning and leads to suboptimal feature quality. To address this issue, we propose MPL-MAE, a masked point learning framework that mitigates positional over-reliance while enhancing the utilization of encoder features. Specifically, we introduce a recalibrated positional embedding module that suppresses metric-dominant coordinate signals while preserving geometric topology, together with a gated positional interface module that dynamically regulates positional injection during reconstruction. These designs promote a more balanced interaction between spatial priors and semantic features, yielding robust and informative representations. Extensive experiments across downstream tasks demonstrate that MPL-MAE consistently achieves competitive performance, validating its effectiveness. Code is available at https://github.com/yanx57/MPL-MAE.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The paper identifies that 3D masked autoencoders are susceptible to positional leakage because they reconstruct spatial coordinates directly, causing the decoder to over-rely on positional information at the expense of semantic features. It proposes MPL-MAE, which introduces a recalibrated positional embedding module (suppressing metric-dominant coordinate signals while preserving geometric topology) and a gated positional interface module (dynamically regulating positional injection). These are claimed to yield a more balanced interaction between spatial priors and semantic features, producing robust representations that achieve competitive performance on downstream tasks. Code is released at https://github.com/yanx57/MPL-MAE.

Significance. If the proposed modules demonstrably strengthen semantic encoder features rather than merely shifting the decoder's coordinate-reconstruction pathway, the work could improve representation quality in 3D self-supervised learning. The public release of code supports reproducibility and is a clear strength.

major comments (1)
  1. [Abstract] Abstract: the claim that the recalibrated embedding and gated interface 'promote a more balanced interaction between spatial priors and semantic features' lacks any supporting quantitative evidence, ablation results, or metrics (such as semantic vs. positional reconstruction error, feature similarity scores, or decoder attention maps). Because the reconstruction target remains coordinates, it is unclear whether the modules mitigate over-reliance or simply reparameterize it under a different gating scheme.
minor comments (1)
  1. [Abstract] Abstract: the statement 'extensive experiments across downstream tasks demonstrate that MPL-MAE consistently achieves competitive performance' provides no specific datasets, baselines, or numerical improvements, making it impossible to evaluate the strength of the empirical support from the given text.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the major comment below and indicate planned revisions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the recalibrated embedding and gated interface 'promote a more balanced interaction between spatial priors and semantic features' lacks any supporting quantitative evidence, ablation results, or metrics (such as semantic vs. positional reconstruction error, feature similarity scores, or decoder attention maps). Because the reconstruction target remains coordinates, it is unclear whether the modules mitigate over-reliance or simply reparameterize it under a different gating scheme.

    Authors: We agree that the abstract claim would be strengthened by direct quantitative support. The manuscript currently demonstrates effectiveness via competitive downstream task results, which serve as indirect evidence of improved semantic encoder features. To address the specific request, we will revise the manuscript to include additional metrics (e.g., feature similarity to semantic labels and decoder attention visualizations) and clarify how the modules reduce positional over-reliance in the encoder rather than simply altering the decoder pathway. These additions will appear in the experiments and method sections. revision: yes

Circularity Check

0 steps flagged

No circularity: proposal introduces new modules without reducing claims to fitted inputs or self-citations

full rationale

The paper's core contribution is a descriptive proposal of two new modules (recalibrated positional embedding and gated positional interface) to address positional leakage in 3D MAE. No equations, derivations, or parameter-fitting steps are presented in the provided text that would allow any claimed prediction or result to reduce by construction to the inputs. The abstract and description contain no self-citations used as load-bearing uniqueness theorems, no ansatz smuggling, and no renaming of known results as novel derivations. The reconstruction target remains coordinates, but the argument does not claim a mathematical equivalence or statistical forcing that collapses the method onto its own data fits. This is a standard engineering proposal whose validity rests on external experiments rather than internal definitional closure.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract only; no mathematical formulation, parameters, or axioms are described.

pith-pipeline@v0.9.1-grok · 5727 in / 1049 out tokens · 24836 ms · 2026-07-01T05:45:31.111005+00:00 · methodology

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

46 extracted references · 4 canonical work pages · 3 internal anchors

  1. [1]

    In: Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition

    Afham, M., Dissanayake, I., Dissanayake, D., Dharmasiri, A., Thilakarathna, K., Rodrigo, R.: Crosspoint: Self-supervised cross-modal contrastive learning for 3d point cloud understanding. In: Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition. pp. 9902–9912 (2022)

    2022

  2. [2]

    In: IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Armeni, I., Sener, O., Zamir, A.R., Jiang, H., Brilakis, I.K., Fischer, M., Savarese, S.: 3d semantic parsing of large-scale indoor spaces. In: IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 1534–1543 (2016)

    2016

  3. [3]

    ShapeNet: An Information-Rich 3D Model Repository

    Chang, A.X., Funkhouser, T., Guibas, L., Hanrahan, P., Huang, Q., Li, Z., Savarese, S., Savva, M., Song, S., Su, H., Xiao, J., Yi, L., Yu, F.: Shapenet: An information-rich 3d model repository. arXiv preprint arXiv:1512.03012 (2015)

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  4. [4]

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

    Chen, A., Zhang, K., Zhang, R., Wang, Z., Lu, Y., Guo, Y., Zhang, S.: PiMAE: Point cloud and image interactive masked autoencoders for 3D object detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 5291–5301 (2023)

    2023

  5. [5]

    In: Advances in Neural In- formation Processing Systems (2023)

    Chen, G., Wang, M., Yang, Y., Yu, K., Yuan, L., Yue, Y.: Pointgpt: Auto- regressively generative pre-training from point clouds. In: Advances in Neural In- formation Processing Systems (2023)

    2023

  6. [6]

    Chhipa, P.C., Upadhyay, R., Saini, R., Lindqvist, L., Nordenskjold, R., Uchida, S., Liwicki, M.: Depth contrast: Self-supervised pretraining on 3DPM images for miningmaterialclassification.In:ProceedingsoftheEuropeanConferenceonCom- puter Vision Workshops. pp. 212–227 (2022)

    2022

  7. [7]

    Dong, R., Qi, Z., Zhang, L., Zhang, J., Sun, J., Ge, Z., Yi, L., Ma, K.: Autoen- coders as Cross-Modal teachers: Can pretrained 2D image transformers help 3D representation learning? In: Proceedings of the International Conference on Learn- ing Representations (2023)

    2023

  8. [8]

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

    Fan, H., Su, H., Guibas, L.: A point set generation network for 3d object recon- struction from a single image. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 2463–2471 (2017)

    2017

  9. [9]

    In: Advances in Neural Information Processing Systems (2020)

    Grill, J.B., Strub, F., Altché, F., Tallec, C., Richemond, P.H., Buchatskaya, E., Do- ersch, C., Pires, B.A., Guo, Z.D., Azar, M.G., Piot, B., Kavukcuoglu, K., Munos, R., Valko, M.: Bootstrap your own latent a new approach to self-supervised learn- ing. In: Advances in Neural Information Processing Systems (2020)

    2020

  10. [10]

    In: Proceedings of the International Joint Conference on Artificial Intelligence (2023)

    Guo, Z., Zhang, R., Qiu, L., Li, X., Heng, P.A.: Joint-mae: 2d-3d joint masked autoencoders for 3d point cloud pre-training. In: Proceedings of the International Joint Conference on Artificial Intelligence (2023)

    2023

  11. [11]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Hou, J., Graham, B., Nießner, M., Xie, S.: Exploring data-efficient 3d scene un- derstanding with contrastive scene contexts. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 15587–15597 (2021)

    2021

  12. [12]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Huang, S., Xie, Y., Zhu, S.C., Zhu, Y.: Spatio-temporal self-supervised representa- tion learning for 3d point clouds. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 6535–6545 (2021)

    2021

  13. [13]

    In: Advances in Neural Information Processing Systems (2022)

    Li, G., Zheng, H., Liu, D., Wang, C., Su, B., Zheng, C.: Semmae: Semantic-guided masking for learning masked autoencoders. In: Advances in Neural Information Processing Systems (2022)

    2022

  14. [14]

    In: Proceedings of the European Conference on Computer Vision

    Liu, H., Cai, M., Lee, Y.J.: Masked discrimination for self-supervised learning on point clouds. In: Proceedings of the European Conference on Computer Vision. pp. 657–675 (2022) MPL-MAE 17

    2022

  15. [15]

    IEEE Transactions on Multime- dia26, 3897–3908 (2024)

    Liu, J., Wu, Y., Gong, M., Liu, Z., Miao, Q., Ma, W.: Inter-modal masked autoen- coder for self-supervised learning on point clouds. IEEE Transactions on Multime- dia26, 3897–3908 (2024)

    2024

  16. [16]

    Journal of Visual Communication and Image Representation115, 104675 (2026)

    Liu, X., Wang, F., Chen, Z., Dong, X.: Dcmae: A dual-branch contrastive masked autoencoder for 3d object detection. Journal of Visual Communication and Image Representation115, 104675 (2026)

    2026

  17. [17]

    Decoupled Weight Decay Regularization

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  18. [18]

    SGDR: Stochastic Gradient Descent with Warm Restarts

    Loshchilov, I., Hutter, F.: Sgdr: Stochastic gradient descent with warm restarts. arXiv preprint arXiv:1608.03983 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  19. [19]

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

    Luo, C., Yang, X., Yuille, A.: Self-supervised pillar motion learning for autonomous driving. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 3183–3192 (2021)

    2021

  20. [20]

    In: Proceedings of the Conference on Robot Learning (2021)

    Mersch, B., Chen, X., Behley, J., Stachniss, C.: Self-supervised point cloud pre- diction using 3d spatio-temporal convolutional networks. In: Proceedings of the Conference on Robot Learning (2021)

    2021

  21. [21]

    arXiv preprint arXiv:2206.09900 (2023)

    Min, C., Xu, X., Zhao, D., Xiao, L., Nie, Y., Dai, B.: Occupancy-mae: Self- supervised pre-training large-scale lidar point clouds with masked occupancy au- toencoders. arXiv preprint arXiv:2206.09900 (2023)

    work page arXiv 2023

  22. [22]

    In: Proceedings of the European Conference on Computer Vision

    Pang, Y., Wang, W., Tay, F.E.H., Liu, W., Tian, Y., Yuan, L.: Masked autoen- coders for point cloud self-supervised learning. In: Proceedings of the European Conference on Computer Vision. pp. 604–621 (2022)

    2022

  23. [23]

    IEEE Transactions on Automation Science and Engineering19(4), 3639–3648 (2022)

    Peng, G., Ren, Z., Wang, H., Li, X., Khyam, M.O.: A self-supervised learning-based 6-dof grasp planning method for manipulator. IEEE Transactions on Automation Science and Engineering19(4), 3639–3648 (2022)

    2022

  24. [24]

    Neural Networks108, 533–543 (2018)

    Phan, A.V., Le Nguyen, M., Nguyen, Y.L.H., Bui, L.T.: Dgcnn: A convolutional neural network over large-scale labeled graphs. Neural Networks108, 533–543 (2018)

    2018

  25. [25]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

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

    2017

  26. [26]

    Advances in neural information processing systems30(2017)

    Qi, C.R., Yi, L., Su, H., Guibas, L.J.: Pointnet++: Deep hierarchical feature learn- ing on point sets in a metric space. Advances in neural information processing systems30(2017)

    2017

  27. [27]

    In: Proceedings of the International Conference on Machine Learning (2023)

    Qi, Z., Dong, R., Fan, G., Ge, Z., Zhang, X., Ma, K., Yi, L.: Contrast with recon- struct: Contrastive 3d representation learning guided by generative pretraining. In: Proceedings of the International Conference on Machine Learning (2023)

    2023

  28. [28]

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

    Sautier, C., Puy, G., Gidaris, S., Boulch, A., Bursuc, A., Marlet, R.: Image-to-lidar self-supervised distillation for autonomous driving data. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 9891– 9901 (2022)

    2022

  29. [29]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Uy, M.A., Pham, Q.H., Hua, B.S., Nguyen, T., Yeung, S.K.: 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. pp. 1588–1597 (2019)

    2019

  30. [30]

    In: Proceedings of the IEEE/CVF interna- tional conference on computer vision

    Wang, H., Liu, Q., Yue, X., Lasenby, J., Kusner, M.J.: Unsupervised point cloud pre-training via occlusion completion. In: Proceedings of the IEEE/CVF interna- tional conference on computer vision. pp. 9782–9792 (2021)

    2021

  31. [31]

    In: Proceedings of the European Conference on Computer Vision

    Wang, H., Bao, Y., Pan, P., Li, Z., Liu, X., Yang, R., Huang, D.: Multi-modal relation distillation for unified 3D representation learning. In: Proceedings of the European Conference on Computer Vision. pp. 364–381 (2024) 18 Y. Xu et al

    2024

  32. [32]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Wang, Y., Solomon, J.M.: Deep closest point: Learning representations for point cloud registration. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 3522–3531 (2019)

    2019

  33. [33]

    In: Proceedings of the European Conference on Computer Vision

    Wu,W.,Wang,Z.Y.,Li,Z.,Liu,W.,Fuxin,L.:Pointpwc-net:Costvolumeonpoint clouds for (self-)supervised scene flow estimation. In: Proceedings of the European Conference on Computer Vision. pp. 88–107 (2020)

    2020

  34. [34]

    In: Proceedings of the IEEE/CVF Conference on computer vision and pattern recognition

    Wu, X., Wen, X., Liu, X., Zhao, H.: Masked scene contrast: A scalable framework for unsupervised 3d representation learning. In: Proceedings of the IEEE/CVF Conference on computer vision and pattern recognition. pp. 9415–9424 (2023)

    2023

  35. [35]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition

    Wu, Z., Song, S., Khosla, A., Yu, F., Zhang, L., Tang, X., Xiao, J.: 3d shapenets: A deep representation for volumetric shapes. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 1912–1920 (2015)

    1912

  36. [36]

    In: European conference on computer vision

    Xie, S., Gu, J., Guo, D., Qi, C.R., Guibas, L., Litany, O.: Pointcontrast: Unsuper- vised pre-training for 3d point cloud understanding. In: European conference on computer vision. pp. 574–591 (2020)

    2020

  37. [37]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Xue, L., Gao, M., Xing, C., Martín-Martín, R., Wu, J., Xiong, C., Xu, R., Niebles, J.C., Savarese, S.: Ulip: Learning a unified representation of language, images, and point clouds for 3d understanding. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 1179–1189 (2023)

    2023

  38. [38]

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

    Xue, L., Yu, N., Zhang, S., Panagopoulou, A., Li, J., Martín-Martín, R., Wu, J., Xiong, C., Xu, R., Niebles, J.C., et al.: Ulip-2: Towards scalable multimodal pre- training for 3d understanding. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 27091–27101 (2024)

    2024

  39. [39]

    IEEE/ASME Transactions on Mechatronics29(1), 625–635 (2023)

    Yu, S., Zhai, D.H., Xia, Y.: Robotic grasp detection based on category-level ob- ject pose estimation with Self-Supervised learning. IEEE/ASME Transactions on Mechatronics29(1), 625–635 (2023)

    2023

  40. [40]

    In: IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Yu, X., Tang, L., Rao, Y., Huang, T., Zhou, J., Lu, J.: Point-bert: Pre-training 3d point cloud transformers with masked point modeling. In: IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 19291–19300 (2022)

    2022

  41. [41]

    In: Proceedings of the AAAI conference on artificial intelligence

    Zha, Y., Ji, H., Li, J., Li, R., Dai, T., Chen, B., Wang, Z., Xia, S.T.: Towards compact 3d representations via point feature enhancement masked autoencoders. In: Proceedings of the AAAI conference on artificial intelligence. vol. 38, pp. 6962– 6970 (2024)

    2024

  42. [42]

    In: Advances in Neural Information Processing Systems

    Zhang, R., Guo, Z., Fang, R., Zhao, B., Wang, D., Qiao, Y., Li, H., Gao, P.: Point- m2ae: Multi-scale masked autoencoders for hierarchical point cloud pre-training. In: Advances in Neural Information Processing Systems. vol. 35, pp. 27061–27074 (2022)

    2022

  43. [43]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Zhang, R., Wang, L., Qiao, Y., Gao, P., Li, H.: Learning 3d representations from 2d pre-trained models via image-to-point masked autoencoders. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 21769– 21780 (2023)

    2023

  44. [44]

    In: Advances in Neural Information Processing Systems

    Zhang, X., Zhang, S., Yan, J.: Pcp-mae: Learning to predict centers for point masked autoencoders. In: Advances in Neural Information Processing Systems. vol. 37, pp. 80303–80327 (2024)

    2024

  45. [45]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (2025)

    Zhang, X., Zhang, S., Yan, J.: Towards more diverse and challenging pre-training for point cloud learning: Self-supervised cross reconstruction with decoupled views. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (2025)

    2025

  46. [46]

    In: Proceedings of the International Conference on Learning Representations (2023)

    Zhou, J., Wang, J., Ma, B., Liu, Y.S., Huang, T., Wang, X.: Uni3d: Exploring unified 3d representation at scale. In: Proceedings of the International Conference on Learning Representations (2023)

    2023