pith. sign in

arxiv: 2506.13183 · v2 · submitted 2025-06-16 · 💻 cs.CV

MT-PCR: Hybrid Mamba-Transformer Network with Spatial Serialization for Point Cloud Registration

Bingxi Liu , An Liu , Hao Chen , Huaqi Tao , Jinqiang Cui , Yiqun Wang , Hong Zhang This is my paper

Pith reviewed 2026-05-19 09:53 UTC · model grok-4.3

classification 💻 cs.CV
keywords point cloud registrationMambaTransformer hybridZ-order curvesspatial serialization3D computer visionstate space modelsgeometric modeling
0
0 comments X

The pith

MT-PCR hybridizes Mamba and Transformer with Z-order serialization to register point clouds more accurately and with far less memory and compute than Transformer baselines.

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

The paper introduces MT-PCR as a point cloud registration method that pairs Mamba's linear-complexity state-space modeling with a Transformer refinement stage. It serializes unordered point features along Z-order space-filling curves so that Mamba can treat the sequence as spatially coherent and capture geometric relations that would otherwise be lost. Removing the usual order-indicator token and placing the Transformer after the Mamba encoder further improves results. On standard benchmarks the approach yields higher registration accuracy while cutting GPU memory and FLOPs compared with pure Transformer and other recent methods.

Core claim

MT-PCR is the first point cloud registration framework that integrates Mamba and Transformer modules. Serializing point cloud features with Z-order space-filling curves enforces spatial locality so that an optimized Mamba encoder can model geometric structure; removing the order-indicator module improves performance in this setting. The serialized features are then refined by a Transformer stage, producing superior accuracy and efficiency with substantially lower GPU memory usage and FLOPs than Transformer-based and other state-of-the-art methods.

What carries the argument

Hybrid pipeline that serializes point features via Z-order curves, feeds the ordered sequence to a Mamba encoder for linear-complexity geometric modeling, then passes the result to a Transformer for feature refinement.

If this is right

  • Higher-resolution point clouds can be registered without the downsampling that currently discards fine detail.
  • Real-time 3D registration in robotics and autonomous driving becomes feasible on hardware with limited memory.
  • Hybrid Mamba-Transformer stacks may reduce quadratic scaling bottlenecks across other 3D vision pipelines.
  • Explicit order tokens are unnecessary once spatial locality is restored by serialization.

Where Pith is reading between the lines

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

  • The same Z-order serialization could let Mamba process other irregular geometric data such as meshes or graphs without custom ordering modules.
  • Performance on dynamic or noisy point clouds would test whether the locality assumption survives real-world sensor artifacts.
  • The removal of the order-indicator token suggests that, for geometrically serialized sequences, learned positional signals may be redundant or even counterproductive.

Load-bearing premise

That serializing point cloud features with Z-order space-filling curves sufficiently enforces spatial locality for Mamba to model geometric structure effectively.

What would settle it

Measure registration accuracy on the same point clouds after randomly permuting the Z-order sequence; a large drop relative to the correctly ordered version would indicate that the serialization step is not carrying the claimed benefit.

Figures

Figures reproduced from arXiv: 2506.13183 by An Liu, Bingxi Liu, Hao Chen, Hong Zhang, Huaqi Tao, Jinqiang Cui, Yiqun Wang.

Figure 1
Figure 1. Figure 1: Efficiency vs. Performance Trade-off of MT-PCR. (a) Registration recall vs. inference time comparison on 3DMatch. Our [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the MT-PCR Framework.The proposed pipeline consists of four stages: multi-scale feature extraction, [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Architecture of the Mamba Encoder and Block. The [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative registration results of CAST and MT [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
read the original abstract

Point cloud registration (PCR) is a fundamental task in 3D computer vision and robotics. Most learning-based PCR methods rely on Transformer architectures, which suffer from quadratic computational complexity. This limitation restricts the resolution of point clouds that can be processed, inevitably leading to information loss. In contrast, Mamba, a recently proposed model based on state-space models, achieves linear computational complexity while maintaining strong long-range contextual modeling capabilities. However, directly applying Mamba to PCR tasks yields suboptimal performance due to the unordered and irregular nature of point cloud data. To address these challenges, we propose MT-PCR, the first point cloud registration framework that integrates Mamba and Transformer modules. Specifically, we serialize point cloud features using Z-order space-filling curves to enforce spatial locality, enabling Mamba to better model the geometric structure of the inputs. Additionally, we remove the order-indicator module commonly used in Mamba-based sequence modeling, leading to improved performance in our setting. The serialized features are then processed by an optimized Mamba encoder, followed by a Transformer-based feature refinement stage. Extensive experiments on multiple benchmarks demonstrate that MT-PCR outperforms Transformer-based and other state-of-the-art methods in both accuracy and efficiency, significantly reducing GPU memory usage and FLOPs.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The manuscript proposes MT-PCR, the first hybrid Mamba-Transformer framework for point cloud registration. It serializes point cloud features via Z-order space-filling curves to impose spatial locality so that a Mamba encoder can model geometric structure, removes the order-indicator module, and follows with Transformer-based refinement. Extensive experiments on multiple benchmarks are reported to demonstrate superior accuracy and efficiency over Transformer-based and other SOTA methods together with large reductions in GPU memory and FLOPs.

Significance. If the central claims are substantiated, the work would be significant: it directly tackles the quadratic complexity barrier of Transformer-only PCR pipelines by introducing a linear-complexity Mamba stage, potentially enabling higher-resolution registration in robotics and autonomous systems. The explicit design choice of Z-order serialization plus removal of the order-indicator constitutes a concrete, testable adaptation of state-space models to unordered 3D data, and the reported efficiency gains (memory and FLOPs) would be practically valuable if reproducible across standard benchmarks.

major comments (2)
  1. [Abstract and §3] Abstract and §3 (Method): The premise that Z-order space-filling curves applied to sparse, non-grid point clouds will impose sufficient spatial locality for the Mamba state-space recurrence to capture local rigidity and correspondence cues is load-bearing for both the accuracy and efficiency claims. Because point clouds require an implicit discretization or coordinate-to-index mapping whose locality properties are not guaranteed to align with true 3D neighborhoods, this step requires either a formal locality analysis or targeted ablations showing that geometrically adjacent points remain contiguous in the serialized sequence; without such evidence the claimed advantage over pure Transformers rests on an unverified assumption.
  2. [§4] §4 (Experiments): The headline performance and efficiency numbers are presented without sufficient detail on baseline implementations, statistical significance testing, or controls for post-hoc hyper-parameter tuning. Given that the soundness assessment is limited by the absence of these elements, the cross-method superiority claims cannot yet be considered fully secured.
minor comments (2)
  1. [§3] Notation for the serialized feature sequence and the precise definition of the Z-order mapping should be introduced with an equation or pseudocode early in §3 to avoid ambiguity for readers unfamiliar with space-filling curves on irregular data.
  2. [Abstract] The abstract would benefit from naming the specific benchmarks (e.g., ModelNet, KITTI, 3DMatch) rather than referring generically to “multiple benchmarks.”

Simulated Author's Rebuttal

2 responses · 0 unresolved

We sincerely thank the referee for the constructive and detailed feedback. The comments highlight important aspects of our methodological assumptions and experimental presentation that we have addressed in the revised manuscript to strengthen the work.

read point-by-point responses

  1. Referee: [Abstract and §3] The premise that Z-order space-filling curves applied to sparse, non-grid point clouds will impose sufficient spatial locality for the Mamba state-space recurrence to capture local rigidity and correspondence cues is load-bearing for both the accuracy and efficiency claims. Because point clouds require an implicit discretization or coordinate-to-index mapping whose locality properties are not guaranteed to align with true 3D neighborhoods, this step requires either a formal locality analysis or targeted ablations showing that geometrically adjacent points remain contiguous in the serialized sequence; without such evidence the claimed advantage over pure Transformers rests on an unverified assumption.

    Authors: We agree that explicit validation of locality preservation is essential for substantiating the design. In the revised manuscript we have added a new subsection (4.3) with targeted ablations that quantify neighborhood preservation under Z-order serialization. These include (i) the fraction of k-nearest neighbors retained within fixed-length windows of the serialized sequence across varying point densities and (ii) visualizations contrasting Z-order ordering with random and grid-based alternatives. The results show that more than 82% of local 3D neighbors remain contiguous within windows of size 64, providing direct empirical support for the assumption and clarifying the advantage relative to pure Transformer pipelines. revision: yes

  2. Referee: [§4] The headline performance and efficiency numbers are presented without sufficient detail on baseline implementations, statistical significance testing, or controls for post-hoc hyper-parameter tuning. Given that the soundness assessment is limited by the absence of these elements, the cross-method superiority claims cannot yet be considered fully secured.

    Authors: We thank the referee for underscoring the need for greater experimental transparency. The revised Section 4 now provides: (1) a supplementary table listing exact baseline implementations, library versions, and hyper-parameter values used for each compared method; (2) performance metrics reported as mean ± standard deviation over five independent runs with different random seeds to demonstrate statistical stability; and (3) an explicit description of the hyper-parameter selection protocol, which applied identical validation-based tuning to all methods before final test evaluation. These additions remove ambiguity and reinforce the reliability of the reported superiority and efficiency gains. revision: yes

Circularity Check

0 steps flagged

No significant circularity: architectural design validated empirically

full rationale

The paper presents MT-PCR as a novel hybrid architecture that applies Z-order serialization to point cloud features to enable effective Mamba processing, removes the order-indicator module, and follows with Transformer refinement. These are explicit design choices justified by the unordered nature of point clouds and validated through benchmark experiments showing accuracy and efficiency gains. No derivation chain reduces a claimed result to its own inputs by construction, no self-citation load-bearing premises appear in the provided text, and performance claims rest on comparative empirical results rather than fitted parameters or self-referential equations.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that Z-order curves create sufficient spatial locality for Mamba on irregular point data and that standard deep-learning training will produce the reported gains; no free parameters or invented entities are explicitly introduced in the abstract.

axioms (1)
  • domain assumption Z-order space-filling curves enforce spatial locality in unordered point cloud features
    Invoked to justify why Mamba can model geometric structure after serialization.

pith-pipeline@v0.9.0 · 5767 in / 1278 out tokens · 29434 ms · 2026-05-19T09:53:44.838376+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

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

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

Reference graph

Works this paper leans on

53 extracted references · 53 canonical work pages · 2 internal anchors

  1. [1]

    Sheng Ao, Qingyong Hu, Hanyun Wang, Kai Xu, and Yulan Guo. 2023. Buffer: Balancing accuracy, efficiency, and generalizability in point cloud registration. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 1255–1264

    work page 2023

  2. [2]

    Sheng Ao, Qingyong Hu, Bo Yang, Andrew Markham, and Yulan Guo. 2021. Spinnet: Learning a general surface descriptor for 3d point cloud registration. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 11753–11762

    work page 2021

  3. [3]

    Simran Arora, Sabri Eyuboglu, Michael Zhang, Aman Timalsina, Silas Alberti, Dylan Zinsley, James Zou, Atri Rudra, and Christopher Ré. 2024. Simple linear attention language models balance the recall-throughput tradeoff. ArXiv preprint abs/2402.18668 (2024). https://arxiv.org/abs/2402.18668

    work page arXiv 2024

  4. [4]

    Xuyang Bai, Zixin Luo, Lei Zhou, Hongbo Fu, Long Quan, and Chiew-Lan Tai

    work page

  5. [5]

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

    D3feat: Joint learning of dense detection and description of 3d local fea- tures. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 6359–6367

    work page

  6. [6]

    Cesar Cadena, Luca Carlone, Henry Carrillo, Yasir Latif, Davide Scaramuzza, José Neira, Ian Reid, and John J Leonard. 2016. Past, present, and future of simultaneous localization and mapping: Toward the robust-perception age. IEEE Transactions on robotics 32, 6 (2016), 1309–1332

    work page 2016

  7. [7]

    Julie Carmigniani, Borko Furht, Marco Anisetti, Paolo Ceravolo, Ernesto Damiani, and Misa Ivkovic. 2011. Augmented reality technologies, systems and applications. Multimedia tools and applications 51 (2011), 341–377

    work page 2011

  8. [8]

    Suyi Chen, Hao Xu, Ru Li, Guanghui Liu, Chi-Wing Fu, and Shuaicheng Liu

    work page

  9. [9]

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

    SIRA-PCR: Sim-to-Real Adaptation for 3D Point Cloud Registration. In Proceedings of the IEEE/CVF International Conference on Computer Vision . 14394– 14405

    work page

  10. [10]

    Zhi Chen, Yufan Ren, Tong Zhang, Zheng Dang, Wenbing Tao, Sabine Süsstrunk, and Mathieu Salzmann. 2023. Diffusionpcr: Diffusion models for robust multi-step point cloud registration. arXiv preprint arXiv:2312.03053 (2023)

    work page arXiv 2023

  11. [11]

    Yuhong Chou, Man Yao, Kexin Wang, Yuqi Pan, Rui-Jie Zhu, Jibin Wu, Yiran Zhong, Yu Qiao, Bo Xu, and Guoqi Li. 2024. MetaLA: Unified optimal linear ap- proximation to softmax attention map. Advances in Neural Information Processing Systems 37 (2024), 71034–71067

    work page 2024

  12. [12]

    Christopher Choy, Jaesik Park, and Vladlen Koltun. 2019. Fully convolutional geometric features. In Proceedings of the IEEE/CVF International Conference on Computer Vision. 8958–8966

    work page 2019

  13. [13]

    Haowen Deng, Tolga Birdal, and Slobodan Ilic. 2018. Ppfnet: Global context aware local features for robust 3d point matching. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition . 195–205

    work page 2018

  14. [14]

    MA FISCHLER AND. 1981. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography. Commun. ACM 24, 6 (1981), 381–395

    work page 1981

  15. [15]

    Andreas Geiger, Philip Lenz, and Raquel Urtasun. 2012. Are we ready for au- tonomous driving? the kitti vision benchmark suite. In IEEE Conference on Com- puter Vision and Pattern Recognition . IEEE, 3354–3361

    work page 2012

  16. [16]

    Zan Gojcic, Caifa Zhou, Jan D Wegner, and Andreas Wieser. 2019. The perfect match: 3d point cloud matching with smoothed densities. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 5545–5554

    work page 2019

  17. [17]

    Albert Gu and Tri Dao. 2023. Mamba: Linear-Time Sequence Modeling with Selective State Spaces. arXiv preprint arXiv:2312.00752 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  18. [18]

    Albert Gu, Karan Goel, Ankit Gupta, and Christopher Ré. 2022. On the parame- terization and initialization of diagonal state space models. Advances in Neural Information Processing Systems 35 (2022), 35971–35983

    work page 2022

  19. [19]

    Shiyi Guo, Yihong Wu, Binjian Xie, Bingxi Liu, and Tong Jia. 2024. Low-Overlap Point Cloud Registration by Semiglobal Block Matching. IEEE Transactions on Industrial Informatics (2024)

    work page 2024

  20. [20]

    Xu Han, Yuan Tang, Zhaoxuan Wang, and Xianzhi Li. 2024. Mamba3d: Enhancing local features for 3d point cloud analysis via state space model. In Proceedings of the 32nd ACM International Conference on Multimedia . 4995–5004

    work page 2024

  21. [21]

    Renlang Huang, Yufan Tang, Jiming Chen, and Liang Li. 2024. A consistency- aware spot-guided transformer for versatile and hierarchical point cloud regis- tration. Proc. Conf. Neural Inf. Process. Syst. (2024)

    work page 2024

  22. [22]

    Shengyu Huang, Zan Gojcic, Mikhail Usvyatsov, Andreas Wieser, and Konrad Schindler. 2021. Predator: Registration of 3d point clouds with low overlap. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 4267–4276

    work page 2021

  23. [23]

    Rudolph Emil Kalman. 1960. A new approach to linear filtering and prediction problems. (1960)

    work page 1960

  24. [24]

    Jiaxin Li and Gim Hee Lee. 2019. Usip: Unsupervised stable interest point detec- tion from 3d point clouds. In Proceedings of the IEEE/CVF International Conference on Computer Vision. 361–370

    work page 2019

  25. [25]

    Dingkang Liang, Xin Zhou, Wei Xu, Xingkui Zhu, Zhikang Zou, Xiaoqing Ye, Xiao Tan, and Xiang Bai. 2024. PointMamba: A Simple State Space Model for Point Cloud Analysis. In Advances in Neural Information Processing Systems

    work page 2024

  26. [26]

    Jiuming Liu, Ruiji Yu, Yian Wang, Yu Zheng, Tianchen Deng, Weicai Ye, and Hesheng Wang. 2024. Point mamba: A novel point cloud backbone based on state space model with octree-based ordering strategy. arXiv preprint arXiv:2403.06467 (2024)

    work page arXiv 2024

  27. [27]

    Yue Liu, Yunjie Tian, Yuzhong Zhao, Hongtian Yu, Lingxi Xie, Yaowei Wang, Qixiang Ye, Jianbin Jiao, and Yunfan Liu. 2024. Vmamba: Visual state space model. Advances in neural information processing systems 37 (2024), 103031–103063

    work page 2024

  28. [28]

    Ilya Loshchilov and Frank Hutter. 2018. Decoupled Weight Decay Regularization. In International Conference on Learning Representations

    work page 2018

  29. [29]

    Fan Lu, Guang Chen, Yinlong Liu, Lijun Zhang, Sanqing Qu, Shu Liu, Rongqi Gu, and Changjun Jiang. 2023. HRegNet: A hierarchical network for efficient and accurate outdoor LiDAR point cloud registration. IEEE Transactions on Pattern Analysis and Machine Intelligence (2023)

    work page 2023

  30. [30]

    Weixin Lu, Yao Zhou, Guowei Wan, Shenhua Hou, and Shiyu Song. 2019. L3-net: Towards learning based lidar localization for autonomous driving. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 6389–6398

    work page 2019

  31. [31]

    Aaron van den Oord, Yazhe Li, and Oriol Vinyals. 2018. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  32. [32]

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Des- maison, Andreas Kopf, Edward Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. 2019. PyTorch: An Imperative Style, Hi...

    work page 2019

  33. [33]

    Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. 2017. Pointnet: Deep learning on point sets for 3d classification and segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition . 652–660

    work page 2017

  34. [34]

    Zheng Qin, Hao Yu, Changjian Wang, Yulan Guo, Yuxing Peng, Slobodan Ilic, Dewen Hu, and Kai Xu. 2023. GeoTransformer: Fast and Robust Point Cloud Registration With Geometric Transformer. IEEE Transactions on Pattern Analysis and Machine Intelligence (2023)

    work page 2023

  35. [35]

    Liliang Ren, Yang Liu, Yadong Lu, Yelong Shen, Chen Liang, and Weizhu Chen

    work page

  36. [36]

    arXiv preprint arXiv:2406.07522 , year=

    Samba: Simple Hybrid State Space Models for Efficient Unlimited Context Language Modeling. ArXiv preprint abs/2406.07522 (2024). https://arxiv.org/abs/ 2406.07522

    work page arXiv 2024

  37. [37]

    Radu Bogdan Rusu, Nico Blodow, and Michael Beetz. 2009. Fast point feature histograms (FPFH) for 3D registration. In 2009 IEEE International Conference on Robotics and Automation. IEEE, 3212–3217

    work page 2009

  38. [38]

    Radu Bogdan Rusu, Nico Blodow, Zoltan Csaba Marton, and Michael Beetz. 2008. Aligning point cloud views using persistent feature histograms. In 2008 IEEE/RSJ International Conference on Intelligent Robots and Systems . IEEE, 3384–3391

    work page 2008

  39. [39]

    Samuele Salti, Federico Tombari, and Luigi Di Stefano. 2014. SHOT: Unique signatures of histograms for surface and texture description. Computer Vision and Image Understanding 125 (2014), 251–264

    work page 2014

  40. [40]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. Advances in neural information processing systems 30 (2017)

    work page 2017

  41. [41]

    Haiping Wang, Yuan Liu, Zhen Dong, and Wenping Wang. 2022. You only hypothesize once: Point cloud registration with rotation-equivariant descriptors. In Proceedings of the 30th ACM International Conference on Multimedia . 1630– 1641

    work page 2022

  42. [42]

    Jingtao Wang and Zechao Li. 2024. 3DPCP-Net: A Lightweight Progressive 3D Correspondence Pruning Network for Accurate and Efficient Point Cloud Regis- tration. In Proceedings of the 32nd ACM International Conference on Multimedia . MM ’25, October 28-31, 2025, Dublin, Ireland Bingxi Liu, An Liu, Hao Chen, Jinqiang Cui, Yiqun Wang, and Hong Zhang ∗ 1885–1894

    work page 2024

  43. [43]

    Yue Wang and Justin M Solomon. 2019. Deep closest point: Learning representa- tions for point cloud registration. In Proceedings of the IEEE/CVF international conference on computer vision . 3523–3532

    work page 2019

  44. [44]

    Fan Yang, Lin Guo, Zhi Chen, and Wenbing Tao. 2022. One-inlier is first: Towards efficient position encoding for point cloud registration. Advances in Neural Information Processing Systems 35 (2022), 6982–6995

    work page 2022

  45. [45]

    Heng Yang, Jingnan Shi, and Luca Carlone. 2020. Teaser: Fast and certifiable point cloud registration. IEEE Transactions on Robotics 37, 2 (2020), 314–333

    work page 2020

  46. [46]

    Zi Jian Yew and Gim Hee Lee. 2018. 3dfeat-net: Weakly supervised local 3d features for point cloud registration. In Proceedings of the European Conference on Computer Vision (ECCV) . 607–623

    work page 2018

  47. [47]

    Zi Jian Yew and Gim Hee Lee. 2022. Regtr: End-to-end point cloud correspon- dences with transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 6677–6686

    work page 2022

  48. [48]

    Hao Yu, Fu Li, Mahdi Saleh, Benjamin Busam, and Slobodan Ilic. 2021. Cofinet: Reliable coarse-to-fine correspondences for robust point cloud registration. Ad- vances in Neural Information Processing Systems 34 (2021), 23872–23884

    work page 2021

  49. [49]

    Hao Yu, Zheng Qin, Ji Hou, Mahdi Saleh, Dongsheng Li, Benjamin Busam, and Slobodan Ilic. 2023. Rotation-invariant transformer for point cloud matching. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 5384–5393

    work page 2023

  50. [50]

    Junle Yu, Luwei Ren, Yu Zhang, Wenhui Zhou, Lili Lin, and Guojun Dai. 2023. PEAL: Prior-embedded explicit attention learning for low-overlap point cloud registration. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 17702–17711

    work page 2023

  51. [51]

    Andy Zeng, Shuran Song, Matthias Nießner, Matthew Fisher, Jianxiong Xiao, and Thomas Funkhouser. 2017. 3dmatch: Learning local geometric descriptors from rgb-d reconstructions. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 1802–1811

    work page 2017

  52. [52]

    Xiyu Zhang, Jiaqi Yang, Shikun Zhang, and Yanning Zhang. 2023. 3D registration with maximal cliques. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 17745–17754

    work page 2023

  53. [53]

    Lianghui Zhu, Bencheng Liao, Qian Zhang, Xinlong Wang, Wenyu Liu, and Xinggang Wang. [n. d.]. Vision Mamba: Efficient Visual Representation Learning with Bidirectional State Space Model. In Forty-first International Conference on Machine Learning

    work page