arxiv: 2605.10456 · v1 · submitted 2026-05-11 · 💻 cs.RO

Recognition: 2 theorem links

· Lean Theorem

Learning Point Cloud Geometry as a Statistical Manifold: Theory and Practice

Jinwoo Lee , Jiwoo Kim , Woojae Shin , Giseop Kim , Hyondong Oh

Authors on Pith no claims yet

Pith reviewed 2026-05-12 05:30 UTC · model grok-4.3

classification 💻 cs.RO

keywords point cloud geometrystatistical manifoldGaussian distributionself-supervised learningrobotic perceptionLiDARlocal geometry estimation

0 comments

The pith

Point cloud local geometry can be represented as a statistical manifold of per-point Gaussians learned self-supervised from sparse observations.

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

The paper establishes that local geometry in point clouds can be explicitly modeled as a statistical manifold where each point is tied to a Gaussian distribution capturing its neighborhood structure. This formulation allows a neural network called POLI to map raw sparse point cloud inputs to these Gaussian parameters using only self-supervised losses, eliminating the need for labeled geometry data. A sympathetic reader would care because LiDAR data is typically incomplete and non-uniform, so better intrinsic geometry estimates should improve downstream tasks like localization and mapping without architectural changes to existing pipelines. The approach preserves geometric inductive biases by grounding the model in a family of Gaussians rather than generic feature learning.

Core claim

We represent local geometry as a statistical manifold induced by a family of Gaussian distributions, where each point is associated with a Gaussian capturing its local geometric structure. Based on this formulation, we introduce Point-to-Ellipsoid (POLI), a deep neural estimator that predicts per-point Gaussian geometry. POLI learns a mapping from point cloud observations to their underlying geometry in a self-supervised manner, removing the need for labeled data while preserving strong geometric inductive biases.

What carries the argument

The statistical manifold induced by a family of Gaussian distributions, one per point, that encodes local geometric structure such as shape and orientation.

If this is right

Self-supervised recovery of per-point geometry parameters directly from sparse inputs without external labels.
Seamless insertion of the Gaussian representation into existing robotic perception pipelines.
Consistent performance gains on tasks including localization, mapping, and object pose estimation.
Reduced reliance on hand-crafted statistics or large supervised datasets for geometry estimation.

Where Pith is reading between the lines

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

The Gaussian manifold could be replaced by other parametric families to handle non-elliptical local structures if the single-Gaussian assumption proves too restrictive.
The same self-supervised training principle might transfer to other 3D sensing modalities such as RGB-D or radar point clouds.
Downstream modules could use the predicted covariance matrices to propagate uncertainty estimates into planning or control loops.

Load-bearing premise

Local scene geometry around each point is adequately captured by a single Gaussian distribution and a self-supervised loss can recover accurate parameters without any labeled geometry data.

What would settle it

Compare the predicted Gaussian parameters against the empirical covariance of densely sampled points in a ground-truth scene; if the predicted ellipsoids systematically deviate from the true local structure or fail to improve downstream perception metrics, the claim is falsified.

Figures

Figures reproduced from arXiv: 2605.10456 by Giseop Kim, Hyondong Oh, Jinwoo Lee, Jiwoo Kim, Woojae Shin.

**Figure 1.** Figure 1: Point cloud geometries estimated by different methods on the HeLiPR dataset, visualized with 300 samples per [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗

**Figure 2.** Figure 2: A visualization illustrating how underlying geometry [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: Probabilistic model as a directed acyclic graph, where [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: Network architecture of covariance estimator. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

**Figure 6.** Figure 6: Performance comparison on object pose estimation [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗

**Figure 5.** Figure 5: Runtime composition of POLI training pipeline. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

**Figure 7.** Figure 7: Comparison of LiDAR odometry performance across different algorithms based on RPE (m). The best results are [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗

**Figure 8.** Figure 8: Comparison of LiDAR odometry trajectories for dif [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗

**Figure 9.** Figure 9: Scan augmentation process using POLI: raw scan, estimated covariances, and augmented scan generated by sampling. [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗

read the original abstract

Point clouds are a fundamental representation for robotic perception tasks such as localization, mapping, and object pose estimation. However, LiDAR-acquired point clouds are inherently sparse and non-uniform, providing incomplete observations of the underlying scene geometry. This makes reliable geometric reasoning challenging and degrades downstream perception performance. Existing approaches attempt to compensate for these limitations by estimating local geometry, but often rely on hand-crafted statistics or end-to-end supervised learning, which can suffer from limited scalability or require large amounts of accurately labeled data. To address these challenges, we explicitly model point cloud geometry under a principled mathematical formulation. We represent local geometry as a statistical manifold induced by a family of Gaussian distributions, where each point is associated with a Gaussian capturing its local geometric structure. Based on this formulation, we introduce Point-to-Ellipsoid (POLI), a deep neural estimator that predicts per-point Gaussian geometry. POLI learns a mapping from point cloud observations to their underlying geometry in a self-supervised manner, removing the need for labeled data while preserving strong geometric inductive biases. The resulting representation integrates seamlessly into existing robotic perception pipelines without architectural modifications. Extensive experiments show that POLI enables accurate and robust geometry estimation and consistently improves performance across diverse robotic perception tasks.

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.

Desk Editor's Note private letter to a colleague

The paper models point cloud geometry as a statistical manifold of per-point Gaussians recovered by a self-supervised network called POLI, which targets sparsity in LiDAR but stays too high-level in the abstract to judge the math or gains.

read the letter

The main takeaway is that they treat local geometry around each point as a Gaussian on a statistical manifold and train POLI to predict those parameters directly from the observed cloud in a self-supervised way. This framing is new in how it ties the manifold idea to per-point distributions for robotic tasks, and it avoids labeled data, which is a practical step forward from supervised alternatives mentioned in the abstract. They also emphasize easy integration into existing localization and mapping pipelines, which could make adoption straightforward if it works. The motivation around sparse, non-uniform LiDAR data is clear and directly relevant to real perception bottlenecks. The self-supervised angle preserves geometric biases without hand-crafted stats, which is a reasonable direction. The soft spots are that no equations, loss terms, or derivation appear in the abstract, so the exact manifold construction and how self-supervision recovers accurate Gaussians remain opaque. A single Gaussian per point may not capture edges or multi-surface regions well, and without numbers or baselines from the experiments it is hard to know if the claimed improvements hold up across tasks. This is aimed at robotic perception researchers who work with LiDAR and want reusable geometric features for mapping or pose estimation. Readers interested in manifold or statistical models for geometry could extract the perspective even if the implementation needs checking. It deserves a serious referee because the high-level formulation is internally consistent and hits a genuine practical issue, though the math and results sections will need close scrutiny.

Referee Report

0 major / 3 minor

Summary. The paper proposes modeling local geometry in point clouds as a statistical manifold induced by a family of per-point Gaussian distributions. It introduces POLI, a deep neural network that predicts these Gaussian parameters from raw point cloud observations in a self-supervised manner, with the goal of improving downstream robotic perception tasks such as localization, mapping, and pose estimation without requiring labeled geometry data.

Significance. If the central claims hold, the work provides a principled integration of statistical manifold theory with self-supervised learning for handling sparse and non-uniform point clouds, a common challenge in robotics. The self-supervised formulation with geometric inductive biases and seamless integration into existing pipelines is a strength; the approach could reduce dependence on large labeled datasets while offering interpretable local geometry estimates.

minor comments (3)

Abstract: The abstract states the high-level formulation but omits any key equation, loss function, or quantitative result, which reduces immediate clarity on the precise technical contribution.
The manuscript would benefit from an explicit statement of the self-supervised loss in the main text (rather than only in supplementary material) to allow readers to verify how the Gaussian parameters are recovered without ground-truth geometry.
Figure captions and axis labels in the experimental section could be expanded to include the exact metrics and baselines used, improving reproducibility.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive summary, significance assessment, and recommendation for minor revision. The report accurately reflects our contribution of casting point cloud geometry as a statistical manifold of per-point Gaussians learned self-supervisedly via POLI. As no specific major comments appear in the report, we provide no point-by-point responses below and will incorporate any minor suggestions into the revised manuscript.

Circularity Check

0 steps flagged

No significant circularity; derivation remains self-contained against external benchmarks

full rationale

The provided abstract and high-level formulation present a modeling choice (local geometry as per-point Gaussians on a statistical manifold) followed by a self-supervised estimator (POLI) without any visible equations, fitted parameters renamed as predictions, or load-bearing self-citations. No derivation chain is stated that reduces a claimed result to its own inputs by construction. The approach is internally consistent at the level of description given, with the central claim resting on the inductive bias of Gaussian modeling and self-supervision rather than on any tautological step. This is the expected honest non-finding for a paper whose core contribution is a new representation and learning method without exhibited self-referential math.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the assumption that local geometry is Gaussian and that self-supervision suffices; no free parameters or invented entities are explicitly named in the abstract.

axioms (2)

domain assumption Local geometry around each point can be represented by a Gaussian distribution
Invoked when stating that each point is associated with a Gaussian capturing local structure.
domain assumption Self-supervised learning can recover accurate Gaussian parameters from incomplete observations
Required for the claim that POLI works without labeled data.

pith-pipeline@v0.9.0 · 5525 in / 1252 out tokens · 76326 ms · 2026-05-12T05:30:46.971216+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

we model point cloud geometry as a statistical manifold, denoted by M^g, induced by Gaussian distributions... each parameterized by a mean x̄_gi ∈ R^3 and a covariance matrix C_gi ∈ S^3_+
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Φ(T; C_q) ≜ ½ Σ [log|2C_qi| + d_i^T (2C_qi)^{-1} d_i] + ½ log|Γ| + ½ ξ(T)^T Γ^{-1} ξ(T)

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

57 extracted references · 57 canonical work pages · 1 internal anchor

[1]

American Mathematical Soc., 2000

Shun-ichi Amari and Hiroshi Nagaoka.Methods of infor- mation geometry, volume 191. American Mathematical Soc., 2000. URL https://books.google.co.kr/books?hl= ko&lr=&id=vc2FWSo7wLUC&oi=fnd&pg=PR7&dq= Methods+of+information+geometry&ots=4JsyKx44L &sig=ZMJpkAT36TZSpRPPkYM-gxyXLLM&redir esc=y#v=onepage&q=Methods%20of%20information% 20geometry&f=false

work page 2000
[2]

DeepFit: 3D Surface Fitting via Neural Network Weighted Least Squares

Yizhak Ben-Shabat and Stephen Gould. DeepFit: 3D Surface Fitting via Neural Network Weighted Least Squares. InEuropean Conference on Computer Vi- sion (ECCV), 2020. doi: 10.1007/978-3-030-58452-8

work page doi:10.1007/978-3-030-58452-8 2020
[3]

URL https://link.springer.com/chapter/10.1007/ 978-3-030-58452-8 2

work page
[4]

Nesti-Net: Normal Estimation for Unstructured 3D Point Clouds Using Convolutional Neural Networks

Yizhak Ben-Shabat, Michael Lindenbaum, and Anath Fischer. Nesti-Net: Normal Estimation for Unstructured 3D Point Clouds Using Convolutional Neural Networks. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),

work page
[5]

URL https://openaccess.thecvf.com/content CVPR 2019/ html/Ben-Shabat Nesti-Net Normal Estimation for Unstructured 3D Point Clouds Using Convolutional CVPR 2019 paper.html

doi: 10.1109/CVPR.2019.01035. URL https://openaccess.thecvf.com/content CVPR 2019/ html/Ben-Shabat Nesti-Net Normal Estimation for Unstructured 3D Point Clouds Using Convolutional CVPR 2019 paper.html

work page doi:10.1109/cvpr.2019.01035 2019
[6]

Besl and Neil D

Paul J. Besl and Neil D. McKay. A Method for Registration of 3D Shapes.IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239– 256, 1992. doi: 10.1109/34.121791. URL https://www. spiedigitallibrary.org/conference-proceedings-of-spie/ 1611/1/Method-for-registration-of-3-D-shapes/10.1117/ 12.57955.short

work page doi:10.1109/34.121791 1992
[7]

Efficient and modular implicit differentiation.Advances in neural information processing systems, 35:5230–5242, 2022

Mathieu Blondel, Quentin Berthet, Marco Cuturi, Roy Frostig, Stephan Hoyer, Felipe Llinares-L ´opez, Fabian Pedregosa, and Jean-Philippe Vert. Efficient and modular implicit differentiation.Advances in neural information processing systems, 35:5230–5242, 2022. URL https: //proceedings.neurips.cc/paper files/paper/2022/hash/ 228b9279ecf9bbafe582406850c5711...

work page 2022
[8]

Convex global 3d registration with lagrangian duality

Jesus Briales and Javier Gonzalez-Jimenez. Convex global 3d registration with lagrangian duality. InPro- ceedings of the IEEE conference on computer vision and pattern recognition, pages 4960–4969, 2017

work page 2017
[9]

A volumetric method for building complex models from range images

Brian Curless and Marc Levoy. A volumetric method for building complex models from range images. InProceed- ings of the 23rd annual conference on Computer graphics and interactive techniques, pages 303–312, 1996. URL https://dl.acm.org/doi/abs/10.1145/237170.237269

work page doi:10.1145/237170.237269 1996
[10]

Courier Dover Publications, 2016

Manfredo P Do Carmo.Differential geometry of curves and surfaces: revised and updated second edition. Courier Dover Publications, 2016

work page 2016
[11]

Are we ready for autonomous driving? the kitti vision bench- mark suite

Andreas Geiger, Philip Lenz, and Raquel Urtasun. Are we ready for autonomous driving? the kitti vision bench- mark suite. In2012 IEEE conference on computer vision and pattern recognition, pages 3354–3361. IEEE,

work page
[12]

URL https://ieeexplore.ieee.org/abstract/document/ 6248074

work page
[13]

EVO: Python Package for the Evaluation of Odometry and SLAM, 2017

Michael Grupp. EVO: Python Package for the Evaluation of Odometry and SLAM, 2017. URL https://github.com/ MichaelGrupp/evo

work page 2017
[14]

Rotational Projection Statistics for 3D Local Surface Description and Object Recognition

Yulan Guo, Ferdous Sohel, Mohammed Bennamoun, Min Lu, and Jianwei Wan. Rotational Projection Statistics for 3D Local Surface Description and Object Recognition. International Journal of Computer Vision, 105(1):63–86,

work page
[15]

URL https:// link.springer.com/article/10.1007/s11263-013-0627-y

doi: 10.1007/s11263-013-0627-y. URL https:// link.springer.com/article/10.1007/s11263-013-0627-y

work page doi:10.1007/s11263-013-0627-y
[16]

On semidefinite relaxations for matrix-weighted state-estimation problems in robotics.IEEE Transac- tions on Robotics, 40:4805–4824, 2024

Connor Holmes, Frederike D ¨umbgen, and Timothy Bar- foot. On semidefinite relaxations for matrix-weighted state-estimation problems in robotics.IEEE Transac- tions on Robotics, 40:4805–4824, 2024. URL https: //ieeexplore.ieee.org/abstract/document/10706005

work page arXiv 2024
[17]

SDPRLayers: Certifiable Backpropagation Through Polynomial Optimization Problems in Robotics

Connor Holmes, Frederike D ¨umbgen, and Timothy D Barfoot. SDPRLayers: Certifiable Backpropagation Through Polynomial Optimization Problems in Robotics. IEEE Transactions on Robotics, 2025. URL https: //ieeexplore.ieee.org/abstract/document/11029144/

work page arXiv 2025
[18]

Surface reconstruction from unorganized points

Hugues Hoppe, Tony DeRose, Tom Duchamp, John McDonald, and Werner Stuetzle. Surface reconstruction from unorganized points. InProceedings of the 19th annual conference on computer graphics and interactive techniques, pages 71–78, 1992. URL https://dl.acm.org/ doi/abs/10.1145/133994.134011

work page doi:10.1145/133994.134011 1992
[19]

HeLiPR: Heterogeneous LiDAR Dataset for Inter-LiDAR Place Recognition Under Spatiotemporal Variations.The In- ternational Journal of Robotics Research, 43(3):1867– 1883, 2024

Minwoo Jung, Wooseong Yang, Dongjae Lee, Hyeon- jae Gil, Giseop Kim, and Ayoung Kim. HeLiPR: Heterogeneous LiDAR Dataset for Inter-LiDAR Place Recognition Under Spatiotemporal Variations.The In- ternational Journal of Robotics Research, 43(3):1867– 1883, 2024. URL https://journals.sagepub.com/doi/full/ 10.1177/02783649241242136

work page doi:10.1177/02783649241242136 2024
[20]

V oxelized GICP for Fast and Accu- rate 3D Point Cloud Registration

Kenji Koide, Masashi Yokozuka, Shuji Oishi, and At- suhiko Banno. V oxelized GICP for Fast and Accu- rate 3D Point Cloud Registration. InProceedings of the IEEE International Conference on Robotics and Automation (ICRA), pages 11054–11059, 2021. URL https://ieeexplore.ieee.org/abstract/document/9560835

work page arXiv 2021
[21]

GLIM: 3D range-inertial localiza- tion and mapping with GPU-accelerated scan match- ing factors.Robotics and Autonomous Systems, 179: 104750, 2024

Kenji Koide, Masashi Yokozuka, Shuji Oishi, and At- suhiko Banno. GLIM: 3D range-inertial localiza- tion and mapping with GPU-accelerated scan match- ing factors.Robotics and Autonomous Systems, 179: 104750, 2024. URL https://www.sciencedirect.com/ science/article/pii/S0921889024001349

work page 2024
[22]

Implicit LiDAR network: LiDAR super-resolution via interpolation weight prediction

Youngsun Kwon, Minhyuk Sung, and Sung-Eui Yoon. Implicit LiDAR network: LiDAR super-resolution via interpolation weight prediction. In2022 international conference on robotics and automation (ICRA), pages 8424–8430. IEEE, 2022. URL https://ieeexplore.ieee.org/ abstract/document/9811992

work page arXiv 2022
[23]

GenZ-ICP: Generalizable and Degeneracy-Robust LiDAR Odometry Using an Adaptive Weighting.IEEE Robotics and Automation Letters, 9(1):456–463, 2024

Daehan Lee, Hyungtae Lim, and Soohee Han. GenZ-ICP: Generalizable and Degeneracy-Robust LiDAR Odometry Using an Adaptive Weighting.IEEE Robotics and Automation Letters, 9(1):456–463, 2024. URL https: //ieeexplore.ieee.org/abstract/document/10753079

work page arXiv 2024
[24]

Deep iterative surface normal estimation

Jan Eric Lenssen, Christian Osendorfer, and Jonathan Masci. Deep iterative surface normal estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11247–11256,

work page
[25]

URL https://openaccess.thecvf.com/content CVPR 2020/html/Lenssen Deep Iterative Surface Normal Estimation CVPR 2020 paper.html

work page 2020
[26]

Learning Normals of Noisy Points by Local Gradient-Aware Surface Filtering

Qing Li, Huifang Feng, Xun Gong, and Yu-Shen Liu. Learning Normals of Noisy Points by Local Gradient-Aware Surface Filtering. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 28828–28838, 2025. URL https://openaccess.thecvf.com/content/ICCV2025/html/ Li Learning Normals of Noisy Points by Local Gradient-Aware Surface Filter...

work page 2025
[27]

High-quality Point Cloud Oriented Normal Estimation via Hybrid Angular and Euclidean Distance Encoding

Yuanqi Li, Jingcheng Huang, Hongshen Wang, Peiyuan Lv, Yansong Liu, Jiuming Zheng, Jie Guo, and Yanwen Guo. High-quality Point Cloud Oriented Normal Estimation via Hybrid Angular and Euclidean Distance Encoding. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 1287–1296, 2025. URL https://openaccess.thecvf.com/ content/CVPR20...

work page 2025
[28]

Quatro++: Robust global registration exploiting ground segmen- tation for loop closing in LiDAR SLAM.The In- ternational Journal of Robotics Research, 43(5):685– 715, 2024

Hyungtae Lim, Beomsoo Kim, Daebeom Kim, Eu- ngchang Mason Lee, and Hyun Myung. Quatro++: Robust global registration exploiting ground segmen- tation for loop closing in LiDAR SLAM.The In- ternational Journal of Robotics Research, 43(5):685– 715, 2024. URL https://journals.sagepub.com/doi/full/10. 1177/02783649231207654

work page 2024
[29]

KISS-Matcher: Fast and Robust Point Cloud Registration Revisited

Hyungtae Lim, Daebeom Kim, Gunhee Shin, Jingnan Shi, Ignacio Vizzo, Hyun Myung, Jaesik Park, and Luca Carlone. KISS-Matcher: Fast and Robust Point Cloud Registration Revisited. InProceedings of the IEEE International Conference on Robotics and Automation,

work page
[30]

URL https://ieeexplore.ieee.org/abstract/document/ 11127458

work page
[31]

Cambridge university press, 2003

David JC MacKay.Information theory, inference and learning algorithms. Cambridge university press, 2003. URL https://books.google.co.kr/ books?hl=ko&lr=&id=AKuMj4PN EMC&oi=fnd& pg=PR11&dq=Information+theory,+inference+ and+learning+algorithms&ots=EOnrla3wBb&sig= 6ziiQReq1O5rpGa9MGPyKNwGdeI&redir esc=y#v= onepage&q=Information%20theory%2C%20inference% 20an...

work page 2003
[32]

Scaling Diffusion Models to Real-World 3D LiDAR Scene Completion

Lucas Nunes, Rodrigo Marcuzzi, Benedikt Mersch, Jens Behley, and Cyrill Stachniss. Scaling Diffusion Models to Real-World 3D LiDAR Scene Completion. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 14770–14780, 2024. URL https://openaccess.thecvf.com/content/CVPR2024/html/ Nunes Scaling Diffusion Models to ...

work page 2024
[33]

Multi- scale feature extraction on point-sampled surfaces

Mark Pauly, Richard Keiser, and Markus Gross. Multi- scale feature extraction on point-sampled surfaces. In Computer Graphics Forum, volume 22, pages 281–289. Wiley Online Library, 2003. URL https://onlinelibrary. wiley.com/doi/full/10.1111/1467-8659.00675

work page doi:10.1111/1467-8659.00675 2003
[34]

Charles Ruizhongtai Qi, Li Yi, Hao Su, and Leonidas J. Guibas. PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space. InAdvances in Neural Information Processing Systems, 2017. URL https://proceedings.neurips.cc/paper/2017/hash/ d8bf84be3800d12f74d8b05e9b89836f-Abstract.html

work page 2017
[35]

Pu-transformer: Point cloud upsampling transformer

Shi Qiu, Saeed Anwar, and Nick Barnes. Pu-transformer: Point cloud upsampling transformer. InProceedings of the Asian conference on computer vision, pages 2475– 2493, 2022. URL https://openaccess.thecvf.com/content/ ACCV2022/html/Qiu PU-Transformer Point Cloud Upsampling Transformer ACCV 2022 paper.html

work page 2022
[36]

Efficient Vari- ants of the ICP Algorithm

Szymon Rusinkiewicz and Marc Levoy. Efficient Vari- ants of the ICP Algorithm. InProceedings of the Interna- tional Conference on 3D Digital Imaging and Modeling,

work page
[37]

URL https://ieeexplore.ieee.org/abstract/document/ 924423

work page
[38]

Marton, Nico Blodow, and Michael Beetz

Radu Bogdan Rusu, Zoltan C. Marton, Nico Blodow, and Michael Beetz. Persistent Point Feature Histograms for 3D Point Clouds. InProceedings of the 10th Inter- national Conference on Intelligent Autonomous Systems (IAS), 2008. URL https://ebooks.iospress.nl/doi/10.3233/ 978-1-58603-887-8-119

work page 2008
[39]

Fast Point Feature Histograms (FPFH) for 3D registra- tion

Radu Bogdan Rusu, Nico Blodow, and Michael Beetz. Fast Point Feature Histograms (FPFH) for 3D registra- tion. In2009 IEEE International Conference on Robotics and Automation, pages 3212–3217. IEEE, 2009

work page 2009
[40]

Generalized-ICP

Avi Segal, Dirk Haehnel, and Sebastian Thrun. Generalized-ICP. InProceedings of Robotics: Science and Systems, 2009. URL https://www.robots.ox.ac.uk/ ∼avsegal/resources/papers/Generalized ICP.pdf

work page 2009
[41]

BUFFER-X: Towards Zero- Shot Point Cloud Registration in Diverse Scenes.arXiv preprint arXiv:2503.07940, 2025

Minkyun Seo, Hyungtae Lim, Kanghee Lee, Luca Car- lone, and Jaesik Park. BUFFER-X: Towards Zero- Shot Point Cloud Registration in Diverse Scenes.arXiv preprint arXiv:2503.07940, 2025. URL https://arxiv.org/ abs/2503.07940

work page arXiv 2025
[42]

ROBIN: a graph-theoretic approach to reject outliers in robust estimation using invariants

Jingnan Shi, Heng Yang, and Luca Carlone. ROBIN: a graph-theoretic approach to reject outliers in robust estimation using invariants. In2021 IEEE International Conference on Robotics and Automation (ICRA), pages 13820–13827. IEEE, 2021. URL https://ieeexplore.ieee. org/abstract/document/9562007

work page arXiv 2021
[43]

Unique Signatures of Histograms for Local Surface Description

Federico Tombari, Samuele Salti, and Luigi Di Stefano. Unique Signatures of Histograms for Local Surface Description. InProceedings of the European Con- ference on Computer Vision (ECCV), pages 356–369,

work page
[44]

URL https://link.springer.com/chapter/10.1007/ 978-3-642-15558-1 26

work page
[45]

Make it dense: Self-supervised geometric scan comple- tion of sparse 3D lidar scans in large outdoor envi- ronments.IEEE Robotics and Automation Letters, 7 (3):8534–8541, 2022

Ignacio Vizzo, Benedikt Mersch, Rodrigo Marcuzzi, Louis Wiesmann, Jens Behley, and Cyrill Stachniss. Make it dense: Self-supervised geometric scan comple- tion of sparse 3D lidar scans in large outdoor envi- ronments.IEEE Robotics and Automation Letters, 7 (3):8534–8541, 2022. URL https://ieeexplore.ieee.org/ abstract/document/9812507

work page arXiv 2022
[46]

KISS-ICP: In Defense of Point-to-Point ICP – Simple, Accurate, and Robust Registration If Done the Right Way.IEEE Robotics and Automation Letters, 8(2):1029– 1036, 2023

Ignacio Vizzo, Tiziano Guadagnino, Benedikt Mersch, Louis Wiesmann, Jens Behley, and Cyrill Stachniss. KISS-ICP: In Defense of Point-to-Point ICP – Simple, Accurate, and Robust Registration If Done the Right Way.IEEE Robotics and Automation Letters, 8(2):1029– 1036, 2023. URL https://ieeexplore.ieee.org/abstract/ document/10015694/

work page arXiv 2023
[47]

F-LOAM: Fast Lidar Odometry and Mapping

Han Wang, Chen Wang, Chun-Lin Chen, and Lihua Xie. F-LOAM: Fast Lidar Odometry and Mapping. In Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems, 2021. URL https: //ieeexplore.ieee.org/abstract/document/9636655

work page arXiv 2021
[48]

FAST-LIO: A Fast, Robust LiDAR-Inertial Odometry Package by Tightly-Coupled Iterated Kalman Filter

Wei Xu, Yixi Cai, Dawei He, Jiarong Lin, and Fu Zhang. FAST-LIO: A Fast, Robust LiDAR-Inertial Odometry Package by Tightly-Coupled Iterated Kalman Filter. IEEE Robotics and Automation Letters, 2021. doi: 10.1109/LRA.2021.3064227. URL https://ieeexplore. ieee.org/abstract/document/9372856/

work page doi:10.1109/lra.2021.3064227 2021
[49]

FAST-LIO2: Fast Direct LiDAR-Inertial Odometry.IEEE Transactions on Robotics, 2022

Wei Xu, Yixi Cai, Dawei He, Jiarong Lin, and Fu Zhang. FAST-LIO2: Fast Direct LiDAR-Inertial Odometry.IEEE Transactions on Robotics, 2022. doi: 10.1109/TRO.2022.3141876. URL https://ieeexplore. ieee.org/abstract/document/9697912/

work page doi:10.1109/tro.2022.3141876 2022
[50]

Tulip: Transformer for upsampling of lidar point clouds

Bin Yang, Patrick Pfreundschuh, Roland Siegwart, Marco Hutter, Peyman Moghadam, and Vaishakh Patil. Tulip: Transformer for upsampling of lidar point clouds. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15354–15364, 2024. URL https://openaccess.thecvf.com/content/CVPR2024/ html/Yang TULIP Transformer for Upsamp...

work page 2024
[51]

Teaser: Fast and certifiable point cloud registration.IEEE Transactions on Robotics, 37(2):314–333, 2020

Heng Yang, Jingnan Shi, and Luca Carlone. Teaser: Fast and certifiable point cloud registration.IEEE Transactions on Robotics, 37(2):314–333, 2020. URL https://ieeexplore.ieee.org/abstract/document/9286491/

work page arXiv 2020
[52]

RPM-Net: Robust Point Matching Using Learned Features

Zi Jian Yew and Gim Hee Lee. RPM-Net: Robust Point Matching Using Learned Features. In2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 11821–11830,

work page
[53]

moco , url=

doi: 10.1109/CVPR42600.2020.01184. URL https://openaccess.thecvf.com/content CVPR 2020/ html/Yew RPM-Net Robust Point Matching Using Learned Features CVPR 2020 paper.html

work page doi:10.1109/cvpr42600.2020.01184 2020
[54]

LOAM: Lidar Odometry and Mapping in Real-time

Ji Zhang and Sanjiv Singh. LOAM: Lidar Odometry and Mapping in Real-time. InProceedings of Robotics: Science and Systems, Berkeley, USA, July 2014. doi: 10. 15607/RSS.2014.X.007. URL https://www.ri.cmu.edu/ pub files/2014/7/Ji LidarMapping RSS2014 v8.pdf

work page 2014
[55]

3D shape generation and completion through point-voxel diffusion

Linqi Zhou, Yilun Du, and Jiajun Wu. 3D shape generation and completion through point-voxel diffusion. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 5826–5835, 2021. URL https://openaccess.thecvf.com/content/ICCV2021/ html/Zhou 3D Shape Generation and Completion Through Point-V oxel Diffusion ICCV 2021 paper. html?ref=ht...

work page 2021
[56]

Fast Global Registration

Qian-Yi Zhou, Jaesik Park, and Vladlen Koltun. Fast Global Registration. InProceedings of the European Conference on Computer Vision (ECCV), pages 766– 782, 2016. URL https://link.springer.com/chapter/10. 1007/978-3-319-46475-6 47

work page 2016
[57]

Open3D: A Modern Library for 3D Data Processing

Qian-Yi Zhou, Jaesik Park, and Vladlen Koltun. Open3D: A Modern Library for 3D Data Processing.arXiv preprint arXiv:1801.09847, 2018. URL https://arxiv.org/ abs/1801.09847. SU P P L E M E N T A R YMA T E R I A L Learning Point Cloud Geometry as a Statistical Manifold: Theory and Practice Jinwoo Lee *,1 Jiwoo Kim *,1 Woojae Shin1 Giseop Kim 2 Hyondong Oh †...

work page internal anchor Pith review arXiv 2018