arxiv: 2604.05060 · v2 · submitted 2026-04-06 · 💻 cs.CV · cs.LG

Recognition: 2 theorem links

· Lean Theorem

R3PM-Net: Real-time, Robust, Real-world Point Matching Network

Yasaman Kashefbahrami , Erkut Akdag , Panagiotis Meletis , Evgeniya Balmashnova , Dip Goswami , Egor Bondarau

Authors on Pith no claims yet

Pith reviewed 2026-05-10 20:21 UTC · model grok-4.3

classification 💻 cs.CV cs.LG

keywords point cloud registrationreal-time processingdeep learningrobust matchingindustrial applicationsphotogrammetryevent camerasCAD alignment

0 comments

The pith

R3PM-Net registers point clouds from noisy real scans to CAD models with perfect fitness in 7 milliseconds.

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

The paper presents R3PM-Net as a lightweight deep network for estimating rigid transformations between two point clouds, specifically built to work on imperfect data from industrial sensors instead of only clean synthetic sets. It releases two new datasets of photogrammetric and event-camera scans to test this capability on real artifacts like noise and occlusion. Experiments show the network reaches a fitness score of 1 and very low error while running seven times faster than earlier methods. If the results hold, this would let precise 3D alignment happen in real time for manufacturing or robotics tasks where delays are not acceptable.

Core claim

R3PM-Net is a global-aware object-level point matching network that delivers a fitness score of 1 and inlier RMSE of 0.029 cm on ModelNet40 in 0.007 seconds, approximately seven times faster than RegTR, with comparable accuracy and low latency on the new Sioux-Cranfield dataset and successful handling of edge cases on Sioux-Scans within 50 ms.

What carries the argument

R3PM-Net, a lightweight deep network for global point matching that estimates rigid transformations between real-world point clouds and reference models.

If this is right

Enables real-time 3D registration in industrial inspection and assembly without long processing delays.
Maintains accuracy on incomplete and noisy scans where earlier methods degrade.
Supports direct alignment of event-camera data to digital CAD models.
Reduces per-pair latency to levels usable in live manufacturing pipelines.

Where Pith is reading between the lines

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

The approach could be adapted for non-rigid or dynamic registration in robotics if the core matching step is extended.
Results point to the value of creating more varied real sensor benchmarks to replace reliance on synthetic test sets.
Faster registration may improve downstream tasks such as object tracking and pose estimation in constrained environments.

Load-bearing premise

The Sioux-Cranfield and Sioux-Scans datasets sufficiently capture the range of noise, occlusion, and sensor artifacts in actual industrial photogrammetric and event-camera deployments.

What would settle it

Testing R3PM-Net on additional real-world point cloud pairs from different industrial sensors or environments that produce markedly lower fitness scores or higher RMSE would disprove the claim of broad real-world robustness.

Figures

Figures reproduced from arXiv: 2604.05060 by Dip Goswami, Egor Bondarau, Erkut Akdag, Evgeniya Balmashnova, Panagiotis Meletis, Yasaman Kashefbahrami.

**Figure 2.** Figure 2: Overview of the R3PM-Net Architecture. Built upon [ [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: Coarse-to-Fine Registration approach to handle noisy, sparse scans. Initially, [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: Qualitative registration results of R3PM-Net on real [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

**Figure 5.** Figure 5: 3D Quality Control Setup developed by Sioux Technolo [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗

**Figure 6.** Figure 6: CAD models of the Sioux-Cranfield dataset. The first six belong to the Cranfield Assembly benchmark [ [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗

**Figure 7.** Figure 7: Sioux-Scans point cloud data overview. Target (blue) and Source (yellow) point clouds for seven distinct objects. [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗

read the original abstract

Accurate Point Cloud Registration (PCR) is an important task in 3D data processing, involving the estimation of a rigid transformation between two point clouds. While deep-learning methods have addressed key limitations of traditional non-learning approaches, such as sensitivity to noise, outliers, occlusion, and initialization, they are developed and evaluated on clean, dense, synthetic datasets (limiting their generalizability to real-world industrial scenarios). This paper introduces R3PM-Net, a lightweight, global-aware, object-level point matching network designed to bridge this gap by prioritizing both generalizability and real-time efficiency. To support this transition, two datasets, Sioux-Cranfield and Sioux-Scans, are proposed. They provide an evaluation ground for registering imperfect photogrammetric and event-camera scans to digital CAD models, and have been made publicly available. Extensive experiments demonstrate that R3PM-Net achieves competitive accuracy with unmatched speed. On ModelNet40, it reaches a perfect fitness score of $1$ and inlier RMSE of $0.029$ cm in only $0.007$s, approximately 7 times faster than the state-of-the-art method RegTR. This performance carries over to the Sioux-Cranfield dataset, maintaining a fitness of $1$ and inlier RMSE of $0.030$ cm with similarly low latency. Furthermore, on the highly challenging Sioux-Scans dataset, R3PM-Net successfully resolves edge cases in under 50 ms. These results confirm that R3PM-Net offers a robust, high-speed solution for critical industrial applications, where precision and real-time performance are indispensable. The code and datasets are available at https://github.com/YasiiKB/R3PM-Net.

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

R3PM-Net delivers clear speed gains on ModelNet40 and introduces two public industrial datasets, but the robustness claims rest on unverified coverage of real sensor artifacts in those sets.

read the letter

The paper's headline result is a lightweight network that hits fitness of 1 and 0.029 cm inlier RMSE on ModelNet40 in 7 ms, roughly seven times faster than RegTR, with similar numbers on the new Sioux-Cranfield set and sub-50 ms handling of edge cases on Sioux-Scans. That speed-accuracy combination is the practical takeaway worth noting first. The work adds a global-aware object-level architecture and releases code plus two datasets aimed at photogrammetric and event-camera scans against CAD models. Making those artifacts public is a concrete step that lets others test the same conditions. The quantitative claims are stated directly as empirical measurements rather than derived fits, and the abstract positions the method as an extension of existing deep point-matching lines rather than a wholesale reinvention. That framing keeps expectations realistic. The soft spot is the domain gap. The Sioux sets are the only real-world evidence offered, yet the stress-test note correctly flags that their artifact distribution may not fully capture the noise, occlusion, density changes, and calibration errors typical in industrial deployments. Without more detail on collection protocols or explicit comparisons to other real scans, the generalizability argument stays provisional. The abstract also omits training protocol, ablation tables, and exact evaluation splits, so independent verification of the speed and accuracy numbers requires the full manuscript. This paper is aimed at applied researchers who need fast registration for robotics or inspection pipelines and who can use the released datasets as a starting benchmark. A reader focused on practical deployment would get immediate value from the timing results and the public data. It is coherent on its own terms and shows honest engagement with the speed-versus-robustness trade-off, so it deserves a serious referee. I would send it to review with the expectation that the authors add dataset characterization and more ablation evidence rather than desk-reject it.

Referee Report

2 major / 0 minor

Summary. The paper introduces R3PM-Net, a lightweight global-aware network for object-level point cloud registration (PCR) that targets real-time efficiency and robustness on imperfect real-world data. It proposes two new public datasets (Sioux-Cranfield and Sioux-Scans) for registering noisy/occluded photogrammetric and event-camera scans to CAD models, and reports that the method achieves a fitness score of 1.0 with 0.029 cm inlier RMSE on ModelNet40 in 0.007 s (approximately 7× faster than RegTR), with comparable near-perfect scores and sub-50 ms latency on the new datasets.

Significance. If the empirical results and dataset representativeness hold, the work would supply a practical, deployable PCR solution for industrial settings that require both speed and tolerance to sensor artifacts, while the public release of code and datasets strengthens reproducibility and provides new benchmarks bridging synthetic and real-world regimes.

major comments (2)

[Abstract and dataset section] Abstract and dataset section: the headline robustness claim for industrial photogrammetric/event-camera scenarios rests on Sioux-Cranfield and Sioux-Scans being representative of actual noise, occlusion, density variation, and calibration errors; the manuscript reports aggregate fitness=1 and RMSE≈0.03 cm but supplies no quantitative characterization (e.g., histograms of point density, occlusion ratios, or noise statistics) or external validation against industrial ground-truth distributions, leaving the domain-gap concern unaddressed.
[Experiments section] Experiments section: the reported speed and accuracy numbers are given without the training protocol, data-augmentation details, or ablation studies on the new datasets, so it is impossible to determine whether the perfect fitness scores reflect genuine generalization or favorable test-set construction.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We are grateful to the referee for the constructive feedback on our submission. The comments highlight important aspects for enhancing the manuscript's rigor, and we respond to each major comment in detail below, indicating the revisions we plan to make.

read point-by-point responses

Referee: [Abstract and dataset section] Abstract and dataset section: the headline robustness claim for industrial photogrammetric/event-camera scenarios rests on Sioux-Cranfield and Sioux-Scans being representative of actual noise, occlusion, density variation, and calibration errors; the manuscript reports aggregate fitness=1 and RMSE≈0.03 cm but supplies no quantitative characterization (e.g., histograms of point density, occlusion ratios, or noise statistics) or external validation against industrial ground-truth distributions, leaving the domain-gap concern unaddressed.

Authors: We acknowledge the importance of characterizing the datasets to support the robustness claims. Although the current manuscript focuses on aggregate performance metrics, we will revise the dataset section to include quantitative analyses, such as histograms of point density variations, occlusion ratios, and noise statistics for the Sioux-Cranfield and Sioux-Scans datasets. For external validation, we note that direct comparison to industrial ground-truth distributions is challenging due to the proprietary nature of many industrial datasets; however, the Sioux datasets are derived from real-world photogrammetric and event-camera captures of industrial components, and we will expand the discussion to better contextualize their representativeness. These additions will clarify that the reported performance reflects handling of realistic imperfections rather than idealized conditions. revision: yes
Referee: [Experiments section] Experiments section: the reported speed and accuracy numbers are given without the training protocol, data-augmentation details, or ablation studies on the new datasets, so it is impossible to determine whether the perfect fitness scores reflect genuine generalization or favorable test-set construction.

Authors: We agree that including the training protocol and related details is essential for reproducibility and to demonstrate genuine generalization. In the revised manuscript, we will provide a comprehensive description of the training protocol, data augmentation techniques applied during training on ModelNet40, and ablation studies specifically on the Sioux-Cranfield and Sioux-Scans datasets. These will include variations in network components and their impact on performance, confirming that the high fitness scores are not due to favorable test-set construction but result from the model's design for real-world robustness. revision: yes

Circularity Check

0 steps flagged

No significant circularity in empirical performance claims

full rationale

The paper reports direct empirical measurements of R3PM-Net performance (fitness, inlier RMSE, runtime) on the standard ModelNet40 benchmark and on the newly introduced Sioux-Cranfield and Sioux-Scans datasets. These quantities are obtained by running the trained network and computing standard registration error metrics against ground-truth transformations; they are not quantities derived from a mathematical chain, fitted parameters, or self-referential definitions that reduce to the inputs by construction. No load-bearing self-citations, uniqueness theorems, or ansatzes appear in the reported results or method description.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on empirical evaluation of a proposed neural network on newly introduced datasets; no explicit free parameters, mathematical axioms, or invented physical entities are stated in the abstract.

pith-pipeline@v0.9.0 · 5637 in / 1110 out tokens · 41438 ms · 2026-05-10T20:21:01.812375+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.

R3PM-Net employs a shared Multilayer Perceptron (MLP) of five linear layers with ReLU activations... final global max-pooling operation aggregates these point-wise features
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

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

Sioux-Cranfield and Sioux-Scans datasets... real-world industrial scenarios

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

42 extracted references · 3 canonical work pages · 1 internal anchor

[1]

PointNetLK: Robust & efficient point cloud registration using pointnet

Yasuhiro Aoki, Hunter Goforth, Rangaprasad Arun Srivatsan, and Simon Lucey. PointNetLK: Robust & efficient point cloud registration using pointnet. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7163–7172, 2019. 1

2019
[2]

Least-squares fitting of two 3-D point sets.IEEE Transactions on Pattern Analysis and Machine Intelligence, PAMI-9(5): 698–700, 1987

K S Arun, T S Huang, and S D Blostein. Least-squares fitting of two 3-D point sets.IEEE Transactions on Pattern Analysis and Machine Intelligence, PAMI-9(5): 698–700, 1987. 3, 4

1987
[3]

Near-linear time algo- rithm for the chamfer distance.Advances in Neural In- formation Processing Systems, 36:66833–66844, 2023

Ainesh Bakshi, Piotr Indyk, Rajesh Jayaram, Sandeep Silwal, and Erik Waingarten. Near-linear time algo- rithm for the chamfer distance.Advances in Neural In- formation Processing Systems, 36:66833–66844, 2023. 3

2023
[4]

A method for registration of 3-D shapes.IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239–256, 1992

P J Besl and Neil D McKay. A method for registration of 3-D shapes.IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239–256, 1992. 1, 2

1992
[5]

A review of rigid point cloud registration based on deep learning.Frontiers in Neu- rorobotics, 17, 2024

Lei Chen, Changzhou Feng, Yunpeng Ma, Yikai Zhao, and Chaorong Wang. A review of rigid point cloud registration based on deep learning.Frontiers in Neu- rorobotics, 17, 2024. 2

2024
[6]

The devel- opment of a european benchmark for the comparison of assembly robot programming systems

K Collins, AJ Palmer, and K Rathmill. The devel- opment of a european benchmark for the comparison of assembly robot programming systems. InRobot Technology and Applications: Proceedings of the 1st Robotics Europe Conference Brussels, June 27–28, 1984, pages 187–199. Springer, 1985. 5, 1, 2

1984
[7]

Compari- son of point cloud gegistration techniques on scanned physical objects.Sensors, 24(7), 2024

Menthy Denayer, Joris De Winter, Evandro Bernardes, Bram Vanderborght, and Tom Verstraten. Compari- son of point cloud gegistration techniques on scanned physical objects.Sensors, 24(7), 2024. 2, 5

2024
[8]

PPFNet: Global context aware local features for ro- bust 3D point matching

Haowen Deng, Tolga Birdal, and Slobodan Ilic. PPFNet: Global context aware local features for ro- bust 3D point matching. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 195–205, 2018. 2

2018
[9]

Robust and correspondence-free point cloud registration: An extended approach with multiple hypotheses evaluation.Intelligent Service Robotics, 17 (6):1109–1124, 2024

Federica Di Lauro, Domenico Giorgio Sorrenti, and Si- mone Fontana. Robust and correspondence-free point cloud registration: An extended approach with multiple hypotheses evaluation.Intelligent Service Robotics, 17 (6):1109–1124, 2024. 3

2024
[10]

Random sam- ple consensus.Communications of the ACM, 24(6): 381–395, 1981

Martin A Fischler and Robert C Bolles. Random sam- ple consensus.Communications of the ACM, 24(6): 381–395, 1981. 1, 2

1981
[11]

A benchmark for point clouds registration algorithms

Simone Fontana, Daniele Cattaneo, Augusto L Bal- lardini, Matteo Vaghi, and Domenico G Sorrenti. A benchmark for point clouds registration algorithms. Robotics and Autonomous Systems, 140:103734, 2021. 2, 3

2021
[12]

Cloud compare

Daniel Girardeau-Montaut et al. Cloud compare. France: EDF R&D Telecom ParisTech, 11(5):2016,

2016
[13]

AliceVision Mesh- room

Carsten Griwodz, Simone Gasparini, Lilian Calvet, Pierre Gurdjos, Fabien Castan, Benoit Maujean, Gre- goire De Lillo, and Yann Lanthony. AliceVision Mesh- room. InProceedings of the 12th ACM Multimedia Systems Conference, pages 241–247, New York, NY , USA, 2021. ACM. 5

2021
[14]

Predator: Regis- tration of 3D point clouds with low overlap

Shengyu Huang, Zan Gojcic, Mikhail Usvyatsov, An- dreas Wieser, and Konrad Schindler. Predator: Regis- tration of 3D point clouds with low overlap. InProceed- ings of the IEEE/CVF Conference on computer vision and pattern recognition, pages 4267–4276, 2021. 2, 3, 5, 6, 7, 4

2021
[15]

A comprehensive survey on point cloud regis- tration.arXiv preprint arXiv:2103.02690, 2021

Xiaoshui Huang, Guofeng Mei, Jian Zhang, and Rana Abbas. A comprehensive survey on point cloud regis- tration.arXiv preprint arXiv:2103.02690, 2021. 2

work page arXiv 2021
[16]

A tutorial review on point cloud registrations: principle, classifi- cation, comparison, and technology challenges, 2021

Leihui Li, Riwei Wang, and Xuping Zhang. A tutorial review on point cloud registrations: principle, classifi- cation, comparison, and technology challenges, 2021. 1

2021
[17]

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, 2017. 2, 3, 6

2017
[18]

Geometric transformer for fast and robust point cloud registration

Zheng Qin, Hao Yu, Changjian Wang, Yulan Guo, Yux- ing Peng, and Kai Xu. Geometric transformer for fast and robust point cloud registration. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11143–11152, 2022. 2, 3, 5, 6, 7, 4

2022
[19]

Learning3D, 2020

Vinit Sarode and Hojun Lee. Learning3D, 2020. 6

2020
[20]

Generalized-ICP

Aleksandr V Segal, Dirk Haehnel, and Sebastian Thrun. Generalized-ICP. InRobotics: Science and Systems,
[21]

A relationship between arbitrary positive matrices and doubly stochastic matrices.The Annals of Mathematical Statistics, 35(2), 1964

Richard Sinkhorn. A relationship between arbitrary positive matrices and doubly stochastic matrices.The Annals of Mathematical Statistics, 35(2), 1964. 3, 4

1964
[22]

Logdesc: Local geometric features aggrega- tion for robust point cloud registration

Karim Slimani, Brahim Tamadazte, and Catherine Achard. Logdesc: Local geometric features aggrega- tion for robust point cloud registration. InProceedings of the Asian Conference on Computer Vision, pages 1952–1968, 2024. 2, 5, 6, 7, 4

1952
[23]

Multi-view convolutional neural networks for 3D shape recognition

Hang Su, Subhransu Maji, Evangelos Kalogerakis, and Erik Learned-Miller. Multi-view convolutional neural networks for 3D shape recognition. InProceedings of 9 the IEEE international conference on computer vision, pages 945–953, 2015. 2

2015
[24]

Optimization of production by 3DoP exhibition work package 4: Automated, ded- icated am production line for affordable 3D printed dental implants and aligners

Sioux Technologies. Optimization of production by 3DoP exhibition work package 4: Automated, ded- icated am production line for affordable 3D printed dental implants and aligners. Technical report, Sioux Technologies, 2023. 5, 1

2023
[25]

Kpconv: Flexible and deformable convolution for point clouds

Hugues Thomas, Charles R Qi, Jean-Emmanuel De- schaud, Beatriz Marcotegui, Fran c ¸ois Goulette, and Leonidas J Guibas. Kpconv: Flexible and deformable convolution for point clouds. InProceedings of the IEEE/CVF international conference on computer vi- sion, pages 6411–6420, 2019. 2

2019
[26]

PRNet: Self-supervised learning for partial-to-partial registration

Y Wang and JM Solomon. PRNet: Self-supervised learning for partial-to-partial registration. arxiv 2019. arXiv preprint arXiv:1910.12240, 1910. 3, 5, 1, 2

work page arXiv 2019
[27]

Deep closest point: Learning representations for point cloud registration

Yue Wang and Justin M Solomon. Deep closest point: Learning representations for point cloud registration. In Proceedings of the IEEE/CVF international conference on computer vision, pages 3523–3532, 2019. 1, 3, 5, 2

2019
[28]

Dy- namic graph CNN for learning on point clouds.ACM Transactions on Graphics (tog), 38(5):1–12, 2019

Yue Wang, Yongbin Sun, Ziwei Liu, Sanjay E Sarma, Michael M Bronstein, and Justin M Solomon. Dy- namic graph CNN for learning on point clouds.ACM Transactions on Graphics (tog), 38(5):1–12, 2019. 2

2019
[29]

Point cloud utils, 2022

Francis Williams. Point cloud utils, 2022. 3

2022
[30]

Point transformer v3: Simpler faster stronger

Xiaoyang Wu, Li Jiang, Peng-Shuai Wang, Zhijian Liu, Xihui Liu, Yu Qiao, Wanli Ouyang, Tong He, and Hengshuang Zhao. Point transformer v3: Simpler faster stronger. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 4840–4851, 2024. 2

2024
[31]

3D shapenets: A deep representation for volumetric shapes

Zhirong Wu, Shuran Song, Aditya Khosla, Fisher Yu, Linguang Zhang, Xiaoou Tang, and Jianxiong Xiao. 3D shapenets: A deep representation for volumetric shapes. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 1912– 1920, 2015. 5, 6, 1

1912
[32]

Object modelling by registration of multiple range images.Image and Vision Computing, 10(3), 1992

Chen Yang and G ´erard Medioni. Object modelling by registration of multiple range images.Image and Vision Computing, 10(3), 1992. 2

1992
[33]

Pare-net: Position-aware rotation- equivariant networks for robust point cloud registration

Runzhao Yao, Shaoyi Du, Wenting Cui, Canhui Tang, and Chengwu Yang. Pare-net: Position-aware rotation- equivariant networks for robust point cloud registration. InEuropean Conference on Computer Vision, pages 287–303. Springer, 2024. 2

2024
[34]

RPM-Net: Robust point matching using learned features

Zi Jian Yew and Gim Hee Lee. RPM-Net: Robust point matching using learned features. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11824–11833, 2020. 1, 2, 3, 4, 5, 6, 7

2020
[35]

Zi Jian Yew and Gim Hee Lee. RegTR: End-to-end point cloud correspondences with transformers.Pro- ceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2022-June: 6667–6676, 2022. 1, 2, 3, 5, 6, 7, 4

2022
[36]

Peal: Prior-embedded explicit attention learning for low-overlap point cloud registra- tion, 2023

Junle Yu, Luwei Ren, Yu Zhang, Wenhui Zhou, Lili Lin, and Guojun Dai. Peal: Prior-embedded explicit attention learning for low-overlap point cloud registra- tion, 2023. 3

2023
[37]

3DMatch: Learning local geometric descriptors from rgb-d reconstructions

Andy Zeng, Shuran Song, Matthias Nießner, Matthew Fisher, Jianxiong Xiao, and Thomas Funkhouser. 3DMatch: Learning local geometric descriptors from rgb-d reconstructions. InProceedings - 30th IEEE Con- ference on Computer Vision and Pattern Recognition, CVPR 2017, 2017. 2

2017
[38]

Fastmac: Stochastic spectral sampling of cor- respondence graph

Yifei Zhang, Hao Zhao, Hongyang Li, and Siheng Chen. Fastmac: Stochastic spectral sampling of cor- respondence graph. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion (CVPR), pages 17857–17867, 2024. 3

2024
[39]

A comprehensive survey and taxonomy on point cloud registration based on deep learning,

Yu-Xin Zhang, Jie Gui, Xiaofeng Cong, Xin Gong, and Wenbing Tao. A comprehensive survey and taxonomy on point cloud registration based on deep learning,
[40]

Deep learning based point cloud registration: an overview

Zhiyuan Zhang, Yuchao Dai, and Jiadai Sun. Deep learning based point cloud registration: an overview. Virtual Reality and Intelligent Hardware, 2(3):222–246,
[41]

Open3D: A Modern Library for 3D Data Processing

QY Zhou, J Park, and V Koltun. Open3D: A mod- ern library for 3D data processing.arXiv preprint arXiv:1801.09847, 1801. 6, 3

work page internal anchor Pith review arXiv
[42]

Deep learning representation using autoencoder for 3D shape retrieval.Neurocomputing, 204:41–50, 2016

Zhuotun Zhu, Xinggang Wang, Song Bai, Cong Yao, and Xiang Bai. Deep learning representation using autoencoder for 3D shape retrieval.Neurocomputing, 204:41–50, 2016. 2 10 R3PM-Net: Real-time, Robust, Real-world Point Matching Network Supplementary Material A.1. Datasets ModelNet40 Dataset.[ 31] is a collection of synthetic 3D CAD models from 40 object cat...

2016