Pith. sign in

REVIEW 3 major objections 4 minor 48 references

Accelerating Outlier-robust Rotation Estimation by Stereographic Projection

T0 review · 3 major / 4 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read Rotation estimation reduces to 2D voting via stereographic projection, with 0.07-second 90%-outlier solutions at 0.01-degree error.

desk verdict The stereographic voting idea is clean and the speed numbers are striking, but the paper never resolves the r/-r sign ambiguity, so the algorithm as written cannot distinguish a rotation from its inverse. read the letter →

arxiv 2502.06337 v1 pith:3GMGCSV3 submitted 2025-02-10 cs.CV cs.RO

Taosi Xu , Yinlong Liu , Xianbo Wang , Zhi-Xin Yang This is my paper · ORCID
classification cs.CVcs.RO
keywords rotationestimationstereographicprojectionoutlierrobustnessspatialvotingpointcloudregistrationmulti-modelfittingaxisandanglemaximumconsensus
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to make outlier-robust rotation estimation fast enough for large real-world inputs by splitting the problem in two. It shows geometrically that every correspondence between source and target points confines the rotation axis to a circle on the unit sphere, so the correct axis is the point where the most circles intersect. Stereographic projection maps those circles to circles in the plane, reducing the axis search to a 2D spatial vote; the angle is then found by a separate one-dimensional histogram vote, and Rodrigues' formula assembles the rotation. The same voting landscape yields multiple rotations at once by reading several peaks. With GPU acceleration, the authors report solving problems with $10^{6}$ correspondences and 90% outliers in about 0.07 seconds with an angular error of 0.01 degrees.

What carries the argument

The central object is the stereographic projection from the unit sphere minus a projection point onto the plane, a conformal map that sends circles on the sphere to circles (or lines) in the plane. It does the work of converting the 3D search for a common intersection of great circles into a planar maximum-intersection problem. The second piece of machinery is a discretized 2D accumulator that counts, for each cell, how many projected constraint circles pass through it; the highest-count cell is back-projected to $S^{2}$ to give the rotation axis. A companion 1D accumulator votes on the angle recovered from each correspondence, and reading the 2D accumulator at multiple local maxima turns single-model estimation into multi-model estimation.

What would settle it

Take a single synthetic problem with a known rotation R and $10^{6}$ correspondences at 90% outliers, run the algorithm, and compare the recovered rotation to both R and $R^{{-1}}$; if the method returns the inverse on a nontrivial fraction of trials, the unstated r-versus-r ambiguity is real and the 0.01-degree accuracy claim depends on an auxiliary tie-breaking rule rather than the described pipeline. A cleaner check is to replace the ground-truth rotation by $R^{{-1}}$ and verify the output flips accordingly.

Watch

Extended reading notes

Core claim

The central discovery is that the axis part of rotation search—usually a 3-degree-of-freedom search over SO(3)—can be decoupled and flattened. For each inlier correspondence the rotation constraint implies r^T(x_i - y_i) = 0, meaning the axis r lies on a great circle on the unit sphere. Because stereographic projection is circle-preserving, the family of these constraint circles becomes a family of circles in the plane, and the axis that satisfies the most correspondences is the point where the most projected circles cross. A discretized 2D accumulator finds that point; back-projection gives the axis, a one-dimensional histogram over per-correspondence angles gives the rotation angle, and Rodrigues' formula reconstructs R. The same accumulator, interpreted at multiple local maxima, provides multiple rotations simultaneously.

Load-bearing premise

The method's success hinges on the rotation axis being recoverable as the unique densest point of intersection of the projected constraint circles; that requires an unstated rule for choosing between r and -r, because both axes satisfy every pairwise constraint and produce the same unsigned angle, yet correspond to different rotations.

Editorial extensions

If this is right

  • If the claim holds, rotation estimation with millions of correspondences and heavy mismatch rates becomes a sub-second operation, making it practical for online autonomous driving and large-scale 3D registration.
  • The axis-angle decoupling plus voting removes the need for a good initialization, a limitation of ICP-style refinement, and avoids the exponential traversal of branch-and-bound rotation search.
  • Because the voting happens in a fixed 2D grid, the algorithm's cost scales roughly linearly with the number of correspondences, and the same accumulator can be reused to find multiple rotations without running a separate consensus loop per model.
  • On the standard indoor and outdoor benchmarks used in the paper, the method reports lower rotation error than the compared baselines at much lower runtime, suggesting the speed does not come at the cost of accuracy.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper leaves the antipodal ambiguity between axes r and -r implicit: every constraint r^T(x_i - y_i) = 0 and the arccos-based angle vote are unchanged by flipping the axis, yet the two choices give inverse rotations. A reader should treat the sign convention as an unverified detail and test whether the method occasionally returns R^{-1}.
  • The multi-rotation experiments assume the number of rotations is known or visible as peaks; an immediate extension is automatic peak detection and model-order selection from the same 2D accumulator, which the paper does not develop.
  • Because the geometric reduction only needs constraints that define circles on a sphere, the same stereographic-voting pipeline could plausibly extend to other consensus problems with circle-valued constraints, such as 1D subspace estimation or homography search; this is an extrapolation, not a claim of the paper.
  • The reported 0.07-second runtime is for GPU; a useful stress test would be an ablation of grid resolution and projection point to see how error and runtime trade off, since the current experiments fix those choices.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper presents a rotation estimation method (AORESP) that decouples the rotation into axis and angle. It derives a geometric constraint r^T(x-y)=0, maps the resulting circles on the unit sphere to a 2D plane via stereographic projection, and uses spatial voting in a 2D accumulator to find the rotation axis. The angle is then recovered by histogram voting of per-correspondence angles computed with Rodrigues' formula. The authors claim state-of-the-art accuracy and speed, including solving 10^6-point problems with 90% outliers in 0.07 s with 0.01° error, and also claim simultaneous estimation of multiple rotations. Experiments are reported on synthetic data, 3DMatch, KITTI, and a custom multi-rotation setup.

Significance. If the algorithmic description were complete and the reported numbers reproducible, the method would be a significant practical advance: the per-correspondence geometric constraint and the stereographic projection are standard tools, but the proposed voting formulation is an interesting way to turn 3D rotation search into two lower-dimensional voting problems. The claimed speed and robustness at 90% outlier rates, together with the extension to multiple rotations, would exceed the reported performance of the compared baselines. However, the paper as written omits several load-bearing implementation details, and one mathematical degeneracy—the r/-r sign ambiguity—is not addressed. These issues currently prevent the central claims from being verified or reproduced.

major comments (3)
  1. [Algorithm 1, line 8] The algorithm does not resolve the antipodal sign ambiguity between r and -r. The axis voting objective in Eq. (5) uses |r^T(x_i-y_i)|, which is invariant under r→-r, and stereographic projection maps both antipodal points to distinct but equally valid intersections. In the angle step, line 15 computes angle_i = arccos(β_i·γ_i/(||β_i||||γ_i||)); replacing r by -r flips both β_i and γ_i, leaving the dot product and the voted angle unchanged. Therefore R(r,θ) and R(-r,θ)=R(r,-θ) receive exactly the same votes for θ∈[0,π], and Algorithm 1 has no step that chooses between them. For a generic rotation with θ not equal to 0 or π, outputting the -r branch yields the inverse rotation, which would produce an error near 180° rather than the reported 0.01°. The authors must either describe the disambiguation step explicitly (for example, a residual check) or change the angle search to a signed angle over a 2π range.
  2. [Algorithm 1, line 8] The sampling formula reads Points3d = α1 cos θj + α2 cos θj, which is a one-dimensional curve, not a circle. This must be α1 cos θj + α2 sin θj for the points to trace a circle on the sphere. As written, the stereographic projection step would not produce the circle intersections the method relies on, and the axis voting would be degenerate. This appears to be a typographical error, but it is a load-bearing one because the entire axis-finding stage depends on sampling circles correctly.
  3. [Sections IV-A, IV-B, and V] Several critical parameters are never specified or analyzed: the resolution of the 2D accumulator, the number of angle samples J for θ, the bin width of the rotation-angle histogram, and the noise tolerance ε in Eq. (5). The reported 0.01° accuracy and 0.07 s runtime cannot be reproduced without these values, and there is no sensitivity study in Section V. This is not a minor omission because the voting method's robustness and efficiency both depend directly on these discretizations, and the claimed superiority over baselines may change with different settings.
minor comments (4)
  1. [Section IV, Eq. (10)] The text has a grammatical error: 'the intersection an plane in S^2' should be 'the intersection of a plane with S^2'. Also, the displayed equation for the projected circle appears incomplete; the right-hand side should probably be (A^2+B^2)/(C+D)^2 + (C-D)/(C+D) with appropriate parentheses.
  2. [Section III-A and Algorithm 1, line 4] The normalization of z_i is inconsistent: Section III-A defines z = (x-y)/||x-y||, while Algorithm 1 sets z_i = y_i - x_i and only then asserts a_i^2+b_i^2+c_i^2=1. The sign and normalization used in the actual voting step should be stated explicitly.
  3. [Section V-A-1] The comparison with TEASER++ and GORE is only described in prose ('memory issues' and 'running time exceeds 12 hours'); no data or plots are shown for these baselines, which makes the head-to-head comparison incomplete.
  4. [Section IV-D] The multi-rotation extension is described only conceptually. The paper does not explain how peaks are selected in the 2D accumulator, how many peaks are assumed, or how the per-peak angle histograms are computed. Without these details, the multi-model experiment in Section V-B cannot be reproduced.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the core axis-angle derivation is self-contained, with only a non-circular antipodal-ambiguity correctness gap.

full rationale

The paper's central derivation chain is self-contained and does not reduce to its own inputs by construction. The axis constraint r^T(x-y)=0 is obtained algebraically from the defining property r^T R = r^T applied to Rx=y (Eqs. 3-4), not assumed as the result. The rotation axis is then found by stereographic projection and spatial voting over circles that encode this constraint, and the rotation angle is computed from Rodrigues' formula per correspondence and aggregated by histogram voting; none of these steps fits a parameter to the target rotation or renames a fitted value as a prediction. The reported experiments are comparisons against external baselines on synthetic and real data, and the only possibly same-author citation ([32], a branch-and-bound dissertation) is used as an example of exhaustive search, not as a load-bearing premise. The unresolved r/-r antipodal ambiguity is a genuine correctness/completeness flaw in the algorithm as written, but it is not circular reasoning: it concerns whether the method uniquely determines the rotation, not whether the derivation is equivalent to its inputs. No circular step could be identified by quoting a specific equation or self-citation that makes the claimed output equal to the input by construction.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The method relies only on standard geometry and voting assumptions; the main hidden degrees of freedom are the discretization parameters, which are not disclosed.

free parameters (4)
  • 2D accumulator resolution = not specified
    The number of cells in the planar voting grid determines the angular resolution of the estimated axis; to achieve 0.01 degrees, the resolution must be very fine, but the paper does not state it.
  • Number of angle samples theta_j = not specified
    Algorithm 1 discretizes [-pi, pi] into J angles for each correspondence to sample the axis circle; J affects both accuracy and runtime.
  • Rotation angle histogram bin width = not specified
    The final angle is chosen as the peak of a histogram; the bin width affects angle accuracy.
  • Noise tolerance epsilon = not specified
    Equation (5) defines the objective with a threshold epsilon, but the voting algorithm does not use an explicit epsilon, so its role is unclear.
assumptions (3)
  • domain assumption For an inlier correspondence (x,y) with y = Rx, the rotation axis r satisfies r^T(x-y) = 0.
    This follows from r^T R = r^T for a rotation about r; it is the geometric constraint used throughout Section III.
  • standard math Stereographic projection maps circles on the unit sphere to circles or lines on the plane.
    This is a classical property of stereographic projection, stated in Definition 1.
  • domain assumption The true rotation axis is the point where the greatest number of projected circles intersect, and outlier-generated circles do not form a competing peak.
    The spatial voting strategy assumes that inliers form a dominant consensus peak, which is reasonable for random outliers but not proven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Accelerating Outlier-robust Rotation Estimation by Stereographic Projection." pith.science (2026). https://pith.science/paper/3GMGCSV3

@misc{pith2026250206337,
  author       = {Pith},
  title        = {Pith review of: Accelerating Outlier-robust Rotation Estimation by Stereographic Projection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3GMGCSV3}},
  note         = {Machine review of arXiv:2502.06337}
}
abstract

Rotation estimation plays a fundamental role in many computer vision and robot tasks. However, efficiently estimating rotation in large inputs containing numerous outliers (i.e., mismatches) and noise is a recognized challenge. Many robust rotation estimation methods have been designed to address this challenge. Unfortunately, existing methods are often inapplicable due to their long computation time and the risk of local optima. In this paper, we propose an efficient and robust rotation estimation method. Specifically, our method first investigates geometric constraints involving only the rotation axis. Then, it uses stereographic projection and spatial voting techniques to identify the rotation axis and angle. Furthermore, our method efficiently obtains the optimal rotation estimation and can estimate multiple rotations simultaneously. To verify the feasibility of our method, we conduct comparative experiments using both synthetic and real-world data. The results show that, with GPU assistance, our method can solve large-scale ($10^6$ points) and severely corrupted (90\% outlier rate) rotation estimation problems within 0.07 seconds, with an angular error of only 0.01 degrees, which is superior to existing methods in terms of accuracy and efficiency.

Figures

Figures reproduced from arXiv: 2502.06337 by the authors.

Figure 1
Figure 1. (a) Given the i-th z i , all possible rotation axes are in a circle (a green dashed circle). (b) Given z 1 and z 2, all possible rotation axes should be in the intersections of two corresponding circles. (c) The stereographic projection is circle-preserving, and a 3D circle on the sphere is into a 2D circle when projected onto the plane. (d) The intersections of 3D circles are projected to the intersections of 2D ci… view at source ↗
Figure 2
Figure 2. Diagram of rotation motion, illustrating that any [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. (a) With the perfect inputs, many project circles are across the same point, corresponding to the rotation axis. (b) With [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Comparative experiments. The left calculates the rotation errors using inputs with different outlier rates when the total [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Comparative experiments. The left calculates the running time using inputs with different outlier rates when the total [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Comparative experiments between our algorithm and Sequential RANSAC for multiple rotation estimations. The first [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: A comparative experiment is conducted between [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: An example from our collected data. Red circles [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 39 canonical work pages

  1. [1]

    Fast rotation search with stereographic projections for 3d registration,

    A. P. Bustos, T.-J. Chin, A. Eriksson, H. Li, and D. Suter, “Fast rotation search with stereographic projections for 3d registration,” IEEE transactions on pattern analysis and machine intelligence , vol. 38, no. 11, pp. 2227–2240, 2016

    work page 2016

  2. [2]

    A comprehensive survey on point cloud registration,

    X. Huang, G. Mei, J. Zhang, and R. Abbas, “A comprehensive survey on point cloud registration,” arXiv preprint arXiv:2103.02690 , 2021

    arXiv 2021

  3. [3]

    Szeliski, Computer vision: algorithms and applications

    R. Szeliski, Computer vision: algorithms and applications . Springer Nature, 2022

    2022

  4. [4]

    Phaser: A robust and correspondence-free global pointcloud registration,

    L. Bernreiter, L. Ott, J. Nieto, R. Siegwart, and C. Cadena, “Phaser: A robust and correspondence-free global pointcloud registration,” IEEE Robotics and Automation Letters , vol. 6, no. 2, pp. 855–862, 2021

    work page 2021

  5. [5]

    Guaranteed outlier removal for rota- tion search,

    A. Parra Bustos and T.-J. Chin, “Guaranteed outlier removal for rota- tion search,” in Proceedings of the IEEE International Conference on Computer Vision, 2015, pp. 2165–2173

    work page 2015

  6. [6]

    Point reg- istration via efficient convex relaxation,

    H. Maron, N. Dym, I. Kezurer, S. Kovalsky, and Y . Lipman, “Point reg- istration via efficient convex relaxation,” ACM Transactions on Graphics (TOG), vol. 35, no. 4, pp. 1–12, 2016

    work page 2016

  7. [7]

    Rosia: Rotation-search-based star identification algorithm,

    C.-K. Chng, A. P. Bustos, B. McCarthy, and T.-J. Chin, “Rosia: Rotation-search-based star identification algorithm,” IEEE Transactions on Aerospace and Electronic Systems , 2023

    work page 2023

  8. [8]

    Quantum multi-model fitting,

    M. Farina, L. Magri, W. Menapace, E. Ricci, V . Golyanik, and F. Ar- rigoni, “Quantum multi-model fitting,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 13 640–13 649

    work page 2023

Show all 48 references
  1. [9]

    Multi-model 3d registration: Finding multiple moving objects in cluttered point clouds,

    D. Jin, S. Karmalkar, H. Zhang, and L. Carlone, “Multi-model 3d registration: Finding multiple moving objects in cluttered point clouds,” arXiv preprint arXiv:2402.10865 , 2024

    2024 arXiv

  2. [10]

    Arcs: Accurate rotation and correspondence search,

    L. Peng, M. C. Tsakiris, and R. Vidal, “Arcs: Accurate rotation and correspondence search,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 11 153–11 163

    2022

  3. [11]

    M ¨obius transformations and stereographic projection,

    D. R. Wilkins, “M ¨obius transformations and stereographic projection,” 2017

    2017

  4. [12]

    Chin and D

    T.-J. Chin and D. Suter, The maximum consensus problem: recent algorithmic advances. Springer Nature, 2022

    2022

  5. [13]

    On the tightness of semidefinite relaxations for rotation estimation,

    L. Brynte, V . Larsson, J. P. Iglesias, C. Olsson, and F. Kahl, “On the tightness of semidefinite relaxations for rotation estimation,” Journal of Mathematical Imaging and Vision , pp. 1–11, 2022

    2022

  6. [14]

    Robust rotation search in computer vision,

    ´A. J. P. Bustos, “Robust rotation search in computer vision,” Ph.D. dissertation, University of Adelaide, School of Computer Science, 2016

    2016

  7. [15]

    Speeding up iterative closest point using stochastic gradient descent,

    F. A. Maken, F. Ramos, and L. Ott, “Speeding up iterative closest point using stochastic gradient descent,” in 2019 International Conference on Robotics and Automation (ICRA) . IEEE, 2019, pp. 6395–6401

    2019

  8. [16]

    Heuristic- based laser scan matching for outdoor 6d slam,

    A. N ¨uchter, K. Lingemann, J. Hertzberg, and H. Surmann, “Heuristic- based laser scan matching for outdoor 6d slam,” in Annual Conference on Artificial Intelligence . Springer, 2005, pp. 304–319

    2005

  9. [17]

    Deep hough voting for robust global registration,

    J. Lee, S. Kim, M. Cho, and J. Park, “Deep hough voting for robust global registration,” 2021

    2021

  10. [18]

    Consac: Robust multi-model fitting by conditional sample consensus,

    F. Kluger, E. Brachmann, H. Ackermann, C. Rother, M. Y . Yang, and B. Rosenhahn, “Consac: Robust multi-model fitting by conditional sample consensus,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2020, pp. 4634–4643

    2020

  11. [19]

    A method for registration for 3-d shapes,

    P. Besl and N. Mckay, “A method for registration for 3-d shapes,” 1992

    1992

  12. [20]

    Numerical optimization,

    S. J. Wright, “Numerical optimization,” 2006

    2006

  13. [21]

    Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography,

    M. A. Fischler and R. C. Bolles, “Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography,” Communications of the ACM, vol. 24, no. 6, pp. 381–395, 1981

    1981

  14. [22]

    Fast global registration,

    Q. Y . Zhou, J. Park, and V . Koltun, “Fast global registration,” in European Conference on Computer Vision , 2016

    2016

  15. [23]

    Deep global registration,

    C. Choy, W. Dong, and V . Koltun, “Deep global registration,” IEEE, 2020

    2020

  16. [24]

    Learning to find good correspondences,

    K. M. Yi, E. Trulls, Y . Ono, V . Lepetit, M. Salzmann, and P. Fua, “Learning to find good correspondences,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 2666– 2674

    2018

  17. [25]

    3dregnet: A deep neural network for 3d point registration,

    G. D. Pais, S. Ramalingam, V . M. Govindu, J. C. Nascimento, R. Chel- lappa, and P. Miraldo, “3dregnet: A deep neural network for 3d point registration,” 2019

    2019

  18. [26]

    Pointdsc: Robust point cloud registration using deep spatial con- sistency,

    X. Bai, Z. Luo, L. Zhou, H. Chen, L. Li, Z. Hu, H. Fu, and C. L. Tai, “Pointdsc: Robust point cloud registration using deep spatial con- sistency,” 2021

    2021

  19. [27]

    Detarnet: Decoupling translation and rotation by siamese network for point cloud registration,

    Z. Chen, F. Yang, and W. Tao, “Detarnet: Decoupling translation and rotation by siamese network for point cloud registration,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 36, no. 1, 2022, pp. 401–409

    2022

  20. [28]

    Deep hough voting for 3d object detection in point clouds,

    C. R. Qi, O. Litany, K. He, and L. J. Guibas, “Deep hough voting for 3d object detection in point clouds,” in proceedings of the IEEE/CVF International Conference on Computer Vision , 2019, pp. 9277–9286

    2019

  21. [29]

    The multiransac algorithm and its application to detect planar homographies,

    M. Zuliani, C. S. Kenney, and B. Manjunath, “The multiransac algorithm and its application to detect planar homographies,” in IEEE International Conference on Image Processing 2005, vol. 3. IEEE, 2005, pp. III–153

    2005

  22. [30]

    T-linkage: A continuous relaxation of j- linkage for multi-model fitting,

    L. Magri and A. Fusiello, “T-linkage: A continuous relaxation of j- linkage for multi-model fitting,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2014, pp. 3954–3961

    2014

  23. [31]

    Progressive-x: Efficient, anytime, multi- model fitting algorithm,

    D. Barath and J. Matas, “Progressive-x: Efficient, anytime, multi- model fitting algorithm,” in Proceedings of the IEEE/CVF international conference on computer vision , 2019, pp. 3780–3788

    2019

  24. [32]

    Globally optimal solutions for unit-norm constrained computer vision problems,

    Y . Liu, “Globally optimal solutions for unit-norm constrained computer vision problems,” Ph.D. dissertation, Technische Universit ¨at M ¨unchen, 2022

    2022

  25. [33]

    Algorithms for the weighted orthogonal procrustes problem and other least squares problems,

    T. Viklands, “Algorithms for the weighted orthogonal procrustes problem and other least squares problems,” Ph.D. dissertation, Datavetenskap, 2006

    2006

  26. [34]

    B. Li. (2017) Math 2220 notes: Section 2.1. [On- line]. Available: https://pi.math.cornell.edu/ ∼boyang/2220%20s2017/ math2220 notes/notes sec 2.1.pdf

    2017

  27. [35]

    Teaser: Fast and certifiable point cloud registration,

    H. Yang, J. Shi, and L. Carlone, “Teaser: Fast and certifiable point cloud registration,” IEEE, no. 2, 2021

    2021

  28. [36]

    Efficient single correspondence voting for point cloud registration,

    X. Xing, Z. Lu, Y . Wang, and J. Xiao, “Efficient single correspondence voting for point cloud registration,” IEEE Transactions on Image Pro- cessing, 2024

    2024

  29. [37]

    Deep closest point: Learning representa- tions for point cloud registration,

    Y . Wang and J. M. Solomon, “Deep closest point: Learning representa- tions for point cloud registration,” 2019

    2019

  30. [38]

    Pointnetlk: Robust & efficient point cloud registration using pointnet,

    Y . Aoki, H. Goforth, R. A. Srivatsan, and S. Lucey, “Pointnetlk: Robust & efficient point cloud registration using pointnet,” 2019

    2019

  31. [39]

    A spectral technique for correspon- dence problems using pairwise constraints,

    Leordeanu, M., Hebert, and M., “A spectral technique for correspon- dence problems using pairwise constraints,” in Computer Vision, 2005. ICCV 2005. Tenth IEEE International Conference on , 2005

    2005

  32. [40]

    Graph-cut ransac,

    D. Barath and J. Matas, “Graph-cut ransac,” in 2018 IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, 2018, pp. 6733–6741

    2018

  33. [41]

    Compatibility-guided sampling consensus for 3-d point cloud registration,

    S. Quan and J. Yang, “Compatibility-guided sampling consensus for 3-d point cloud registration,” IEEE Transactions on Geoscience and Remote Sensing, vol. PP, no. 99, pp. 1–13, 2020

    2020

  34. [42]

    Sc2-pcr: A second order spatial compatibility for efficient and robust point cloud registration,

    Z. Chen, K. Sun, F. Yang, and W. Tao, “Sc2-pcr: A second order spatial compatibility for efficient and robust point cloud registration,” in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 13 211–13 221

    2022

  35. [43]

    Guaranteed outlier removal for point cloud registration with correspondences,

    A. P. Bustos and T.-J. Chin, “Guaranteed outlier removal for point cloud registration with correspondences,” IEEE transactions on pattern analysis and machine intelligence, vol. 40, no. 12, pp. 2868–2882, 2017

    2017

  36. [44]

    3dmatch: Learning local geometric descriptors from rgb-d reconstruc- tions,

    A. Zeng, S. Song, M. Nießner, M. Fisher, J. Xiao, and T. Funkhouser, “3dmatch: Learning local geometric descriptors from rgb-d reconstruc- tions,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 1802–1811

    2017

  37. [45]

    Are we ready for autonomous driving? the kitti vision benchmark suite,

    A. Geiger, P. Lenz, and R. Urtasun, “Are we ready for autonomous driving? the kitti vision benchmark suite,” in 2012 IEEE conference on computer vision and pattern recognition . IEEE, 2012, pp. 3354–3361

    2012

  38. [46]

    Fast point feature histograms (fpfh) for 3d registration,

    R. B. Rusu, N. Blodow, and M. Beetz, “Fast point feature histograms (fpfh) for 3d registration,” in 2009 IEEE international conference on robotics and automation . IEEE, 2009, pp. 3212–3217

    2009

  39. [47]

    Fully convolutional geometric fea- tures,

    C. Choy, J. Park, and V . Koltun, “Fully convolutional geometric fea- tures,” in Proceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 8958–8966

    2019

  40. [48]

    A comparative study of sift and its variants,

    J. Wu, Z. Cui, V . S. Sheng, P. Zhao, D. Su, and S. Gong, “A comparative study of sift and its variants,”Measurement science review, vol. 13, no. 3, pp. 122–131, 2013

    2013

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.