pith. machine review for the scientific record. sign in

arxiv: 2602.22639 · v2 · submitted 2026-02-26 · 💻 cs.CV · cs.NA· math.NA· math.OC

Recognition: 2 theorem links

· Lean Theorem

QuadSync: Quadrifocal Tensor Synchronization via Tucker Decomposition

Daniel Miao , Gilad Lerman , Joe Kileel
Authors on Pith no claims yet

Pith reviewed 2026-05-15 19:19 UTC · model grok-4.3

classification 💻 cs.CV cs.NAmath.NAmath.OC
keywords quadrifocal tensorTucker decompositioncamera synchronizationstructure from motionmultilinear rankADMMiteratively reweighted least squares
0
0 comments X

The pith

The block quadrifocal tensor formed from n cameras admits a Tucker decomposition of multilinear rank (4,4,4,4) whose factors are the stacked camera matrices.

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

The paper constructs a single block quadrifocal tensor from a collection of n views and proves it factors via Tucker decomposition into four matrices that are precisely the stacked camera matrices. Because the multilinear rank stays fixed at (4,4,4,4) no matter how large n grows, the decomposition directly yields all camera matrices at once. The authors turn this algebraic fact into a practical synchronization algorithm that alternates between Tucker decomposition, ADMM, and iteratively reweighted least squares. They also relate the block quadrifocal tensor to its bifocal and trifocal counterparts and supply a joint synchronization procedure for all three. Experiments on contemporary datasets show that the higher-order tensors can be estimated accurately enough for the low-rank recovery to succeed.

Core claim

The block quadrifocal tensor admits a Tucker decomposition whose factor matrices are exactly the stacked camera matrices, and therefore possesses a multilinear rank of (4,4,4,4) that is independent of the number of views. This low-rank structure is recovered by an ADMM-IRLS procedure that synchronizes the collection of quadrifocal tensors, and the same construction extends to joint synchronization of bifocal, trifocal, and quadrifocal tensors.

What carries the argument

Tucker decomposition of the block quadrifocal tensor, whose four factor matrices recover the stacked camera matrices.

Load-bearing premise

Accurate quadrifocal tensors can be estimated from image correspondences and the ADMM-IRLS procedure recovers the exact low-rank factors without being trapped by local minima or inconsistent tensor estimates.

What would settle it

A numerical check on real or synthetic data showing that the multilinear rank of the assembled block quadrifocal tensor exceeds (4,4,4,4) for any n greater than 4 would falsify the central algebraic claim.

Figures

Figures reproduced from arXiv: 2602.22639 by Daniel Miao, Gilad Lerman, Joe Kileel.

Figure 1
Figure 1. Figure 1: Mean location error for ETH3D datasets [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Mean location error for EPFL datasets When there is no noise in any of the initial trifocal tensors, aP1 i =P 4 i H43 and bP1 j =P 4 j H43 for some nonzero scales a,b∈R. However, when there is noise, this will not hold. Our heuristic d(i, j, k, l) can be defined as a tuple measuring the distance between the first two and last two cameras: d(i, j, k, l):=g(d(P 1 i , P4 i H43), d(P 1 j , P4 j H43)). (11) In … view at source ↗
Figure 3
Figure 3. Figure 3: QuadSync retrieved camera poses on near-collinear views from plant [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Randomized updates in QuadSync tested on ETH3D ‘relief’ dataset. [PITH_FULL_IMAGE:figures/full_fig_p024_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Ground truth location of 10 cameras. × is the origin. ◦ denotes the camera centers. We conduct a small set of synthetic experiments to demonstrate that our algorithm can successfully recover camera poses in the collinear setting. A simple diagram can show the situation, see [PITH_FULL_IMAGE:figures/full_fig_p025_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Retrieved poses (colored, where each cluster has a distinct color) vs. ground truth poses (black) for CastleP30 with [PITH_FULL_IMAGE:figures/full_fig_p026_6.png] view at source ↗
read the original abstract

In structure from motion, quadrifocal tensors capture more information than their pairwise counterparts (essential matrices), yet they have often been thought of as impractical and only of theoretical interest. In this work, we challenge such beliefs by providing a new framework to recover $n$ cameras from the corresponding collection of quadrifocal tensors. We form the block quadrifocal tensor and show that it admits a Tucker decomposition whose factor matrices are the stacked camera matrices, and which thus has a multilinear rank of (4,~4,~4,~4) independent of $n$. We develop the first synchronization algorithm for quadrifocal tensors, using Tucker decomposition, alternating direction method of multipliers, and iteratively reweighted least squares. We further establish relationships between the block quadrifocal, trifocal, and bifocal tensors, and introduce an algorithm that jointly synchronizes these three entities. Numerical experiments demonstrate the effectiveness of our methods on modern datasets, indicating the potential and importance of using higher-order information in synchronization.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 3 minor

Summary. The paper introduces QuadSync, a synchronization method for n cameras from a collection of quadrifocal tensors. It constructs a block quadrifocal tensor by assembling the individual 3x3x3x3 tensors into a larger 3n x 3n x 3n x 3n array and claims this block tensor admits an exact Tucker decomposition whose four factor matrices are identical and equal to the stacked 3n x 4 camera matrix P, yielding multilinear rank exactly (4,4,4,4) independent of n. An ADMM-IRLS algorithm is derived to recover the factors from noisy tensor estimates; relationships to block trifocal and bifocal tensors are established, enabling a joint synchronization procedure. Numerical experiments on modern SfM datasets are reported to demonstrate practical effectiveness.

Significance. If the algebraic claim holds, the work supplies the first practical synchronization algorithm that directly exploits quadrifocal constraints, which encode more geometric information than essential matrices or trifocal tensors. The joint quad-tri-bi synchronization and the parameter-free multilinear-rank property (once the block tensor is defined) are genuine strengths. The approach could improve robustness in large-scale SfM when accurate higher-order tensors can be estimated. However, the absence of error bounds, noise-model analysis, or ablation on tensor estimation quality limits immediate impact; the result is plausible but its practical significance depends on how well the low-rank recovery behaves under realistic correspondence noise.

major comments (3)
  1. [§3] §3 (Block quadrifocal tensor construction): The central claim that the assembled block tensor T exactly equals a Tucker product with core C and identical factor matrices P (the stacked camera matrix) yielding multilinear rank (4,4,4,4) independent of n is load-bearing. The geometric quadrifocal tensor is defined via determinants or null-space contractions (Hartley & Zisserman, §15.3) that are not multilinear in the camera rows; an explicit algebraic derivation showing why the block assembly nevertheless produces the claimed Tucker form is required, together with a short proof that the rank remains exactly (4,4,4,4) for arbitrary n.
  2. [§4] §4 (ADMM-IRLS algorithm): No convergence analysis or recovery guarantee is provided for the non-convex IRLS step under the tensor estimation noise model that arises from image correspondences. Because the method’s correctness rests on recovering the exact low-rank factors, a brief discussion of local-minima behavior or a synthetic-noise experiment quantifying breakdown thresholds is needed.
  3. [§5] §5 (Numerical experiments): The reported results lack an ablation on the accuracy of the input quadrifocal tensors (e.g., varying correspondence noise or outlier rates). Without this, it is impossible to separate the contribution of the Tucker synchronization from the quality of the tensor estimator, weakening the claim that higher-order information improves synchronization.
minor comments (3)
  1. [Abstract] The abstract states the multilinear rank result but supplies no derivation outline; a single sentence sketching the key algebraic step would improve readability.
  2. [§3] Notation for the block tensor dimensions (3n versus 3n×3n×3n×3n) should be introduced once and used consistently in all equations.
  3. [§5] Figure captions for the synthetic and real-data plots should explicitly state the noise model and the number of trials averaged.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below and will revise the manuscript to incorporate the suggested improvements.

read point-by-point responses

  1. Referee: [§3] §3 (Block quadrifocal tensor construction): The central claim that the assembled block tensor T exactly equals a Tucker product with core C and identical factor matrices P (the stacked camera matrix) yielding multilinear rank (4,4,4,4) independent of n is load-bearing. The geometric quadrifocal tensor is defined via determinants or null-space contractions (Hartley & Zisserman, §15.3) that are not multilinear in the camera rows; an explicit algebraic derivation showing why the block assembly nevertheless produces the claimed Tucker form is required, together with a short proof that the rank remains exactly (4,4,4,4) for arbitrary n.

    Authors: We agree that making the algebraic foundation fully explicit will strengthen the paper. Section 3 already sketches the construction, but the revised manuscript will include a detailed step-by-step derivation showing how the block assembly of individual quadrifocal tensors yields the Tucker product with identical factor matrices P (the stacked 3n×4 camera matrix) and a short proof establishing that the multilinear rank is exactly (4,4,4,4) for any n, independent of the non-multilinear nature of the individual tensor definitions. revision: yes

  2. Referee: [§4] §4 (ADMM-IRLS algorithm): No convergence analysis or recovery guarantee is provided for the non-convex IRLS step under the tensor estimation noise model that arises from image correspondences. Because the method’s correctness rests on recovering the exact low-rank factors, a brief discussion of local-minima behavior or a synthetic-noise experiment quantifying breakdown thresholds is needed.

    Authors: We acknowledge the lack of formal analysis. In the revision we will add a concise discussion of local-minima behavior for the non-convex IRLS step and include synthetic experiments that inject controlled correspondence noise to quantify breakdown thresholds, thereby clarifying the practical recovery behavior under realistic tensor estimation noise. revision: yes

  3. Referee: [§5] §5 (Numerical experiments): The reported results lack an ablation on the accuracy of the input quadrifocal tensors (e.g., varying correspondence noise or outlier rates). Without this, it is impossible to separate the contribution of the Tucker synchronization from the quality of the tensor estimator, weakening the claim that higher-order information improves synchronization.

    Authors: We agree that an ablation study is necessary to isolate the synchronization contribution. The revised experimental section will include additional results ablating input tensor accuracy by varying correspondence noise levels and outlier rates, allowing readers to assess the robustness of the Tucker-based synchronization independently of the tensor estimator quality. revision: yes

Circularity Check

0 steps flagged

No circularity: algebraic derivation of Tucker form from block tensor construction is self-contained

full rationale

The paper forms the block quadrifocal tensor by stacking estimated individual quadrifocal tensors into a larger 4-mode array and algebraically demonstrates that this block tensor admits a Tucker decomposition whose factor matrices are exactly the stacked camera matrices P (with multilinear rank (4,4,4,4) independent of n). This follows directly from the definition of the block construction and the algebraic properties of the quadrifocal tensor; it is not obtained by fitting parameters to data, renaming a known result, or reducing to a self-citation chain. No load-bearing steps match the enumerated circularity patterns. The derivation stands on its own algebraic steps without self-referential inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The framework rests on standard properties of Tucker decomposition and multilinear rank from tensor algebra, plus the geometric definition of quadrifocal tensors from prior structure-from-motion literature; no new free parameters or invented entities are introduced in the abstract.

axioms (2)
  • standard math Tucker decomposition exists and its multilinear rank is well-defined for the constructed block tensor
    Invoked when stating the rank-(4,4,4,4) property independent of n.
  • domain assumption Quadrifocal tensors can be formed from image correspondences and stacked into a consistent block tensor
    Underlying the formation of the block quadrifocal tensor from n cameras.

pith-pipeline@v0.9.0 · 5478 in / 1456 out tokens · 32018 ms · 2026-05-15T19:19:44.155475+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

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

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

Reference graph

Works this paper leans on

61 extracted references · 61 canonical work pages

  1. [1]

    Radial multi-focal tensors: Applications to omnidirectional camera calibration

    SriRam Thirthala and Marc Pollefeys. Radial multi-focal tensors: Applications to omnidirectional camera calibration. IJCV, 96(2):195–211, 2012. 1

    work page 2012

  2. [2]

    Photo Tourism: Exploring photo collections in 3D

    Noah Snavely, Steven M Seitz, and Richard Szeliski. Photo Tourism: Exploring photo collections in 3D. In SIGGRAPH, pages 835–846. 2006. 1

    work page 2006

  3. [3]

    Structure-from-motion revisited

    Johannes L Schonberger and Jan-Michael Frahm. Structure-from-motion revisited. InCVPR, pages 4104–4113,

    work page

  4. [4]

    A new rank constraint on multi-view fundamental matrices, and its application to camera location recovery

    Soumyadip Sengupta, Tal Amir, Meirav Galun, Tom Goldstein, David W Jacobs, Amit Singer, and Ronen Basri. A new rank constraint on multi-view fundamental matrices, and its application to camera location recovery. In CVPR, pages 4798–4806, 2017. 2, 11

    work page 2017

  5. [5]

    GPSfM: Global projective sfm using algebraic constraints on multi-view fundamental matrices

    Y oni Kasten, Amnon Geifman, Meirav Galun, and Ronen Basri. GPSfM: Global projective sfm using algebraic constraints on multi-view fundamental matrices. InCVPR, pages 3264–3272, 2019. 2

    work page 2019

  6. [6]

    Algebraic characterization of essential matrices and their averaging in multiview settings

    Y oni Kasten, Amnon Geifman, Meirav Galun, and Ronen Basri. Algebraic characterization of essential matrices and their averaging in multiview settings. InICCV, pages 5895–5903, 2019. 2

    work page 2019

  7. [7]

    Averaging essential and fundamental matrices in collinear camera settings

    Amnon Geifman, Y oni Kasten, Meirav Galun, and Ronen Basri. Averaging essential and fundamental matrices in collinear camera settings. InCVPR, pages 13–19, 2020. 2, 4

    work page 2020

  8. [8]

    On the recovery of cameras from fundamental matrices

    Rakshith Madhavan and Federica Arrigoni. On the recovery of cameras from fundamental matrices. InICCV, pages 20934–20943, 2025. 2

    work page 2025

  9. [9]

    Lie-algebraic averaging for globally consistent motion estimation

    V enu Madhav Govindu. Lie-algebraic averaging for globally consistent motion estimation. InCVPR, volume 1, pages 1–8, 2004. 2

    work page 2004

  10. [10]

    Efficient and robust large-scale rotation averaging

    Avishek Chatterjee and V enu Madhav Govindu. Efficient and robust large-scale rotation averaging. InICCV, pages 521–528, 2013

    work page 2013

  11. [11]

    Rotation averaging.IJCV, 103(3):267–305, 2013

    Richard Hartley, Jochen Trumpf, Y uchao Dai, and Hongdong Li. Rotation averaging.IJCV, 103(3):267–305, 2013. 13

    work page 2013

  12. [12]

    Robust relative rotation averaging.IEEE Transactions on P attern Analysis and Machine Intelligence, 40(4):958–972, 2017

    Avishek Chatterjee and V enu Madhav Govindu. Robust relative rotation averaging.IEEE Transactions on P attern Analysis and Machine Intelligence, 40(4):958–972, 2017

    work page 2017

  13. [13]

    Message passing least squares framework and its application to rotation synchronization

    Y unpeng Shi and Gilad Lerman. Message passing least squares framework and its application to rotation synchronization. InICML, pages 8796–8806. PMLR, 2020. 11

    work page 2020

  14. [14]

    Global motion estimation from point matches

    Mica Arie-Nachimson, Shahar Z Kovalsky, Ira Kemelmacher-Shlizerman, Amit Singer, and Ronen Basri. Global motion estimation from point matches. In2012 Second International Conference on 3D Imaging, Modeling, Processing, V isualization & Transmission, pages 81–88. IEEE, 2012. 2

    work page 2012

  15. [15]

    Robust global translations with 1DSFM

    Kyle Wilson and Noah Snavely. Robust global translations with 1DSFM. InECCV, pages 61–75. Springer, 2014. 2

    work page 2014

  16. [16]

    Stable camera motion estimation using convex programming.SIAM Journal on Imaging Sciences, 8(2):1220–1262, 2015

    Onur Ozyesil, Amit Singer, and Ronen Basri. Stable camera motion estimation using convex programming.SIAM Journal on Imaging Sciences, 8(2):1220–1262, 2015. 11

    work page 2015

  17. [17]

    Shapefit and Shapekick for robust, scalable structure from motion

    Thomas Goldstein, Paul Hand, Choongbum Lee, Vladislav V oroninski, and Stefano Soatto. Shapefit and Shapekick for robust, scalable structure from motion. InECCV, pages 289–304. Springer, 2016

    work page 2016

  18. [18]

    Baseline desensitizing in translation averaging

    Bingbing Zhuang, Loong-Fah Cheong, and Gim Hee Lee. Baseline desensitizing in translation averaging. InCVPR, pages 4539–4547, 2018. 11

    work page 2018

  19. [19]

    Cycle-Sync: Robust global camera pose estimation through enhanced cycle-consistent synchronization.arXiv preprint arXiv:2511.02329, 2025

    Shaohan Li, Y unpeng Shi, and Gilad Lerman. Cycle-Sync: Robust global camera pose estimation through enhanced cycle-consistent synchronization.arXiv preprint arXiv:2511.02329, 2025. 2, 11

    work page arXiv 2025

  20. [20]

    Spectral synchronization of multiple views in SE (3).SIAM Journal on Imaging Sciences, 9(4):1963–1990, 2016

    Federica Arrigoni, Beatrice Rossi, and Andrea Fusiello. Spectral synchronization of multiple views in SE (3).SIAM Journal on Imaging Sciences, 9(4):1963–1990, 2016. 2

    work page 1963

  21. [21]

    SE-Sync: A certifiably correct algorithm for synchronization over the special Euclidean group.The International Journal of Robotics Research, 38(2-3):95–125, 2019

    David M Rosen, Luca Carlone, Afonso S Bandeira, and John J Leonard. SE-Sync: A certifiably correct algorithm for synchronization over the special Euclidean group.The International Journal of Robotics Research, 38(2-3):95–125, 2019

    work page 2019

  22. [22]

    Sensor network localization by eigenvector synchronization over the Euclidean group.ACM Transactions on Sensor Networks (TOSN), 8(3):1–42, 2012

    Mihai Cucuringu, Y aron Lipman, and Amit Singer. Sensor network localization by eigenvector synchronization over the Euclidean group.ACM Transactions on Sensor Networks (TOSN), 8(3):1–42, 2012

    work page 2012

  23. [23]

    Cartan-Sync: Fast and global SE(d)-synchronization.IEEE Robotics and Automation Letters, 2(4):2127–2134, 2017

    Jesus Briales and Javier Gonzalez-Jimenez. Cartan-Sync: Fast and global SE(d)-synchronization.IEEE Robotics and Automation Letters, 2(4):2127–2134, 2017. 2

    work page 2017

  24. [24]

    Global structure-from-motion revisited

    Linfei Pan, D´aniel Bar´ath, Marc Pollefeys, and Johannes L Sch¨onberger. Global structure-from-motion revisited. InECCV, pages 58–77. Springer, 2024. 2

    work page 2024

  25. [25]

    Theia: A fast and scalable structure-from-motion library

    Christopher Sweeney, Tobias Hollerer, and Matthew Turk. Theia: A fast and scalable structure-from-motion library. InACMMM, pages 693–696, 2015. 2

    work page 2015

  26. [26]

    Efficient detection of long consistent cycles and its application to distributed synchronization

    Shaohan Li, Y unpeng Shi, and Gilad Lerman. Efficient detection of long consistent cycles and its application to distributed synchronization. InCVPR, pages 5260–5269, 2024. 2

    work page 2024

  27. [27]

    Tensor-based synchronization and the low-rankness of the block trifocal tensor

    Daniel Miao, Gilad Lerman, and Joe Kileel. Tensor-based synchronization and the low-rankness of the block trifocal tensor. InNeurIPS, volume 37, pages 69505–69532, 2024. 3, 10, 11, 16

    work page 2024

  28. [28]

    Higher-order group synchronization.arXiv preprint arXiv:2505.21932, 2025

    Adriana L Duncan and Joe Kileel. Higher-order group synchronization.arXiv preprint arXiv:2505.21932, 2025. 2, 9

    work page arXiv 2025

  29. [29]

    Four-view geometry with unknown radial distortion

    Petr Hruby, Viktor Korotynskiy, Timothy Duff, Luke Oeding, Marc Pollefeys, Tomas Pajdla, and Viktor Larsson. Four-view geometry with unknown radial distortion. InCVPR, pages 8990–9000, 2023. 2

    work page 2023

  30. [30]

    Accurate quadrifocal tracking for robust 3D visual odometry

    Andrew I Comport, Ezio Malis, and Patrick Rives. Accurate quadrifocal tracking for robust 3D visual odometry. InProceedings 2007 IEEE International Conference on Robotics and Automation, pages 40–45. IEEE, 2007. 2

    work page 2007

  31. [31]

    Real-time quadrifocal visual odometry.The International Journal of Robotics Research, 29(2-3):245–266, 2010

    Andrew I Comport, Ezio Malis, and Patrick Rives. Real-time quadrifocal visual odometry.The International Journal of Robotics Research, 29(2-3):245–266, 2010. 2

    work page 2010

  32. [32]

    3D line mapping revisited

    Shaohui Liu, Yifan Y u, R´emi Pautrat, Marc Pollefeys, and Viktor Larsson. 3D line mapping revisited. InCVPR, pages 21445–21455, 2023. 3

    work page 2023

  33. [33]

    GlueStick: Robust image matching by sticking points and lines together

    R´emi Pautrat, Iago Su´arez, Yifan Y u, Marc Pollefeys, and Viktor Larsson. GlueStick: Robust image matching by sticking points and lines together. InICCV, pages 9706–9716, 2023. 3, 12

    work page 2023

  34. [34]

    Cambridge university press,

    Richard Hartley and Andrew Zisserman.Multiple View Geometry in Computer V ision. Cambridge university press,

    work page

  35. [35]

    Computation of the quadrifocal tensor

    Richard I Hartley. Computation of the quadrifocal tensor. InECCV, pages 20–35. Springer, 1998

    work page 1998

  36. [36]

    The quadrifocal variety.Linear Algebra and its Applications, 512:306–330, 2017

    Luke Oeding. The quadrifocal variety.Linear Algebra and its Applications, 512:306–330, 2017. 3

    work page 2017

  37. [37]

    On the structure and properties of the quadrifocal tensor

    Amnon Shashua and Lior Wolf. On the structure and properties of the quadrifocal tensor. InECCV, pages 710–724,

    work page

  38. [38]

    Line-based relative pose estimation

    Ali Elqursh and Ahmed Elgammal. Line-based relative pose estimation. InCVPR, pages 3049–3056, 2011. 3

    work page 2011

  39. [39]

    Pose estimation with unknown focal length using points, directions and lines

    Y ubin Kuang and Kalle Astrom. Pose estimation with unknown focal length using points, directions and lines. InICCV, pages 529–536, 2013

    work page 2013

  40. [40]

    A clever elimination strategy for efficient minimal solvers

    Zuzana Kukelova, Joe Kileel, Bernd Sturmfels, and Tomas Pajdla. A clever elimination strategy for efficient minimal solvers. InCVPR, pages 4912–4921, 2017

    work page 2017

  41. [41]

    A minimal closed-form solution for multi-perspective pose estimation using points and lines

    Pedro Miraldo, Tiago Dias, and Srikumar Ramalingam. A minimal closed-form solution for multi-perspective pose estimation using points and lines. InECCV, pages 474–490, 2018

    work page 2018

  42. [42]

    Distortion varieties.F oundations of Computational Mathematics, 18(4):1043–1071, 2018

    Joe Kileel, Zuzana Kukelova, Tomas Pajdla, and Bernd Sturmfels. Distortion varieties.F oundations of Computational Mathematics, 18(4):1043–1071, 2018. 3

    work page 2018

  43. [43]

    Minimal problems for the calibrated trifocal variety.SIAM Journal on Applied Algebra and Geometry, 1(1):575–598, 2017

    Joe Kileel. Minimal problems for the calibrated trifocal variety.SIAM Journal on Applied Algebra and Geometry, 1(1):575–598, 2017

    work page 2017

  44. [44]

    The ideal of the trifocal variety.Mathematics of Computation, 83(289):2553–2574,

    Chris Aholt and Luke Oeding. The ideal of the trifocal variety.Mathematics of Computation, 83(289):2553–2574,

    work page

  45. [45]

    Snapshot of algebraic vision.AMS Proceedings of Symposia in Pure Mathematics, 111:323–363, 2025

    Joe Kileel and Kathlen Kohn. Snapshot of algebraic vision.AMS Proceedings of Symposia in Pure Mathematics, 111:323–363, 2025. 3

    work page 2025

  46. [46]

    Tensor decompositions and applications.SIAM Review, 51(3):455–500, 2009

    Tamara G Kolda and Brett W Bader. Tensor decompositions and applications.SIAM Review, 51(3):455–500, 2009. 4

    work page 2009

  47. [47]

    A subspace-constrained Tyler’s estimator and its applications to structure from motion

    Feng Y u, Teng Zhang, and Gilad Lerman. A subspace-constrained Tyler’s estimator and its applications to structure from motion. InCVPR, pages 14575–14584, 2024. 11

    work page 2024

  48. [48]

    Sch¨onberger, Silvano Galliani, Torsten Sattler, Konrad Schindler, Marc Pollefeys, and Andreas Geiger

    Thomas Sch¨ops, Johannes L. Sch¨onberger, Silvano Galliani, Torsten Sattler, Konrad Schindler, Marc Pollefeys, and Andreas Geiger. A multi-view stereo benchmark with high-resolution images and multi-camera videos. In CVPR, pages 3260–3269, 2017. 11

    work page 2017

  49. [49]

    BAD SLAM: Bundle adjusted direct RGB-D SLAM

    Thomas Sch¨ops, Torsten Sattler, and Marc Pollefeys. BAD SLAM: Bundle adjusted direct RGB-D SLAM. In CVPR, pages 134–144, 2019. 11

    work page 2019

  50. [50]

    A critical review of the trifocal tensor estimation

    Laura F Juli`a and Pascal Monasse. A critical review of the trifocal tensor estimation. InP acific-Rim Symposium on Image and V ideo T echnology, pages 337–349. Springer, 2017. 11

    work page 2017

  51. [51]

    On benchmarking camera calibration and multi-view stereo for high resolution imagery

    Christoph Strecha, Wolfgang V on Hansen, Luc V an Gool, Pascal Fua, and Ulrich Thoennessen. On benchmarking camera calibration and multi-view stereo for high resolution imagery. InCVPR, pages 1–8, 2008. 12

    work page 2008

  52. [52]

    Graph-cut RANSAC

    Daniel Barath and Jiˇr´ı Matas. Graph-cut RANSAC. InCVPR, pages 6733–6741, 2018. 12

    work page 2018

  53. [53]

    Dummit and Richard M

    David S. Dummit and Richard M. Foote.Abstract Algebra. John Wiley & Sons, 3rd edition, 2004. 16

    work page 2004

  54. [54]

    Rigid components identification and rigidity control in bearing-only localization using the graph cycle basis

    Roberto Tron, Luca Carlone, Frank Dellaert, and Kostas Daniilidis. Rigid components identification and rigidity control in bearing-only localization using the graph cycle basis. In2015 American Control Conference (ACC), pages 3911–3918. IEEE, 2015. 23

    work page 2015

  55. [55]

    Camera motion from group synchronization

    Federica Arrigoni, Andrea Fusiello, and Beatrice Rossi. Camera motion from group synchronization. In2016 F ourth International Conference on 3D V ision (3DV), pages 546–555. IEEE, 2016. 23 15 Appendices

    work page 2016

  56. [56]

    Line Projection Matrices LetA∈R 3×4 be a projection matrix, factoring asKR[I|−t]whereR∈SO(3)andt∈R 3

    Additional Background Material 7.1. Line Projection Matrices LetA∈R 3×4 be a projection matrix, factoring asKR[I|−t]whereR∈SO(3)andt∈R 3. Then, the exterior square is P=   A2∧A3 A3∧A1 A1∧A2   where Ai indicates the ith row of the camera matrix A. Here, ∧ is the wedge product, or the exterior product, be- tween two vectors. The order of the basis is gi...

    work page

  57. [57]

    Block Trifocal Tensor in Collinear Cases The reference [27] includes a characterization of the low multilinear rank for the block trifocal tensor under the non- collinear setting

    Proofs for Theorems 8.1. Block Trifocal Tensor in Collinear Cases The reference [27] includes a characterization of the low multilinear rank for the block trifocal tensor under the non- collinear setting. We extend this to the collinear setting, showing that the block trifocal tensor will then have a low multilinear rank of(5,4,4). Theorem 8.1.Given a blo...

    work page

  58. [58]

    We now see thatλ-entries with two1-indices agree

    So (13) looks the same but with indices permuted and possibly a sign flip. We now see thatλ-entries with two1-indices agree. Indeed, takingi=j above givesλπ1(i1i1)=λ π2(ii11) for all π1 and π2 that fix(ii11)and (i1i1)respectively. Soλii11=λπ(ii11) for allπ. Takingi̸=j givesλii11=λπ(i1j1)=λjj11 for allπ. Together, there existsc∈R∗ such thatc=λ π(ij11) for ...

    work page

  59. [59]

    QuadSync We present pseudocode for QuadSync in Algorithm 1 in this section

    Algorithm Details 9.1. QuadSync We present pseudocode for QuadSync in Algorithm 1 in this section. Algorithm 1Quadsync IRLS-ADMM Input:Q n ∈R3n×3n×3n×3n,ρ>0∈R Ωset of observed block indices Output: ¯C∈R 3n×4 NormalizeQn so that each block has norm 1 ObtainCi,Bfrom first four singular vectors in first factor matrix of HOSVD(Qn) Calculate initial IRLS weigh...

    work page

  60. [60]

    Randomized Updates in QuadSync We test the effect of randomized updates forCi in QuadSync for the ETH3D ‘relief’ dataset

    Additional Experiments 10.1. Randomized Updates in QuadSync We test the effect of randomized updates forCi in QuadSync for the ETH3D ‘relief’ dataset. Our ‘relief’ dataset contains 13 images, and the maximum number of columns for updatingCi is 27×13 3 = 59319. We try randomized updates using columns whose number range from 20 to 59319. Using 30 random col...

    work page 1944

  61. [61]

    We report the completion rates and an idea of the runtime in Table 3

    Detailed Numerical Results In this section, we include more comprehensive results accompanying the results in the main paper. We report the completion rates and an idea of the runtime in Table 3. We also report the detailed mean location error, median location error, mean rotation error, median rotation error in Tables 4, 5, 6, 7 respectively. Locations a...

    work page arXiv 2068