Pith. sign in

REVIEW 4 major objections 3 minor 1 cited by

Camera Pose Refinement via 3D Gaussian Splatting

T0 review · 4 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Camera pose refinement using epipolar constraints over multiple 3DGS-rendered views reduces median translation error by 53.3% on 7-Scenes.

desk verdict Plausible training-free pose refinement via 3DGS renderings, but the abstract leaves open whether the 3DGS model is independent of the poses being refined. read the letter →

arxiv 2508.17876 v1 pith:NFKUSIS5 submitted 2025-08-25 cs.CV

Lulu Hao , Lipu Zhou , Zhenzhong Wei , Xu Wang This is my paper
classification cs.CV
keywords cameraposerefinement3DGaussianSplattingepipolargeometrynovelviewsynthesisvisuallocalizationiterativeoptimizationtraining-free7-Scenes
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

This paper introduces GS-SMC, a camera pose refinement method built on 3D Gaussian Splatting. The idea is to take an existing 3DGS model of a scene, render several nearby views from it, and then adjust the query camera pose so that epipolar geometry between the rendered views and the query image is satisfied. Because it only needs a 3DGS model and any off-the-shelf feature extractor/matcher, the method works across scenes without retraining or per-scene reconstruction. On the 7-Scenes and Cambridge Landmarks datasets, the refinement reduces median translation and rotation errors by roughly half relative to current refinement approaches. The paper's central claim is that geometry constraints from multiple rendered views, not just feature similarity, are enough to correct initial pose estimates.

What carries the argument

The key mechanism is GS-SMC's iterative pose optimization under epipolar constraints between the query image and several 3DGS-rendered views. The 3DGS model supplies photorealistic novel views from arbitrary poses, so the epipolar geometry between a rendered view and the query image provides a geometric error signal that can be optimized with respect to the pose parameters, without needing scene-specific training or descriptors.

What would settle it

Run GS-SMC on a scene where the 3DGS model is deliberately trained from the same inaccurate camera poses that provide the initial query pose, and compare the refinement result to using a 3DGS model trained from ground-truth poses. If the accuracy gain largely disappears in the first case, the method is relying on the model's internal accuracy rather than correcting pose errors itself.

Watch

Extended reading notes

Core claim

GS-SMC treats camera pose refinement as an optimization problem: given a query image with an initial pose and a pretrained 3D Gaussian Splatting model, the method renders multiple views around the initial pose, extracts and matches features between those renders and the query, and minimizes an epipolar-geometric cost over the pose. The central discovery is that this multi-view epipolar constraint, which does not require 2D-3D correspondences or a network trained for the specific scene, is sufficient to substantially improve pose accuracy. The paper reports that this approach outperforms state-of-the-art refinement methods, with median translation and rotation error reductions of 53.3% and 56

Load-bearing premise

The method assumes the pre-existing 3DGS model is an accurate scene representation and that rendered views from it, when matched to the query image, produce epipolar constraints whose optimization converges to the true pose; if the model's poses are biased by the very errors the refinement should correct, the rendered constraints can perpetuate those errors.

Editorial extensions

If this is right

  • Any scene that already has a 3DGS model can immediately use GS-SMC for pose refinement, since no per-scene training or descriptor-specific reconstruction is needed.
  • The refinement method is indifferent to which feature extractor or matcher is used, so improvements in general-purpose matching directly improve pose accuracy.
  • On the reported benchmarks, the median translation and rotation errors drop by roughly half, a gain large enough to matter for downstream AR and robot localization.
  • Because the method uses rendered images rather than explicit 2D-3D correspondences, it can be applied to scenes where traditional matching is unreliable.

Reading between the lines

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

  • If the pre-existing 3DGS model was trained from the same inaccurate poses that the initial query pose is derived from, the rendered views will inherit those errors and the epipolar constraints may pull the refined pose toward the model's bias; a test comparing models built from accurate versus erroneous poses would reveal whether the reported gains survive.
  • The optimization landscape over epipolar costs and rendered views may be non-convex; the paper's evaluations start from relatively good initializations, so how the method behaves under large pose perturbations remains open.
  • The multi-view rendered-image formulation could be extended from a single query image to refining a whole trajectory of poses simultaneously, potentially integrating with SLAM back-ends that already maintain a 3DGS map.
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

4 major / 3 minor

Summary. The paper proposes GS-SMC, a camera pose refinement framework built on 3D Gaussian Splatting. Given an existing 3DGS model, the method renders multiple novel views and iteratively refines the query camera pose by enforcing epipolar geometric constraints between the query image and these rendered images. The claimed advantages are that the method is training-free, scene-agnostic, and can use arbitrary feature extractors and matchers. The abstract reports large percentage reductions in median translation and rotation errors on 7-Scenes (53.3% and 56.9%) and Cambridge Landmarks (40.7% and 53.2%), allegedly outperforming state-of-the-art pose refinement approaches.

Significance. If the reported gains hold, the method would be a practically useful, lightweight refinement tool that leverages the widespread availability of 3DGS models without per-scene retraining. The idea of using rendered views to impose epipolar constraints is a reasonable and potentially novel alternative to descriptor-specific 2D-3D correspondence pipelines. However, the evidence presented in the abstract is only summary statistics. There are no absolute errors, baseline descriptions, error bars, or ablations, and the provenance of the 3DGS model is unspecified. The core contribution is plausible, but its validity cannot be assessed from the provided material.

major comments (4)
  1. [Abstract] The central quantitative claims (53.3%/56.9% on 7-Scenes and 40.7%/53.2% on Cambridge) are reported only as percentage reductions relative to unspecified baselines. No absolute translation/rotation errors are given, no error bars or statistical significance are reported, and the 'state-of-the-art' baselines are not named. These numbers are not independently verifiable from the abstract and should be presented with full experimental protocol, including absolute errors, standard deviations, and a table of all compared methods.
  2. [Abstract] The method assumes 'an existing 3DGS model' without specifying how that model was obtained. If the 3DGS model was reconstructed from the same initial pose estimates that the refinement aims to correct, then the rendered views encode those same pose errors. Matching the query to such views and enforcing epipolar consistency could reinforce the bias rather than correct it, making the reported reductions a form of self-consistency rather than absolute accuracy. The paper must clarify model provenance (e.g., trained with ground-truth poses or an independent reconstruction) and evaluate the sensitivity of the refinement to model pose error.
  3. [Abstract] The method is described as an 'iterative optimization approach' using epipolar constraints among multiple rendered views, but no convergence guarantees, initialization requirements, or characterization of the optimization landscape are provided. Since the feature extractors and matchers are 'flexibly chosen,' it is unclear whether the objective is well-behaved for arbitrary descriptors or prone to local minima. The paper should include an ablation over feature extractors/matchers and an analysis of convergence (e.g., success rate as a function of initial pose error).
  4. [Abstract] The abstract states that most refinement approaches rely on 2D-3D correspondences with specific descriptors or dedicated networks, but it does not compare against any specific recent method. The claimed superiority over 'state-of-the-art camera pose refinement approaches' is not substantiated without naming the baselines and reporting the comparison protocol (e.g., same initial poses, same number of refinement iterations, same runtime constraints).
minor comments (3)
  1. [Abstract] The acronym GS-SMC is not expanded. If SMC stands for something like 'Sampling-based Motion Compensation' or 'Sparse Multi-view Constraints,' it should be spelled out.
  2. [Abstract] The phrase 'epipolar geometric constraints among the query and multiple rendered images' is vague. It is unclear whether the constraints are pairwise between the query and each rendered view, or also among rendered views. A precise formulation (equations) would help.
  3. [Abstract] The term 'training-free' is somewhat ambiguous: the refinement does not train a network, but the 3DGS model itself is trained. The paper should clarify what exactly requires no training, especially if the 3DGS model is assumed to be pre-existing.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the abstract reports an external benchmark comparison against a pre-existing 3DGS model; no fitted parameter is renamed as prediction.

full rationale

The abstract describes a pose-refinement framework that renders novel views from an existing 3DGS model, establishes epipolar constraints between the query image and those rendered views, iteratively optimizes the pose, and evaluates on standard benchmarks (7-Scenes, Cambridge Landmarks). The claimed improvements are empirical comparisons against other methods on held-out test data, so they are not derived from the method's own inputs by construction. No equation or definition in the abstract identifies a fitted quantity with the reported output, and no load-bearing self-citation appears. The only conceivable circularity would arise if the 3DGS model had been reconstructed from the same erroneous poses being refined, but the abstract neither states nor implies this, and no quoted text exhibits such a reduction. Per the hard rules, speculation about unstated model provenance is not sufficient to claim circularity. Therefore the honest finding is no significant circularity.

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

No numerical free parameters are disclosed in the abstract; the full paper likely contains optimization hyperparameters and feature-extractor choices that would be listed in a full review. The assumptions listed are the load-bearing premises of the described pipeline.

assumptions (4)
  • domain assumption An existing 3DGS model of the target scene is accurate enough to render useful novel views.
    Abstract: 'our method can employ an existing 3DGS model to render novel views'. The whole pipeline depends on model availability and accuracy.
  • domain assumption Rendered views from nearby poses overlap with the query image sufficiently for feature matching to succeed.
    Abstract describes matching query and multiple rendered images; large baseline or occlusion cases are not discussed.
  • domain assumption The 7-Scenes and Cambridge Landmarks evaluations follow standard protocols and the reported reductions compare to strong baselines.
    Abstract reports only percentage reductions; the weakest link in the empirical claim is the unstated baseline set.
  • standard math Feature extractor and matcher correctness follows standard epipolar geometry.
    The method relies on textbook epipolar constraints; no new mathematical axioms are introduced in the abstract.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Camera Pose Refinement via 3D Gaussian Splatting." pith.science (2026). https://pith.science/paper/NFKUSIS5

@misc{pith2026250817876,
  author       = {Pith},
  title        = {Pith review of: Camera Pose Refinement via 3D Gaussian Splatting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NFKUSIS5}},
  note         = {Machine review of arXiv:2508.17876}
}
read the original abstract

Camera pose refinement aims at improving the accuracy of initial pose estimation for applications in 3D computer vision. Most refinement approaches rely on 2D-3D correspondences with specific descriptors or dedicated networks, requiring reconstructing the scene again for a different descriptor or fully retraining the network for each scene. Some recent methods instead infer pose from feature similarity, but their lack of geometry constraints results in less accuracy. To overcome these limitations, we propose a novel camera pose refinement framework leveraging 3D Gaussian Splatting (3DGS), referred to as GS-SMC. Given the widespread usage of 3DGS, our method can employ an existing 3DGS model to render novel views, providing a lightweight solution that can be directly applied to diverse scenes without additional training or fine-tuning. Specifically, we introduce an iterative optimization approach, which refines the camera pose using epipolar geometric constraints among the query and multiple rendered images. Our method allows flexibly choosing feature extractors and matchers to establish these constraints. Extensive empirical evaluations on the 7-Scenes and the Cambridge Landmarks datasets demonstrate that our method outperforms state-of-the-art camera pose refinement approaches, achieving 53.3% and 56.9% reductions in median translation and rotation errors on 7-Scenes, and 40.7% and 53.2% on Cambridge.

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Swimm3R: Splatting with Medium-aware SfM for Underwater 3D Reconstruction

    cs.CV 2026-08 conditional novelty 6.0 of 10

    Swimm3R couples a scattering-aware, feed-forward structure-from-motion backbone with underwater Beta splatting to reconstruct and render 3D scenes from turbid underwater video, improving rendering PSNR and localizatio...

Reference graph

Works this paper leans on

75 extracted references · 72 canonical work pages · cited by 1 Pith paper

  1. [1]

    Huang, S

    Y. Huang, S. Meng, Y. Zhu, J. Wang, J. Cao, X. Qiao, and X. Su, ``Webarnav: Mobile web ar indoor navigation with edge-assisted vision localization,'' IEEE Transactions on Mobile Computing, 2025

    work page 2025

  2. [2]

    H. Bao, W. Xie, Q. Qian, D. Chen, S. Zhai, N. Wang, and G. Zhang, ``Robust tightly-coupled visual-inertial odometry with pre-built maps in high latency situations,'' IEEE transactions on visualization and computer graphics, vol. 28, no. 5, pp. 2212--2222, 2022

    work page 2022

  3. [3]

    J. H. Jung, Y. Choe, and C. G. Park, ``Photometric visual-inertial navigation with uncertainty-aware ensembles,'' IEEE Transactions on Robotics, vol. 38, no. 4, pp. 2039--2052, 2022

    work page

  4. [4]

    H. Zhai, G. Huang, Q. Hu, G. Li, H. Bao, and G. Zhang, ``Nis-slam: Neural implicit semantic rgb-d slam for 3d consistent scene understanding,'' IEEE Transactions on Visualization and Computer Graphics, 2024

    work page 2024

  5. [5]

    X. Lei, M. Wang, W. Zhou, and H. Li, ``Gaussnav: Gaussian splatting for visual navigation,'' IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 47, no. 5, pp. 4108--4121, 2025

    work page 2025

  6. [6]

    T. Chen, O. Shorinwa, J. Bruno, A. Swann, J. Yu, W. Zeng, K. Nagami, P. Dames, and M. Schwager, ``Splat-nav: Safe real-time robot navigation in gaussian splatting maps,'' IEEE Transactions on Robotics, vol. 41, pp. 2765--2784, 2025

    work page 2025

  7. [7]

    J. L. Schönberger and J.-M. Frahm, ``Structure-from-motion revisited,'' in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 4104--4113

    work page 2016

  8. [8]

    H. Cui, X. Gao, S. Shen, and Z. Hu, ``Hsfm: Hybrid structure-from-motion,'' in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 1212--1221

    work page 2017

Show all 75 references
  1. [9]

    Sarlin, C

    P.-E. Sarlin, C. Cadena, R. Siegwart, and M. Dymczyk, ``From coarse to fine: Robust hierarchical localization at large scale,'' in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 12\,716--12\,725

    2019

  2. [10]

    F. Ge, Y. Zhang, L. Wang, S. Coleman, and D. Kerr, ``Double-domain adaptation semantics for retrieval-based long-term visual localization,'' IEEE Transactions on Multimedia, vol. 26, pp. 6050--6064, 2024

    2024

  3. [11]

    Piasco, D

    N. Piasco, D. Sidib \'e , C. Demonceaux, and V. Gouet-Brunet, ``Perspective-n-learned-point: Pose estimation from relative depth,'' in British machine vision conference (BMVC), 2019

    2019

  4. [12]

    FISCHLER AND, ``Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography,'' Commun

    M. FISCHLER AND, ``Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography,'' Commun. ACM, vol. 24, no. 6, pp. 381--395, 1981

    1981

  5. [13]

    Brachmann and C

    E. Brachmann and C. Rother, ``Visual camera re-localization from rgb and rgb-d images using dsac,'' IEEE transactions on pattern analysis and machine intelligence, vol. 44, no. 9, pp. 5847--5865, 2021

    2021

  6. [14]

    Brachmann, T

    E. Brachmann, T. Cavallari, and V. A. Prisacariu, ``Accelerated coordinate encoding: Learning to relocalize in minutes using rgb and poses,'' in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 5044--5053

    2023

  7. [15]

    F. Wang, X. Jiang, S. Galliani, C. Vogel, and M. Pollefeys, ``Glace: Global local accelerated coordinate encoding,'' in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 21\,562--21\,571

    2024

  8. [16]

    B. Wang, C. Chen, C. X. Lu, P. Zhao, N. Trigoni, and A. Markham, ``Atloc: Attention guided camera localization,'' in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 34, no. 06, 2020, pp. 10\,393--10\,401

    2020

  9. [17]

    Shavit, R

    Y. Shavit, R. Ferens, and Y. Keller, ``Learning multi-scene absolute pose regression with transformers,'' in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 2733--2742

    2021

  10. [18]

    S. Chen, X. Li, Z. Wang, and V. A. Prisacariu, ``Dfnet: Enhance absolute pose regression with direct feature matching,'' in European Conference on Computer Vision. 1em plus 0.5em minus 0.4em Springer, 2022, pp. 1--17

    2022

  11. [19]

    Moreau, N

    A. Moreau, N. Piasco, D. Tsishkou, B. Stanciulescu, and A. de La Fortelle, ``Lens: Localization enhanced by nerf synthesis,'' in Conference on Robot Learning. 1em plus 0.5em minus 0.4em PMLR, 2022, pp. 1347--1356

    2022

  12. [20]

    J. Miao, K. Jiang, T. Wen, Y. Wang, P. Jia, B. Wijaya, X. Zhao, Q. Cheng, Z. Xiao, J. Huang et al., ``A survey on monocular re-localization: From the perspective of scene map representation,'' IEEE Transactions on Intelligent Vehicles, 2024

    2024

  13. [21]

    Sarlin, A

    P.-E. Sarlin, A. Unagar, M. Larsson, H. Germain, C. Toft, V. Larsson, M. Pollefeys, V. Lepetit, L. Hammarstrand, F. Kahl et al., ``Back to the feature: Learning robust camera localization from pixels to pose,'' in Proceedings of the IEEE/CVF conference on computer vision and p...

    2021

  14. [22]

    Hyeon, J

    J. Hyeon, J. Kim, and N. Doh, ``Pose correction for highly accurate visual localization in large-scale indoor spaces,'' in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 15\,974--15\,983

    2021

  15. [23]

    Mildenhall, P

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, and R. Ng, ``Nerf: Representing scenes as neural radiance fields for view synthesis,'' Communications of the ACM, vol. 65, no. 1, pp. 99--106, 2021

    2021

  16. [24]

    Yen-Chen, P

    L. Yen-Chen, P. Florence, J. T. Barron, A. Rodriguez, P. Isola, and T.-Y. Lin, ``Inerf: Inverting neural radiance fields for pose estimation,'' in 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). 1em plus 0.5em minus 0.4em IEEE, 2021, pp. 1323--1330

    2021

  17. [25]

    Y. Lin, T. M \"u ller, J. Tremblay, B. Wen, S. Tyree, A. Evans, P. A. Vela, and S. Birchfield, ``Parallel inversion of neural radiance fields for robust pose estimation,'' in 2023 IEEE International Conference on Robotics and Automation (ICRA). 1em plus 0.5em minus 0.4em IEEE,...

    2023

  18. [26]

    Moreau, N

    A. Moreau, N. Piasco, M. Bennehar, D. Tsishkou, B. Stanciulescu, and A. de La Fortelle, ``Crossfire: Camera relocalization on self-supervised features from an implicit representation,'' in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 252--262

    2023

  19. [27]

    S. Chen, Y. Bhalgat, X. Li, J.-W. Bian, K. Li, Z. Wang, and V. A. Prisacariu, ``Neural refinement for absolute pose regression with feature synthesis,'' in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2024, pp. 20\,987--20\,996

    2024

  20. [28]

    C. Liu, S. Chen, Y. Zhao, H. Huang, V. Prisacariu, and T. Braud, ``Hr-apr: Apr-agnostic framework with uncertainty estimation and hierarchical refinement for camera relocalisation,'' in 2024 IEEE International Conference on Robotics and Automation (ICRA). 1em plus 0.5em minus ...

    2024

  21. [29]

    Kerbl, G

    B. Kerbl, G. Kopanas, T. Leimk \"u hler, and G. Drettakis, ``3d gaussian splatting for real-time radiance field rendering,'' ACM Trans. Graph., vol. 42, no. 4, pp. 139--1, 2023

    2023

  22. [30]

    B. Fei, J. Xu, R. Zhang, Q. Zhou, W. Yang, and Y. He, ``3d gaussian splatting as new era: A survey,'' IEEE Transactions on Visualization and Computer Graphics, 2024

    2024

  23. [31]

    Y. Wang, J. Wang, R. Gao, Y. Qu, W. Duan, S. Yang, and Y. Qi, ``Look at the sky: Sky-aware efficient 3d gaussian splatting in the wild,'' IEEE Transactions on Visualization and Computer Graphics, 2025

    2025

  24. [32]

    Botashev, V

    K. Botashev, V. Pyatov, G. Ferrer, and S. Lefkimmiatis, ``Gsloc: Visual localization with 3d gaussian splatting,'' in 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). 1em plus 0.5em minus 0.4em IEEE, 2024, pp. 5664--5671

    2024

  25. [33]

    Z. Zhou, F. Hui, Y. Wu, and Y. Liu, ``Six-dof pose estimation with efficient 3-d gaussian splatting representation for visual relocalization,'' IEEE/ASME Transactions on Mechatronics, pp. 1--10, 2024

    2024

  26. [34]

    C. Liu, S. Chen, Y. S. Bhalgat, S. HU, M. Cheng, Z. Wang, V. A. Prisacariu, and T. Braud, `` GS - CPR : Efficient camera pose refinement via 3d gaussian splatting,'' in The Thirteenth International Conference on Learning Representations, 2025, pp. 1--20. [Online]. Available: h...

    2025

  27. [35]

    Trivigno, C

    G. Trivigno, C. Masone, B. Caputo, and T. Sattler, ``The unreasonable effectiveness of pre-trained features for camera pose refinement,'' in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 12\,786--12\,798

    2024

  28. [36]

    H. Li, C. Xu, W. Yang, L. Mi, H. Yu, H. Zhang, and G.-S. Xia, ``Unsupervised multi-view uav image geo-localization via iterative rendering,'' IEEE Transactions on Geoscience and Remote Sensing, pp. 1--1, 2025

    2025

  29. [37]

    Sattler, B

    T. Sattler, B. Leibe, and L. Kobbelt, ``Improving image-based localization by active correspondence search,'' in European conference on computer vision. 1em plus 0.5em minus 0.4em Springer, 2012, pp. 752--765

    2012

  30. [38]

    Reyes-Aviles, P

    F. Reyes-Aviles, P. Fleck, D. Schmalstieg, and C. Arth, ``Bag of world anchors for instant large-scale localization,'' IEEE Transactions on Visualization and Computer Graphics, vol. 29, no. 11, pp. 4730--4739, 2023

    2023

  31. [39]

    D. G. Lowe, ``Distinctive image features from scale-invariant keypoints,'' International journal of computer vision, vol. 60, pp. 91--110, 2004

    2004

  32. [40]

    J. Wang, Y. Zhong, Y. Dai, S. Birchfield, K. Zhang, N. Smolyanskiy, and H. Li, ``Deep two-view structure-from-motion revisited,'' in Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition, 2021, pp. 8953--8962

    2021

  33. [41]

    J. Wang, N. Karaev, C. Rupprecht, and D. Novotny, ``Vggsfm: Visual geometry grounded deep structure from motion,'' in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 21\,686--21\,697

    2024

  34. [42]

    C. Chen, B. Wang, C. X. Lu, N. Trigoni, and A. Markham, ``Deep learning for visual localization and mapping: A survey,'' IEEE Transactions on Neural Networks and Learning Systems, 2023

    2023

  35. [43]

    Brachmann, F

    E. Brachmann, F. Michel, A. Krull, M. Y. Yang, S. Gumhold, and C. Rother, ``Uncertainty-driven 6d pose estimation of objects and scenes from a single rgb image,'' in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 3364--3372

    2016

  36. [44]

    Brachmann and C

    E. Brachmann and C. Rother, ``Learning less is more-6d camera localization via 3d surface regression,'' in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 4654--4662

    2018

  37. [45]

    Kendall, M

    A. Kendall, M. Grimes, and R. Cipolla, ``Posenet: A convolutional network for real-time 6-dof camera relocalization,'' in Proceedings of the IEEE international conference on computer vision, 2015, pp. 2938--2946

    2015

  38. [46]

    Kendall and R

    A. Kendall and R. Cipolla, ``Geometric loss functions for camera pose regression with deep learning,'' in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 5974--5983

    2017

  39. [47]

    Walch, C

    F. Walch, C. Hazirbas, L. Leal-Taixe, T. Sattler, S. Hilsenbeck, and D. Cremers, ``Image-based localization using lstms for structured feature correlation,'' in Proceedings of the IEEE international conference on computer vision, 2017, pp. 627--637

    2017

  40. [48]

    H. Li, X. Yang, H. Zhai, Y. Liu, H. Bao, and G. Zhang, ``Vox-surf: Voxel-based implicit surface representation,'' IEEE Transactions on Visualization and Computer Graphics, vol. 30, no. 3, pp. 1743--1755, 2022

    2022

  41. [49]

    Y. Shi, Y. Tao, M. Yang, Y. Liu, L. Yi, and B. Zhou, ``Articulated motion-aware nerf for 3d dynamic appearance and geometry reconstruction by implicit motion states,'' IEEE transactions on visualization and computer graphics, 2024

    2024

  42. [50]

    S. Chen, Z. Wang, and V. Prisacariu, ``Direct-posenet: Absolute pose regression with photometric consistency,'' in 2021 International Conference on 3D Vision (3DV). 1em plus 0.5em minus 0.4em IEEE, 2021, pp. 1175--1185

    2021

  43. [51]

    H. Li, H. Zhai, X. Yang, Z. Wu, Y. Zheng, H. Wang, J. Wu, H. Bao, and G. Zhang, ``Imtooth: Neural implicit tooth for dental augmented reality,'' IEEE Transactions on Visualization and Computer Graphics, vol. 29, no. 5, pp. 2837--2846, 2023

    2023

  44. [52]

    Maggio, M

    D. Maggio, M. Abate, J. Shi, C. Mario, and L. Carlone, ``Loc-nerf: Monte carlo localization using neural radiance fields,'' in 2023 IEEE International Conference on Robotics and Automation (ICRA). 1em plus 0.5em minus 0.4em IEEE, 2023, pp. 4018--4025

    2023

  45. [53]

    Germain, D

    H. Germain, D. DeTone, G. Pascoe, T. Schmidt, D. Novotny, R. Newcombe, C. Sweeney, R. Szeliski, and V. Balntas, ``Feature query networks: Neural surface description for camera pose refinement,'' in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogniti...

    2022

  46. [54]

    S. Chen, T. Cavallari, V. A. Prisacariu, and E. Brachmann, ``Map-relative pose regression for visual re-localization,'' in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 20\,665--20\,674

    2024

  47. [55]

    T. Lu, M. Yu, L. Xu, Y. Xiangli, L. Wang, D. Lin, and B. Dai, ``Scaffold-gs: Structured 3d gaussians for view-adaptive rendering,'' in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 20\,654--20\,664

    2024

  48. [56]

    Y. Sun, X. Wang, Y. Zhang, J. Zhang, C. Jiang, Y. Guo, and F. Wang, ``icomma: Inverting 3d gaussian splatting for camera pose estimation via comparing and matching,'' arXiv preprint arXiv:2312.09031, 2023

    2023 arXiv

  49. [57]

    Z. Niu, Z. Tan, J. Zhang, X. Yang, and D. Hu, ``Hgsloc: 3dgs-based heuristic camera pose refinement,'' arXiv preprint arXiv:2409.10925, 2024

    2024 arXiv

  50. [58]

    R. Zhu, Y. Liang, H. Chang, J. Deng, J. Lu, W. Yang, T. Zhang, and Y. Zhang, ``Motiongs: Exploring explicit motion guidance for deformable 3d gaussian splatting,'' Advances in Neural Information Processing Systems, vol. 37, pp. 101\,790--101\,817, 2024

    2024

  51. [59]

    W. Meng, T. Wu, H. Yin, and F. Zhang, ``Nurf: Nudging the particle filter in radiance fields for robot visual localization,'' IEEE Transactions on Cognitive and Developmental Systems, pp. 1--10, 2025

    2025

  52. [60]

    H. Zhai, X. Zhang, B. Zhao, H. Li, Y. He, Z. Cui, H. Bao, and G. Zhang, ``Splatloc: 3d gaussian splatting-based visual localization for augmented reality,'' IEEE Transactions on Visualization and Computer Graphics, vol. 31, no. 5, pp. 3591--3601, 2025

    2025

  53. [61]

    L. Yang, X. Zhao, Q. Sun, K. Wang, A. Chen, and P. Kang, ``Splatpose: Geometry-aware 6-dof pose estimation from single rgb image via 3d gaussian splatting,'' arXiv preprint arXiv:2503.05174, 2025

    2025 arXiv

  54. [62]

    Engel, V

    J. Engel, V. Koltun, and D. Cremers, ``Direct sparse odometry,'' IEEE transactions on pattern analysis and machine intelligence, vol. 40, no. 3, pp. 611--625, 2017

    2017

  55. [63]

    Z. Yan, M. Ye, and L. Ren, ``Dense visual slam with probabilistic surfel map,'' IEEE transactions on visualization and computer graphics, vol. 23, no. 11, pp. 2389--2398, 2017

    2017

  56. [64]

    Von Stumberg, P

    L. Von Stumberg, P. Wenzel, N. Yang, and D. Cremers, ``Lm-reloc: Levenberg-marquardt based direct visual relocalization,'' in 2020 International Conference on 3D Vision (3DV). 1em plus 0.5em minus 0.4em IEEE, 2020, pp. 968--977

    2020

  57. [65]

    Von Stumberg, P

    L. Von Stumberg, P. Wenzel, Q. Khan, and D. Cremers, ``Gn-net: The gauss-newton loss for multi-weather relocalization,'' IEEE Robotics and Automation Letters, vol. 5, no. 2, pp. 890--897, 2020

    2020

  58. [66]

    J. Zhu, S. Yan, L. Wang, S. Zhang, Y. Liu, and M. Zhang, ``Lod-loc: Aerial visual localization using lod 3d map with neural wireframe alignment,'' Advances in Neural Information Processing Systems, vol. 37, pp. 119\,063--119\,098, 2024

    2024

  59. [67]

    J. Zhu, S. Peng, L. Wang, H. Tan, Y. Liu, M. Zhang, and S. Yan, ``Lod-loc v2: Aerial visual localization over low level-of-detail city models using explicit silhouette alignment,'' arXiv preprint arXiv:2507.00659, 2025

    2025 arXiv

  60. [68]

    J. B. Diederik P. Kingma, ``Adam: A method for stochastic optimization,'' International Conference on Learning Representations (ICLR), 2015

    2015

  61. [69]

    Pietrantoni, G

    M. Pietrantoni, G. Csurka, M. Humenberger, and T. Sattler, ``Self-supervised learning of neural implicit feature fields for camera pose refinement,'' in 2024 International Conference on 3D Vision (3DV). 1em plus 0.5em minus 0.4em IEEE, 2024, pp. 484--494

    2024

  62. [70]

    Y. Liu, R. Wu, S. Yan, X. Cheng, J. Zhu, Y. Liu, and M. Zhang, ``Atloc: Aerial thermal images localization via view synthesis,'' IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1--13, 2024

    2024

  63. [71]

    DeTone, T

    D. DeTone, T. Malisiewicz, and A. Rabinovich, ``Superpoint: Self-supervised interest point detection and description,'' in Proceedings of the IEEE conference on computer vision and pattern recognition workshops, 2018, pp. 224--236

    2018

  64. [72]

    Lindenberger, P.-E

    P. Lindenberger, P.-E. Sarlin, and M. Pollefeys, ``Lightglue: Local feature matching at light speed,'' in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 17\,627--17\,638

    2023

  65. [73]

    J. J. Mor \'e , ``The levenberg-marquardt algorithm: implementation and theory,'' in Numerical analysis: proceedings of the biennial Conference held at Dundee, June 28--July 1, 1977. 1em plus 0.5em minus 0.4em Springer, 2006, pp. 105--116

    1977

  66. [74]

    Shotton, B

    J. Shotton, B. Glocker, C. Zach, S. Izadi, A. Criminisi, and A. Fitzgibbon, ``Scene coordinate regression forests for camera relocalization in rgb-d images,'' in Proceedings of the IEEE conference on computer vision and pattern recognition, 2013, pp. 2930--2937

    2013

  67. [75]

    Brachmann, M

    E. Brachmann, M. Humenberger, C. Rother, and T. Sattler, ``On the limits of pseudo ground truth in visual camera re-localisation,'' in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 6218--6228

    2021

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.