pith. machine review for the scientific record. sign in

arxiv: 2605.07351 · v1 · submitted 2026-05-08 · 💻 cs.CV

Recognition: 2 theorem links

· Lean Theorem

Disambiguating 2D-3D Correspondences in Gaussian Splatting-based Feature Fields for Visual Localization

Miso Lee , Sangeek Hyun , Yerim Jeon , Jae-Pil Heo
Authors on Pith no claims yet

Pith reviewed 2026-05-11 02:17 UTC · model grok-4.3

classification 💻 cs.CV
keywords Gaussian SplattingFeature FieldsVisual Localization2D-3D CorrespondencePnP Pose EstimationMixture of GaussiansMulti-view Consistency
0
0 comments X

The pith

Splitting Gaussians creates precise one-to-one 2D-3D matches that stabilize PnP pose estimation.

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

Existing Gaussian Splatting feature fields produce many-to-one pixel-to-point mappings and features that lack consistency across views, which prevents reliable camera pose recovery. The paper addresses this by decomposing each Gaussian into smaller ones and selecting only those that contribute strongly and consistently in rasterization. These steps replace ambiguous mappings with exact correspondences while preserving discriminative power. The resulting fields support stable PnP solving and reach state-of-the-art accuracy on localization benchmarks. The approach works without per-scene retraining or iterative refinement steps.

Core claim

By applying Mixture-of-Gaussians splitting to decompose each Gaussian and using rasterization composition weights to retain only multi-view consistent Gaussians, the method replaces many-to-one 2D-3D mappings with precise one-to-one correspondences and produces compact, discriminative feature fields that enable stable PnP convergence for visual localization.

What carries the argument

Mixture-of-Gaussians-based splitting together with composition-weight selection from GS rasterization, which decomposes ambiguous Gaussians into smaller ones and retains only those with strong multi-view contributions.

If this is right

  • Precise one-to-one correspondences allow the PnP solver to converge reliably without additional refinement.
  • The selected Gaussians remain compact while retaining high discriminability across views.
  • Photometric Gaussian fields become directly usable for accurate localization without extra per-scene optimization.
  • State-of-the-art performance is reached on standard visual localization benchmarks.
  • No iterative pose refinement is required after initial matching.

Where Pith is reading between the lines

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

  • The same splitting-plus-selection pattern could be tested on other Gaussian-based tasks such as novel view synthesis to reduce correspondence noise.
  • If the weight threshold generalizes across scenes, the method might allow a single set of hyperparameters to work for many environments.
  • In scenes with moving objects the consistency filter might automatically suppress transient Gaussians.
  • Combining this disambiguation with learned feature descriptors could further tighten the 2D-3D matches.

Load-bearing premise

That decomposing Gaussians will replace many-to-one mappings with exact one-to-one correspondences and that weight-based selection will enforce genuine multi-view consistency without losing discriminability or creating new artifacts.

What would settle it

Run the splitting and selection on a benchmark scene and check whether rendered feature maps still show many pixels mapping to the same 3D point or whether PnP fails to converge on held-out query images.

Figures

Figures reproduced from arXiv: 2605.07351 by Jae-Pil Heo, Miso Lee, Sangeek Hyun, Yerim Jeon.

Figure 1
Figure 1. Figure 1: Motivation and experimental results. (a) Each 3D Gaussian covers a volumetric region and can thus correspond to multiple image pixels. However, PnP takes 2D-3D point correspondences as input; when each Gaussian is approximated by a single 3D point (e.g., its center), this leads to many￾to-one correspondences. We resolve this by decomposing Gaussians into smaller ones, increasing valid one-to-one correspond… view at source ↗
Figure 2
Figure 2. Figure 2: PlugGS-Loc pipeline. The figure illustrates how to construct GSFFs for localization when pretrained RGB 3DGS and training views are given. (i) For each view Ii , we render and identify Gaussians primarily rasterized above the weight threshold τ in Eq. 1 for each pixel, and select a representative image-Gaussian weight among them for memory efficiency as in Eq. 2. After iterating overall training views, the… view at source ↗
Figure 3
Figure 3. Figure 3: Ablation on mapping hyperparameters. We compare the average per-scene median translation error under varying weight thresholds τ , splitting factor β, and the map size on Cambridge Landmarks [10]. (a) Mild pruning causes feature blurring and aggressive pruning limits coverage of the scene, bringing suboptimal results. (b) Localization accuracy increases monotonically with β within the feasible region, sugg… view at source ↗
read the original abstract

While Gaussian Splatting-based Feature Fields (GSFFs) have shown promise for visual localization, this paper highlights that photometrically optimized GSFFs are inherently ill-suited for 2D-3D matching. The volumetric extent of each Gaussian induces many-to-one pixel-to-point mappings that destabilize PnP-based pose estimation, while photometric optimization gives rise to superfluous Gaussians devoid of multi-view consistency. To address these issues, we propose SplitGS-Loc, a localization-specialized GSFFs construction framework that disambiguates 2D-3D correspondences by exploiting Gaussian attributes. Our key design, Mixture-of-Gaussians-based splitting, decomposes each Gaussian into smaller Gaussians, replacing ambiguous many-to-one with precise one-to-one correspondences. In parallel, we exploit composition weights from GS rasterization to select Gaussians that significantly and consistently contribute across multiple views and aggregate discriminative features through strong pixel-Gaussian associations, enforcing multi-view consistency. The resulting compact yet discriminative feature fields enable stable PnP convergence, achieving state-of-the-art performance on localization benchmarks. Extensive experiments validate that SplitGS-Loc extends the utility of photometric GSFFs to accurate and efficient localization by exploiting Gaussian attributes, without per-scene training or iterative pose refinement.

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

2 major / 2 minor

Summary. The paper claims that photometrically optimized Gaussian Splatting-based Feature Fields (GSFFs) are ill-suited for visual localization due to volumetric many-to-one pixel-to-Gaussian mappings that destabilize PnP and superfluous Gaussians lacking multi-view consistency. It proposes SplitGS-Loc, which uses Mixture-of-Gaussians splitting to decompose Gaussians into smaller ones for precise one-to-one correspondences and composition-weight selection from rasterization to retain only strongly contributing multi-view consistent Gaussians while aggregating discriminative features. The resulting compact feature fields enable stable PnP and achieve state-of-the-art performance on localization benchmarks without per-scene training or iterative refinement.

Significance. If the two core operations (MoG splitting and weight selection) reliably produce the claimed disambiguation and consistency without side effects, the work would meaningfully extend GSFF utility to accurate visual localization by exploiting existing Gaussian attributes rather than requiring new training regimes. This could improve efficiency in pose estimation pipelines that rely on 2D-3D matching.

major comments (2)
  1. [§3.2] §3.2 (Mixture-of-Gaussians-based splitting): The central disambiguation claim rests on splitting converting many-to-one mappings to precise one-to-one correspondences. However, no analysis or bound is supplied showing that projected sub-Gaussians cannot still overlap in the image plane (e.g., near surfaces or under viewpoint change), which would leave residual ambiguities and undermine PnP stability. This assumption is load-bearing for the abstract's performance claim.
  2. [§3.3] §3.3 (composition-weight selection): The selection step assumes high-contribution Gaussians (via rasterization weights) are exactly the multi-view-consistent and discriminative ones. No derivation, ablation, or quantitative check is provided that this does not discard useful features or retain inconsistent ones, which directly affects the 'compact yet discriminative' field needed for the SOTA localization result.
minor comments (2)
  1. [Abstract] Abstract: The statement of 'state-of-the-art performance' would benefit from naming the specific benchmarks and reporting at least one key metric (e.g., median translation/rotation error) to allow immediate assessment of the improvement magnitude.
  2. [§3] Notation: The distinction between original Gaussians and sub-Gaussians after splitting should be made explicit in equations and figures to avoid reader confusion when tracing the correspondence mapping.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and the opportunity to clarify our contributions. We address each major comment below.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (Mixture-of-Gaussians-based splitting): The central disambiguation claim rests on splitting converting many-to-one mappings to precise one-to-one correspondences. However, no analysis or bound is supplied showing that projected sub-Gaussians cannot still overlap in the image plane (e.g., near surfaces or under viewpoint change), which would leave residual ambiguities and undermine PnP stability. This assumption is load-bearing for the abstract's performance claim.

    Authors: We agree that a theoretical bound on residual overlaps would strengthen the claim. However, the splitting is performed by modeling each Gaussian as a mixture and decomposing it into sub-Gaussians with reduced scale, which empirically reduces the number of Gaussians projecting to the same pixel. Our experiments show that this leads to more stable PnP without requiring additional training. In the revised manuscript, we will include an analysis of the overlap reduction before and after splitting to address this concern. revision: partial

  2. Referee: [§3.3] §3.3 (composition-weight selection): The selection step assumes high-contribution Gaussians (via rasterization weights) are exactly the multi-view-consistent and discriminative ones. No derivation, ablation, or quantitative check is provided that this does not discard useful features or retain inconsistent ones, which directly affects the 'compact yet discriminative' field needed for the SOTA localization result.

    Authors: The composition weights from rasterization directly measure the contribution of each Gaussian to the rendered feature at each pixel. By selecting those with high weights across multiple views, we retain Gaussians that are consistently visible and contributing, which correlates with multi-view consistency. The manuscript includes ablations demonstrating the impact of this selection on localization performance. To further validate, we will add quantitative metrics on the multi-view consistency of selected vs. discarded Gaussians in the revision. revision: partial

Circularity Check

0 steps flagged

No circularity: constructive pipeline validated externally

full rationale

The paper presents SplitGS-Loc as a constructive pipeline that decomposes Gaussians via Mixture-of-Gaussians splitting and selects via composition weights drawn from existing rasterization attributes. These steps are defined directly from Gaussian Splatting properties without any equations that reduce the final feature field or PnP performance metric to a fitted quantity defined on the same localization data. Claims of stable convergence and SOTA results are supported by benchmark experiments rather than by construction or self-referential definitions. No load-bearing premise collapses to a self-citation chain or ansatz smuggled from prior author work; the derivation remains independent of the target performance numbers.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach rests on two domain assumptions about Gaussian Splatting behavior and one design choice whose effectiveness is asserted rather than derived from first principles.

axioms (2)
  • domain assumption Volumetric extent of each Gaussian induces many-to-one pixel-to-point mappings
    Invoked in the abstract as the root cause of unstable PnP.
  • domain assumption Photometric optimization produces superfluous Gaussians that lack multi-view consistency
    Stated as an inherent property of standard GSFF training.

pith-pipeline@v0.9.0 · 5539 in / 1183 out tokens · 36118 ms · 2026-05-11T02:17:08.618162+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

55 extracted references · 55 canonical work pages

  1. [1]

    Complete solution classification for the perspective-three-point problem.IEEE transactions on pattern analysis and machine intelligence, 25(8):930–943, 2003

    Xiao-Shan Gao, Xiao-Rong Hou, Jianliang Tang, and Hang-Fei Cheng. Complete solution classification for the perspective-three-point problem.IEEE transactions on pattern analysis and machine intelligence, 25(8):930–943, 2003

    work page 2003

  2. [2]

    Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography.Communications of the ACM, 24(6):381–395, 1981

    Martin A Fischler and Robert C Bolles. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography.Communications of the ACM, 24(6):381–395, 1981

    work page 1981

  3. [3]

    3d gaussian splatting for real-time radiance field rendering.ACM Trans

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Trans. Graph., 42(4):139–1, 2023

    work page 2023

  4. [4]

    Lerf: Language embedded radiance fields

    Justin Kerr, Chung Min Kim, Ken Goldberg, Angjoo Kanazawa, and Matthew Tancik. Lerf: Language embedded radiance fields. InProceedings of the IEEE/CVF international conference on computer vision, pages 19729–19739, 2023

    work page 2023

  5. [5]

    Feature 3dgs: Supercharging 3d gaussian splatting to enable distilled feature fields

    Shijie Zhou, Haoran Chang, Sicheng Jiang, Zhiwen Fan, Zehao Zhu, Dejia Xu, Pradyumna Chari, Suya You, Zhangyang Wang, and Achuta Kadambi. Feature 3dgs: Supercharging 3d gaussian splatting to enable distilled feature fields. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21676–21685, 2024

    work page 2024

  6. [6]

    Langsplat: 3d language gaus- sian splatting

    Minghan Qin, Wanhua Li, Jiawei Zhou, Haoqian Wang, and Hanspeter Pfister. Langsplat: 3d language gaus- sian splatting. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20051–20060, 2024

    work page 2024

  7. [7]

    Gsplatloc: Grounding keypoint descriptors into 3d gaussian splatting for improved visual localization

    Gennady Sidorov, Malik Mohrat, Denis Gridusov, Ruslan Rakhimov, and Sergey Kolyubin. Gsplatloc: Grounding keypoint descriptors into 3d gaussian splatting for improved visual localization. In2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 12601–12607. IEEE, 2025

    work page 2025

  8. [8]

    Gaussian splatting feature fields for (privacy- preserving) visual localization

    Maxime Pietrantoni, Gabriela Csurka, and Torsten Sattler. Gaussian splatting feature fields for (privacy- preserving) visual localization. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 1082–1092, 2025

    work page 2025

  9. [9]

    From sparse to dense: Camera relocalization with scene-specific detector from feature gaussian splatting

    Zhiwei Huang, Hailin Yu, Yichun Shentu, Jin Yuan, and Guofeng Zhang. From sparse to dense: Camera relocalization with scene-specific detector from feature gaussian splatting. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 27059–27069, 2025

    work page 2025

  10. [10]

    Posenet: A convolutional network for real-time 6-dof camera relocalization

    Alex Kendall, Matthew Grimes, and Roberto Cipolla. Posenet: A convolutional network for real-time 6-dof camera relocalization. InProceedings of the IEEE international conference on computer vision, pages 2938–2946, 2015

    work page 2015

  11. [11]

    Spec-Gaussian: Anisotropic view-dependent appear- ance for 3D gaussian splatting

    Ziyi Yang, Xinyu Gao, Yang-Tian Sun, Yi-Hua Huang, Xiaoyang Lyu, Wen Zhou, Shaohui Jiao, Xiaojuan Qi, and Xiaogang Jin. Spec-Gaussian: Anisotropic view-dependent appear- ance for 3D gaussian splatting. InAdvances in Neural Information Processing Systems, vol- ume 37, 2024. URL https://proceedings.neurips.cc/paper_files/paper/2024/hash/ 708e0d691a22212e1e3...

    work page 2024

  12. [12]

    Pup 3d-gs: Principled uncertainty pruning for 3d gaussian splatting

    Alex Hanson, Allen Tu, Vasu Singla, Mayuka Jayawardhana, Matthias Zwicker, and Tom Goldstein. Pup 3d-gs: Principled uncertainty pruning for 3d gaussian splatting. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 5949–5958, 2025

    work page 2025

  13. [13]

    Opengaussian: Towards point-level 3d gaussian-based open vocabulary understanding.Advances in Neural Information Processing Systems, 37:19114–19138, 2024

    Yanmin Wu, Jiarui Meng, Haijie Li, Chenming Wu, Yahao Shi, Xinhua Cheng, Chen Zhao, Haocheng Feng, Errui Ding, Jingdong Wang, et al. Opengaussian: Towards point-level 3d gaussian-based open vocabulary understanding.Advances in Neural Information Processing Systems, 37:19114–19138, 2024

    work page 2024

  14. [14]

    Kim Jun-Seong, GeonU Kim, Kim Yu-Ji, Yu-Chiang Frank Wang, Jaesung Choe, and Tae-Hyun Oh. Dr. splat: Directly referring 3d gaussian splatting via direct language embedding registration. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 14137–14146, 2025

    work page 2025

  15. [15]

    Scene coordinate regression forests for camera relocalization in rgb-d images

    Jamie Shotton, Ben Glocker, Christopher Zach, Shahram Izadi, Antonio Criminisi, and Andrew Fitzgibbon. Scene coordinate regression forests for camera relocalization in rgb-d images. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 2930–2937, 2013

    work page 2013

  16. [16]

    Location recognition using prioritized feature matching

    Yunpeng Li, Noah Snavely, and Daniel P Huttenlocher. Location recognition using prioritized feature matching. InEuropean conference on computer vision, pages 791–804. Springer, 2010. 10

    work page 2010

  17. [17]

    Efficient & effective prioritized matching for large- scale image-based localization.IEEE transactions on pattern analysis and machine intelligence, 39(9): 1744–1756, 2016

    Torsten Sattler, Bastian Leibe, and Leif Kobbelt. Efficient & effective prioritized matching for large- scale image-based localization.IEEE transactions on pattern analysis and machine intelligence, 39(9): 1744–1756, 2016

    work page 2016

  18. [18]

    Are large-scale 3d models really necessary for accurate visual localization? InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 1637–1646, 2017

    Torsten Sattler, Akihiko Torii, Josef Sivic, Marc Pollefeys, Hajime Taira, Masatoshi Okutomi, and Tomas Pajdla. Are large-scale 3d models really necessary for accurate visual localization? InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 1637–1646, 2017

    work page 2017

  19. [19]

    From coarse to fine: Robust hierarchical localization at large scale

    Paul-Edouard Sarlin, Cesar Cadena, Roland Siegwart, and Marcin Dymczyk. From coarse to fine: Robust hierarchical localization at large scale. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12716–12725, 2019

    work page 2019

  20. [20]

    Robust im- age retrieval-based visual localization using kapture.arXiv preprint arXiv:2007.13867, 2020

    Martin Humenberger, Yohann Cabon, Nicolas Guerin, Julien Morat, Vincent Leroy, Jérôme Revaud, Philippe Rerole, Noé Pion, Cesar De Souza, and Gabriela Csurka. Robust image retrieval-based visual localization using kapture.arXiv preprint arXiv:2007.13867, 2020

    work page arXiv 2007

  21. [21]

    Back to the feature: Learning robust camera localization from pixels to pose

    Paul-Edouard Sarlin, Ajaykumar Unagar, Mans Larsson, Hugo Germain, Carl Toft, Viktor Larsson, Marc Pollefeys, Vincent Lepetit, Lars Hammarstrand, Fredrik Kahl, et al. Back to the feature: Learning robust camera localization from pixels to pose. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3247–3257, 2021

    work page 2021

  22. [22]

    Learning to produce semi-dense correspondences for visual localization

    Khang Truong Giang, Soohwan Song, and Sungho Jo. Learning to produce semi-dense correspondences for visual localization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19468–19478, 2024

    work page 2024

  23. [23]

    24/7 place recogni- tion by view synthesis

    Akihiko Torii, Relja Arandjelovic, Josef Sivic, Masatoshi Okutomi, and Tomas Pajdla. 24/7 place recogni- tion by view synthesis. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 1808–1817, 2015

    work page 2015

  24. [24]

    Netvlad: Cnn architecture for weakly supervised place recognition

    Relja Arandjelovic, Petr Gronat, Akihiko Torii, Tomas Pajdla, and Josef Sivic. Netvlad: Cnn architecture for weakly supervised place recognition. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 5297–5307, 2016

    work page 2016

  25. [25]

    Distinctive image features from scale-invariant keypoints.International journal of computer vision, 60(2):91–110, 2004

    David G Lowe. Distinctive image features from scale-invariant keypoints.International journal of computer vision, 60(2):91–110, 2004

    work page 2004

  26. [26]

    Superglue: Learning feature matching with graph neural networks

    Paul-Edouard Sarlin, Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. Superglue: Learning feature matching with graph neural networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4938–4947, 2020

    work page 2020

  27. [27]

    Superpoint: Self-supervised interest point detection and description

    Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. Superpoint: Self-supervised interest point detection and description. InProceedings of the IEEE conference on computer vision and pattern recognition workshops, pages 224–236, 2018

    work page 2018

  28. [28]

    Learning less is more-6d camera localization via 3d surface regression

    Eric Brachmann and Carsten Rother. Learning less is more-6d camera localization via 3d surface regression. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 4654–4662, 2018

    work page 2018

  29. [29]

    Visual camera re-localization from rgb and rgb-d images using dsac

    Eric Brachmann and Carsten Rother. Visual camera re-localization from rgb and rgb-d images using dsac. IEEE transactions on pattern analysis and machine intelligence, 44(9):5847–5865, 2021

    work page 2021

  30. [30]

    Accelerated coordinate encoding: Learning to relocalize in minutes using rgb and poses

    Eric Brachmann, Tommaso Cavallari, and Victor Adrian Prisacariu. Accelerated coordinate encoding: Learning to relocalize in minutes using rgb and poses. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5044–5053, 2023

    work page 2023

  31. [31]

    Glace: Global local accelerated coordinate encoding

    Fangjinhua Wang, Xudong Jiang, Silvano Galliani, Christoph V ogel, and Marc Pollefeys. Glace: Global local accelerated coordinate encoding. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21562–21571, 2024

    work page 2024

  32. [32]

    Geometric loss functions for camera pose regression with deep learning

    Alex Kendall and Roberto Cipolla. Geometric loss functions for camera pose regression with deep learning. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 5974–5983, 2017

    work page 2017

  33. [33]

    Geometry-aware learning of maps for camera localization

    Samarth Brahmbhatt, Jinwei Gu, Kihwan Kim, James Hays, and Jan Kautz. Geometry-aware learning of maps for camera localization. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 2616–2625, 2018. 11

    work page 2018

  34. [34]

    Understanding the limitations of cnn-based absolute camera pose regression

    Torsten Sattler, Qunjie Zhou, Marc Pollefeys, and Laura Leal-Taixe. Understanding the limitations of cnn-based absolute camera pose regression. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3302–3312, 2019

    work page 2019

  35. [35]

    Reloc3r: Large-scale training of relative camera pose regression for generalizable, fast, and accurate visual localization

    Siyan Dong, Shuzhe Wang, Shaohui Liu, Lulu Cai, Qingnan Fan, Juho Kannala, and Yanchao Yang. Reloc3r: Large-scale training of relative camera pose regression for generalizable, fast, and accurate visual localization. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 16739–16752, 2025

    work page 2025

  36. [36]

    Map-relative pose regression for visual re-localization

    Shuai Chen, Tommaso Cavallari, Victor Adrian Prisacariu, and Eric Brachmann. Map-relative pose regression for visual re-localization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20665–20674, 2024

    work page 2024

  37. [37]

    Coarse-to-fine multi-scene pose regression with transformers

    Yoli Shavit, Ron Ferens, and Yosi Keller. Coarse-to-fine multi-scene pose regression with transformers. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023

    work page 2023

  38. [38]

    Ufvl-net: A unified framework for visual localization across multiple indoor scenes.IEEE Transactions on Instrumentation and Measurement, 72:1–16, 2023

    Tao Xie, Zhiqiang Jiang, Shuozhan Li, Yukun Zhang, Kun Dai, Ke Wang, Ruifeng Li, and Lijun Zhao. Ufvl-net: A unified framework for visual localization across multiple indoor scenes.IEEE Transactions on Instrumentation and Measurement, 72:1–16, 2023

    work page 2023

  39. [39]

    Activating self-attention for multi-scene absolute pose regression

    Miso Lee, Jihwan Kim, and Jae-Pil Heo. Activating self-attention for multi-scene absolute pose regression. Advances in Neural Information Processing Systems, 37:38508–38529, 2024

    work page 2024

  40. [40]

    Direct-posenet: Absolute pose regression with photometric consistency

    Shuai Chen, Zirui Wang, and Victor Prisacariu. Direct-posenet: Absolute pose regression with photometric consistency. In2021 International Conference on 3D Vision (3DV), pages 1175–1185. IEEE, 2021

    work page 2021

  41. [41]

    Lens: Localization enhanced by nerf synthesis

    Arthur Moreau, Nathan Piasco, Dzmitry Tsishkou, Bogdan Stanciulescu, and Arnaud de La Fortelle. Lens: Localization enhanced by nerf synthesis. InConference on Robot Learning, pages 1347–1356. PMLR, 2022

    work page 2022

  42. [42]

    Dfnet: Enhance absolute pose regression with direct feature matching

    Shuai Chen, Xinghui Li, Zirui Wang, and Victor A Prisacariu. Dfnet: Enhance absolute pose regression with direct feature matching. InEuropean Conference on Computer Vision, pages 1–17, 2022

    work page 2022

  43. [43]

    Crossfire: Camera relocalization on self-supervised features from an implicit representation

    Arthur Moreau, Nathan Piasco, Moussab Bennehar, Dzmitry Tsishkou, Bogdan Stanciulescu, and Arnaud de La Fortelle. Crossfire: Camera relocalization on self-supervised features from an implicit representation. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 252–262, 2023

    work page 2023

  44. [44]

    Nerf: Representing scenes as neural radiance fields for view synthesis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. InEuropean Conference on Computer Vision, pages 405–421. Springer, 2020

    work page 2020

  45. [45]

    Nerf in the wild: Neural radiance fields for unconstrained photo collections

    Ricardo Martin-Brualla, Noha Radwan, Mehdi SM Sajjadi, Jonathan T Barron, Alexey Dosovitskiy, and Daniel Duckworth. Nerf in the wild: Neural radiance fields for unconstrained photo collections. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7210–7219, 2021

    work page 2021

  46. [46]

    Instant neural graphics primitives with a multiresolution hash encoding.ACM transactions on graphics (TOG), 41(4):1–15, 2022

    Thomas Müller, Alex Evans, Christoph Schied, and Alexander Keller. Instant neural graphics primitives with a multiresolution hash encoding.ACM transactions on graphics (TOG), 41(4):1–15, 2022

    work page 2022

  47. [47]

    Pnerfloc: Visual localization with point-based neural radiance fields

    Boming Zhao, Luwei Yang, Mao Mao, Hujun Bao, and Zhaopeng Cui. Pnerfloc: Visual localization with point-based neural radiance fields. InProceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 7450–7459, 2024

    work page 2024

  48. [48]

    The nerfect match: Exploring nerf features for visual localization

    Qunjie Zhou, Maxim Maximov, Or Litany, and Laura Leal-Taixé. The nerfect match: Exploring nerf features for visual localization. InEuropean Conference on Computer Vision, pages 108–127, 2024

    work page 2024

  49. [49]

    GS-CPR: Efficient camera pose refinement via 3d gaussian splatting

    Changkun Liu, Shuai Chen, Yash Sanjay Bhalgat, Siyan HU, Ming Cheng, Zirui Wang, Victor Adrian Prisacariu, and Tristan Braud. GS-CPR: Efficient camera pose refinement via 3d gaussian splatting. InThe Thirteenth International Conference on Learning Representations, 2025. URL https://openreview. net/forum?id=mP7uV59iJM

    work page 2025

  50. [50]

    Generaliz- able visual localization for gaussian splatting scene representations

    Fadi Khatib, Dror Moran, Guy Trostianetsky, Yoni Kasten, Meirav Galun, and Ronen Basri. Generaliz- able visual localization for gaussian splatting scene representations. InProceedings of the IEEE/CVF International Conference on Computer Vision Workshops, 2025

    work page 2025

  51. [51]

    Adversarial exploitation of data diversity improves visual localization

    Sihang Li, Siqi Tan, Bowen Chang, Jing Zhang, Chen Feng, and Yiming Li. Adversarial exploitation of data diversity improves visual localization. InInternational Conference on Computer Vision (ICCV), 2025. 12

    work page 2025

  52. [52]

    3d gaussian splatting based scene- independent relocalization with unidirectional and bidirectional feature fusion

    Junyi Wang, Yuze Wang, Wantong Duan, Meng Wang, and Yue Qi. 3d gaussian splatting based scene- independent relocalization with unidirectional and bidirectional feature fusion. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025. URL https://openreview.net/forum? id=ewgZItWaHh

    work page 2025

  53. [53]

    Ludvig: Learning-free uplifting of 2d visual features to gaussian splatting scenes

    Juliette Marrie, Romain Ménégaux, Michael Arbel, Diane Larlus, and Julien Mairal. Ludvig: Learning-free uplifting of 2d visual features to gaussian splatting scenes. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 7440–7450, 2025

    work page 2025

  54. [54]

    On the Limits of Pseudo Ground Truth in Visual Camera Re-Localisation

    Eric Brachmann, Martin Humenberger, Carsten Rother, and Torsten Sattler. On the Limits of Pseudo Ground Truth in Visual Camera Re-Localisation. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021

    work page 2021

  55. [55]

    PoseLib - Minimal Solvers for Camera Pose Estimation, 2020

    Viktor Larsson and contributors. PoseLib - Minimal Solvers for Camera Pose Estimation, 2020. URL https://github.com/vlarsson/PoseLib. 13 A Derivation of the Splitting Parameters This section provides a detailed derivation of the mixture parameters obtained by matching the first four moments. As mentioned in the main paper, we define a parent 1D Gaussian d...

    work page 2020