arxiv: 2604.04642 · v1 · submitted 2026-04-06 · 💻 cs.RO

Recognition: 2 theorem links

· Lean Theorem

WaterSplat-SLAM: Photorealistic Monocular SLAM in Underwater Environment

Kangxu Wang , Shaofeng Zou , Chenxing Jiang , Yixiang Dai , Siang Chen , Shaojie Shen , Guijin Wang

Authors on Pith no claims yet

Pith reviewed 2026-05-10 19:31 UTC · model grok-4.3

classification 💻 cs.RO

keywords underwater SLAMmonocular SLAMGaussian splattingsemantic filteringphotorealistic mappingdense reconstructionunderwater roboticsmedium modeling

0 comments

The pith

WaterSplat-SLAM achieves robust pose estimation and photorealistic dense mapping in underwater environments by coupling semantic medium filtering into two-view 3D reconstruction and using an online medium-aware Gaussian map.

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

The paper targets monocular SLAM in underwater settings, where scattering and absorption from water degrade depth estimates and visual quality in standard systems. It integrates semantic medium filtering directly into the two-view 3D reconstruction step to adapt camera tracking and depth recovery to underwater conditions. A semantic-guided rendering process then works with an adaptive map management strategy built around an online medium-aware Gaussian map that represents the scene in a compact yet visually accurate way. This produces maps suitable for navigation and inspection where earlier underwater SLAM approaches produced lower-fidelity results. The work matters for any application that needs reliable visual positioning and detailed scene reconstruction from a single camera moving through water.

Core claim

WaterSplat-SLAM achieves robust pose estimation and photorealistic dense mapping in underwater environments by coupling semantic medium filtering into two-view 3D reconstruction prior to enable underwater-adapted camera tracking and depth estimation, and by presenting a semantic-guided rendering and adaptive map management strategy with an online medium-aware Gaussian map that models the underwater environment in a photorealistic and compact manner.

What carries the argument

The online medium-aware Gaussian map combined with semantic medium filtering, which adapts the representation to water effects during tracking, depth estimation, and rendering.

If this is right

Camera tracking remains accurate despite water-induced distortions that defeat standard monocular SLAM.
Dense maps support high-fidelity visual rendering suitable for inspection tasks.
The representation stays compact while incorporating medium effects through adaptive management.
Improved mapping directly benefits autonomous underwater vehicle navigation and marine archaeology documentation.

Where Pith is reading between the lines

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

The same semantic filtering idea for handling scattering media could transfer to other low-visibility domains such as fog or dust.
Adding the medium-aware map to multi-camera or sensor-fusion setups might reduce reliance on monocular depth cues.
Adaptive map management offers a template for SLAM systems facing gradual environmental changes beyond water.

Load-bearing premise

Semantic medium filtering can be effectively coupled into two-view 3D reconstruction to enable accurate underwater-adapted camera tracking and depth estimation without introducing errors that degrade overall performance.

What would settle it

Running the system on an underwater dataset with high or varying turbidity where the reported pose estimation error exceeds that of prior monocular SLAM baselines or where PSNR/SSIM metrics for rendered views fall below competing methods would falsify the central performance claim.

Figures

Figures reproduced from arXiv: 2604.04642 by Chenxing Jiang, Guijin Wang, Kangxu Wang, Shaofeng Zou, Shaojie Shen, Siang Chen, Yixiang Dai.

**Figure 1.** Figure 1: System overview of WaterSplat-SLAM: The system takes an RGB sequence as input and generates an online medium-aware Gaussian map. The [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Figure 2: Illustration of medium-aware Gaussian mapping: Encoded ray vectors [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: Gaussian primitives merging pipeline: When consecutive keyframes [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: Detailed comparison of reconstruction results for Curacao, JapRedSea, and Panama sequences on SeaThru-NeRF dataset. All three sequences exhibit [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

**Figure 5.** Figure 5: Detailed reconstruction comparisons for Big [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

**Figure 6.** Figure 6: NVS result w/o Water Mask module. adjustment and merging strategy. We conduct our ablation study on each module of the four modules in Panama sequence and our dataset. For the Panama scene, as in Table. V, our semanticguided rendering module contributes to the system performance crucially. For Pool loop scene, as in Table. V, our primitives adjustment module also contributes to the system performance cru… view at source ↗

read the original abstract

Underwater monocular SLAM is a challenging problem with applications from autonomous underwater vehicles to marine archaeology. However, existing underwater SLAM methods struggle to produce maps with high-fidelity rendering. In this paper, we propose WaterSplat-SLAM, a novel monocular underwater SLAM system that achieves robust pose estimation and photorealistic dense mapping. Specifically, we couple semantic medium filtering into two-view 3D reconstruction prior to enable underwater-adapted camera tracking and depth estimation. Furthermore, we present a semantic-guided rendering and adaptive map management strategy with an online medium-aware Gaussian map, modeling underwater environment in a photorealistic and compact manner. Experiments on multiple underwater datasets demonstrate that WaterSplat-SLAM achieves robust camera tracking and high-fidelity rendering in underwater environments.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

WaterSplat-SLAM adds semantic medium filtering to two-view reconstruction and medium-aware Gaussians for underwater monocular SLAM, but the validation leaves the error-propagation risk unaddressed.

read the letter

WaterSplat-SLAM couples semantic medium filtering into the two-view geometry stage and then runs semantic-guided rendering on an online medium-aware Gaussian map. That combination is the actual new element; prior underwater SLAM work has not packaged the filtering and the adaptive Gaussian management together in this way. The approach targets a concrete gap—standard monocular methods lose accuracy and photorealism once scattering and attenuation enter the picture—and the reported experiments on several underwater datasets show usable tracking plus dense rendering, which is a practical step forward for AUV mapping or archaeology tasks. The architecture itself is straightforward to follow and builds directly on recent Gaussian splatting pipelines with only the domain-specific additions. The soft spot is exactly the one the stress-test flags. Semantic segmentation underwater is brittle under color shift and low visibility, and any mislabeled pixels would corrupt the filtered correspondences and depth priors that feed the tracker and the map. The paper gives no ablation that isolates the filter’s contribution, no quantitative bound on segmentation accuracy, and no failure-case analysis when the medium labels are noisy. Without those, it is hard to tell whether the claimed robustness comes from the new components or from dataset-specific tuning. The citation pattern is normal for the subfield and does not hide gaps. This paper is for the narrow group working on dense underwater SLAM and marine robotics; a reader already building Gaussian-based systems for AUVs could extract the filtering and map-management tricks and test them themselves. It is worth sending to peer review because the problem is real, the method is concrete, and the experiments at least demonstrate feasibility on real data, even though the referee will almost certainly ask for the missing robustness checks.

Referee Report

2 major / 1 minor

Summary. The paper proposes WaterSplat-SLAM, a monocular SLAM system for underwater environments that achieves robust pose estimation and photorealistic dense mapping. It couples semantic medium filtering into two-view 3D reconstruction to produce an underwater-adapted prior for camera tracking and depth estimation, and introduces semantic-guided rendering together with adaptive map management using an online medium-aware Gaussian map. Experiments on multiple underwater datasets are reported to demonstrate the system's performance.

Significance. If the central claims are substantiated with quantitative evidence, the work would represent a meaningful advance in underwater SLAM by addressing medium-induced distortions (scattering, attenuation, color shift) through semantic filtering and Gaussian splatting, enabling higher-fidelity rendering than prior underwater methods. This has direct relevance to AUV navigation and marine archaeology. The approach of integrating semantics into both geometry estimation and map management is conceptually promising, but its practical impact cannot be assessed without the missing evaluation details.

major comments (2)

[Abstract] Abstract: the central claim that the system 'achieves robust camera tracking and high-fidelity rendering' rests on experiments on 'multiple underwater datasets,' yet no quantitative metrics (e.g., ATE, RPE, PSNR, SSIM), error bars, baseline comparisons, or ablation studies isolating the semantic medium filter are provided. This absence prevents verification that the filter improves rather than degrades two-view geometry under low-visibility conditions.
[Abstract] Abstract (description of semantic medium filtering): the assumption that coupling semantic medium filtering into two-view 3D reconstruction yields an 'underwater-adapted' prior without introducing new failure modes is load-bearing, but no analysis of segmentation error propagation, no bound on segmentation accuracy, and no discussion of how mislabeled medium pixels affect subsequent Gaussian optimization or adaptive map management are supplied.

minor comments (1)

[Abstract] Abstract: the phrase 'online medium-aware Gaussian map' is introduced without a brief definition or reference to the underlying representation (e.g., 3D Gaussian splatting), which would help readers immediately grasp the technical contribution.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and recommendation for major revision. The points raised highlight areas where additional quantitative evidence and analysis will strengthen the manuscript. We address each major comment below and will incorporate the suggested revisions.

read point-by-point responses

Referee: [Abstract] Abstract: the central claim that the system 'achieves robust camera tracking and high-fidelity rendering' rests on experiments on 'multiple underwater datasets,' yet no quantitative metrics (e.g., ATE, RPE, PSNR, SSIM), error bars, baseline comparisons, or ablation studies isolating the semantic medium filter are provided. This absence prevents verification that the filter improves rather than degrades two-view geometry under low-visibility conditions.

Authors: We agree that the abstract would be more informative with explicit metrics. The full manuscript reports ATE, RPE, PSNR, SSIM, and LPIPS results on multiple underwater datasets with baseline comparisons in the Experiments section. To directly address this comment, we will revise the abstract to highlight key quantitative improvements and expand the experimental section with error bars, statistical details, and a dedicated ablation isolating the semantic medium filter's contribution to two-view geometry under low-visibility conditions. revision: yes
Referee: [Abstract] Abstract (description of semantic medium filtering): the assumption that coupling semantic medium filtering into two-view 3D reconstruction yields an 'underwater-adapted' prior without introducing new failure modes is load-bearing, but no analysis of segmentation error propagation, no bound on segmentation accuracy, and no discussion of how mislabeled medium pixels affect subsequent Gaussian optimization or adaptive map management are supplied.

Authors: This is a valid observation. The current version emphasizes the benefits of semantic filtering but lacks explicit error analysis. In the revision, we will add a dedicated discussion covering bounds on segmentation accuracy (via IoU on annotated data), propagation of mislabeling errors to depth estimation and Gaussian optimization, and handling strategies in adaptive map management such as confidence-based weighting. This will include potential failure cases and mitigation to substantiate the assumption. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected in WaterSplat-SLAM derivation

full rationale

The paper proposes a novel monocular underwater SLAM system by introducing semantic medium filtering coupled into two-view 3D reconstruction, semantic-guided rendering, and an adaptive map management strategy with an online medium-aware Gaussian map. These elements are presented as original methodological contributions, with performance claims supported by experiments on multiple underwater datasets rather than by reducing to fitted parameters, self-definitions, or load-bearing self-citations. No equations or steps in the abstract or described chain equate outputs to inputs by construction; the derivation chain remains self-contained through independent algorithmic proposals and empirical validation.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based on abstract only; no explicit free parameters, axioms, or invented entities are detailed. The system likely builds on standard SLAM and Gaussian splatting assumptions but introduces 'medium-aware' adaptations whose specifics cannot be audited without the full text.

pith-pipeline@v0.9.0 · 5446 in / 1133 out tokens · 64387 ms · 2026-05-10T19:31:50.309351+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Foundation/AlphaCoordinateFixation.lean alpha_pin_under_high_calibration unclear

?

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

Medium Network (MLP) for acquiring medium parameters σ_attn, σ_bs, and c_med

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

35 extracted references · 13 canonical work pages

[1]

Object perception in underwater environments: a survey on sensors and sensing methodologies,

D. Q. Huy, N. Sadjoli, A. B. Azam, B. Elhadidi, Y . Cai, and G. Seet, “Object perception in underwater environments: a survey on sensors and sensing methodologies,”Ocean Engineering, vol. 267, p. 113202, 2023

2023
[2]

Qualitative evalua- tion of state-of-the-art dso and orb-slam-based monocular visual slam algorithms for underwater applications,

J. Drupt, C. Dune, A. I. Comport, and V . Hugel, “Qualitative evalua- tion of state-of-the-art dso and orb-slam-based monocular visual slam algorithms for underwater applications,” inOCEANS 2023 - Limerick, 2023, pp. 1–7

2023
[3]

Monocular orb-slam appli- cation in underwater scenarios,

F. Hidalgo, C. Kahlefendt, and T. Br ¨aunl, “Monocular orb-slam appli- cation in underwater scenarios,” in2018 OCEANS - MTS/IEEE Kobe Techno-Oceans (OTO), 2018, pp. 1–4

2018
[4]

Detecting loop closure using enhanced image for underwater vins-mono,

H. Zhao, R. Zheng, M. Liu, and S. Zhang, “Detecting loop closure using enhanced image for underwater vins-mono,” inGlobal Oceans 2020: Singapore–US Gulf Coast. IEEE, 2020, pp. 1–6

2020
[5]

Acoustic camera-based pose graph slam for dense 3-d mapping in underwater environments,

Y . Wang, Y . Ji, H. Woo, Y . Tamura, H. Tsuchiya, A. Yamashita, and H. Asama, “Acoustic camera-based pose graph slam for dense 3-d mapping in underwater environments,”IEEE Journal of Oceanic Engineering, vol. 46, no. 3, pp. 829–847, 2020

2020
[6]

Hybrid-vins: Underwater tightly coupled hybrid visual inertial dense slam for auv,

Y . Ou, J. Fan, C. Zhou, P. Zhang, and Z.-G. Hou, “Hybrid-vins: Underwater tightly coupled hybrid visual inertial dense slam for auv,” IEEE Transactions on Industrial Electronics, vol. 72, no. 3, pp. 2821– 2831, 2025

2025
[7]

An underwater, fault-tolerant, laser-aided robotic multi-modal dense slam system for continuous underwater in-situ observation,

Y . Ou, J. Fan, C. Zhou, P. Zhang, Z. Shen, Y . Fu, X. Liu, and Z. Hou, “An underwater, fault-tolerant, laser-aided robotic multi-modal dense slam system for continuous underwater in-situ observation,”arXiv preprint arXiv:2504.21826, 2025

work page arXiv 2025
[8]

Structured light-based underwater collision-free navigation and dense mapping system for refined exploration in unknown dark environments,

Y . Ou, J. Fan, C. Zhou, S. Kang, Z. Zhang, Z.-G. Hou, and M. Tan, “Structured light-based underwater collision-free navigation and dense mapping system for refined exploration in unknown dark environments,” IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 55, no. 1, pp. 110–123, 2024

2024
[9]

Gaussian splatting slam,

H. Matsuki, R. Murai, P. H. Kelly, and A. J. Davison, “Gaussian splatting slam,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2024, pp. 18 039–18 048

2024
[10]

Gaussian-slam: Photo-realistic dense slam with gaussian splatting,

V . Yugay, Y . Li, T. Gevers, and M. R. Oswald, “Gaussian-slam: Photo-realistic dense slam with gaussian splatting,” 2024. [Online]. Available: https://arxiv.org/abs/2312.10070

work page arXiv 2024
[11]

Photo-slam: Real- time simultaneous localization and photorealistic mapping for monocular stereo and rgb-d cameras,

H. Huang, L. Li, H. Cheng, and S.-K. Yeung, “Photo-slam: Real- time simultaneous localization and photorealistic mapping for monocular stereo and rgb-d cameras,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 21 584–21 593

2024
[12]

Hi-slam2: Geometry-aware gaussian slam for fast monocular scene reconstruction,

W. Zhang, Q. Cheng, D. Skuddis, N. Zeller, D. Cremers, and N. Haala, “Hi-slam2: Geometry-aware gaussian slam for fast monocular scene reconstruction,”arXiv preprint arXiv:2411.17982, 2024

work page arXiv 2024
[13]

Mvs-gs: High-quality 3d gaussian splatting mapping via online multi-view stereo,

B. Lee, J. Park, K. T. Giang, S. Jo, and S. Song, “Mvs-gs: High-quality 3d gaussian splatting mapping via online multi-view stereo,” 2024. [Online]. Available: https://arxiv.org/abs/2412.19130

work page arXiv 2024
[14]

Svin2: An underwater slam system using sonar, visual, inertial, and depth sensor,

S. Rahman, A. Q. Li, and I. Rekleitis, “Svin2: An underwater slam system using sonar, visual, inertial, and depth sensor,” in2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2019, pp. 1861–1868

2019
[15]

Robust underwater visual slam fusing acoustic sensing,

E. Vargas, R. Scona, J. S. Willners, T. Luczynski, Y . Cao, S. Wang, and Y . R. Petillot, “Robust underwater visual slam fusing acoustic sensing,” in2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2021, pp. 2140–2146

2021
[16]

Sm/vio: Robust underwater state estimation switching between model-based and visual inertial odometry,

B. Joshi, H. Damron, S. Rahman, and I. Rekleitis, “Sm/vio: Robust underwater state estimation switching between model-based and visual inertial odometry,”arXiv preprint arXiv:2304.01988, 2023

work page arXiv 2023
[17]

Hybrid visual inertial odometry for robust underwater estimation,

B. Joshi, C. Bandara, I. Poulakakis, H. G. Tanner, and I. Rekleitis, “Hybrid visual inertial odometry for robust underwater estimation,” in OCEANS 2023-MTS/IEEE US Gulf Coast. IEEE, 2023, pp. 1–7

2023
[18]

Deepvl: Dynamics and inertial measurements- based deep velocity learning for underwater odometry,

M. Singh and K. Alexis, “Deepvl: Dynamics and inertial measurements- based deep velocity learning for underwater odometry,”arXiv preprint arXiv:2502.07726, 2025

work page arXiv 2025
[19]

Real-time monocular visual odometry for turbid and dynamic underwater envi- ronments,

M. Ferrera, J. Moras, P. Trouv ´e-Peloux, and V . Creuze, “Real-time monocular visual odometry for turbid and dynamic underwater envi- ronments,”Sensors, vol. 19, no. 3, p. 687, 2019

2019
[20]

Improving self-consistency in underwa- ter mapping through laser-based loop closure,

T. Hitchcox and J. R. Forbes, “Improving self-consistency in underwa- ter mapping through laser-based loop closure,”IEEE Transactions on Robotics, vol. 39, no. 3, pp. 1873–1892, 2023

2023
[21]

Splat-slam: Globally optimized rgb-only slam with 3d gaussians.arXiv preprint arXiv:2405.16544, 2024

E. Sandstr ¨om, K. Tateno, M. Oechsle, M. Niemeyer, L. Van Gool, M. R. Oswald, and F. Tombari, “Splat-slam: Globally optimized rgb-only slam with 3d gaussians,”arXiv preprint arXiv:2405.16544, 2024

work page arXiv 2024
[22]

Glorie-slam: Globally optimized rgb-only implicit encoding point cloud slam.arXiv preprint arXiv:2403.19549, 2024

G. Zhang, E. Sandstr ¨om, Y . Zhang, M. Patel, L. Van Gool, and M. R. Oswald, “Glorie-slam: Globally optimized rgb-only implicit encoding point cloud slam,”arXiv preprint arXiv:2403.19549, 2024

work page arXiv 2024
[23]

Hi-slam: Monocular real-time dense mapping with hybrid implicit fields,

W. Zhang, T. Sun, S. Wang, Q. Cheng, and N. Haala, “Hi-slam: Monocular real-time dense mapping with hybrid implicit fields,”IEEE Robotics and Automation Letters, vol. 9, no. 2, pp. 1548–1555, 2023

2023
[24]

Rgb-only gaus- sian splatting slam for unbounded outdoor scenes,

S. Yu, C. Cheng, Y . Zhou, X. Yang, and H. Wang, “Rgb-only gaus- sian splatting slam for unbounded outdoor scenes,”arXiv preprint arXiv:2502.15633, 2025

work page arXiv 2025
[25]

Dust3r: Geometric 3d vision made easy,

S. Wang, V . Leroy, Y . Cabon, B. Chidlovskii, and J. Revaud, “Dust3r: Geometric 3d vision made easy,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 20 697–20 709

2024
[26]

Outdoor monocular slam with global scale-consistent 3d gaussian pointmaps,

C. Cheng, S. Yu, Z. Wang, Y . Zhou, and H. Wang, “Outdoor monocular slam with global scale-consistent 3d gaussian pointmaps,”arXiv preprint arXiv:2507.03737, 2025

work page arXiv 2025
[27]

Grounding image matching in 3d with mast3r,

V . Leroy, Y . Cabon, and J. Revaud, “Grounding image matching in 3d with mast3r,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 71–91

2024
[28]

Watersplatting: Fast underwater 3d scene reconstruction using gaussian splatting,

H. Li, W. Song, T. Xu, A. Elsig, and J. Kulhanek, “Watersplatting: Fast underwater 3d scene reconstruction using gaussian splatting,”arXiv preprint arXiv:2408.08206, 2024

work page arXiv 2024
[29]

Seathru-nerf: Neural radiance fields in scattering media,

D. Levy, A. Peleg, N. Pearl, D. Rosenbaum, D. Akkaynak, S. Korman, and T. Treibitz, “Seathru-nerf: Neural radiance fields in scattering media,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 56–65

2023
[30]

Mast3r-slam: Real- time dense slam with 3d reconstruction priors,

R. Murai, E. Dexheimer, and A. J. Davison, “Mast3r-slam: Real- time dense slam with 3d reconstruction priors,”arXiv preprint arXiv:2412.12392, 2024

work page arXiv 2024
[31]

Image segmentation using text and image prompts,

T. L ¨uddecke and A. Ecker, “Image segmentation using text and image prompts,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2022, pp. 7086–7096

2022
[32]

Mast3r-sfm: a fully-integrated solution for unconstrained structure-from-motion,

B. Duisterhof, L. Zust, P. Weinzaepfel, V . Leroy, Y . Cabon, and J. Revaud, “Mast3r-sfm: a fully-integrated solution for unconstrained structure-from-motion,”arXiv preprint arXiv:2409.19152, 2024

work page arXiv 2024
[33]

Go-slam: Global optimization for consistent 3d instant reconstruction,

Y . Zhang, F. Tosi, S. Mattoccia, and M. Poggi, “Go-slam: Global optimization for consistent 3d instant reconstruction,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2023, pp. 3727–3737

2023
[34]

Learning and aggregating deep local descriptors for instance-level recognition,

G. Tolias, T. Jenicek, and O. Chum, “Learning and aggregating deep local descriptors for instance-level recognition,” inEuropean Conference on Computer Vision. Springer, 2020, pp. 460–477

2020
[35]

Orb-slam3: An accurate open-source library for visual, visual– inertial, and multimap slam,

C. Campos, R. Elvira, J. J. G. Rodr ´ıguez, J. M. Montiel, and J. D. Tard´os, “Orb-slam3: An accurate open-source library for visual, visual– inertial, and multimap slam,”IEEE transactions on robotics, vol. 37, no. 6, pp. 1874–1890, 2021

2021