Pith. sign in

REVIEW 4 major objections 6 minor 35 references

MR-Voxel-SVIO shows that a multi-resolution voxelized LiDAR prior map can improve stereo visual-inertial odometry accuracy while transmitting only 44 bytes per queried feature (32 bytes up, 12 bytes down) between edge device and cloud.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 18:17 UTC pith:ETYWUQ2W

load-bearing objection Solid incremental extension of Voxel-SVIO with a genuinely new multi-resolution voxel mechanism and 3D-DDA association; the central accuracy claim holds up in the reported experiments, but the saliency selection needs a proper ablation before I'd fully trust it. the 4 major comments →

arxiv 2607.17332 v2 pith:ETYWUQ2W submitted 2026-07-19 cs.RO

Multi-Resolution Voxelized Map-Based Stereo Visual-Inertial Odometry

Shuyi Pan , Hangtian Wang , Zhaoxing Zhang , Chengliang Zhang , Zikang Yuan , Xin Yang This is my paper
classification cs.RO
keywords multi-resolution voxel mapstereo visual-inertial odometryLiDAR prior map3D-DDA ray castingbandwidth-efficient edge-cloud VIOcross-modal data associationEuRoC MAVKAIST dataset
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper claims that the reason prior-map VIO is impractical under bandwidth limits is that maps are too large to ship to an edge device—and that this can be fixed by compressing a LiDAR map into a multi-resolution voxel grid that keeps only one salient point per voxel. During online operation, the device sends just a ray origin, direction, distance, and a resolution index, and receives back a single 3D point. The choice of voxel resolution is tied to the physical footprint of each pixel at its estimated depth, so nearby features use fine voxels and distant features use coarse ones. Across the EuRoC and KAIST datasets, the system reports lower absolute trajectory error than prior Voxel-SVIO and other baselines, including when the prior map is queried with ground-truth poses. A sympathetic reader would see this as evidence that map-based VIO accuracy can be decoupled from map-transmission bandwidth.

Core claim

The central claim is that a prior LiDAR map, voxelized at multiple resolutions with a single representative point per voxel, can serve as a compact 3D prior for stereo visual-inertial odometry without ever loading the map onto the edge device. Each voxel stores the point with the highest normal-weighted intensity gradient (photometric saliency), so the map is sparse but visually distinctive. At run time, triangulated feature depths define a projection cone whose intercept picks the matching voxel resolution, and a 3D digital differential analyzer (3D-DDA) ray traversal returns the single stored point along the line of sight. The retrieved point is used twice in a two-stage MSCKF update: firs

What carries the argument

The key object is the multi-resolution voxelized prior map M = {M_r1, ..., M_rn}, where each voxel keeps exactly one LiDAR point selected by maximizing the photometric saliency score S_photo (Eq. 1). Resolution selection uses the depth-projected pixel footprint: r* = argmin_r |r - d/f|, matching voxel size to the real-world size of the pixel at its triangulated depth. The 3D-DDA ray traversal then walks the line of sight through the chosen resolution until it hits a valid voxel. Together these turn map association into a fixed-size query/response protocol: the edge sends ray parameters and receives one 3D point, which is what makes the bandwidth claim concrete.

Load-bearing premise

The load-bearing premise is that the photometric saliency score computed from LiDAR intensity gradients (Eq. 1) picks out the same structures that KLT will track in the camera image; if LiDAR texture and image texture part ways in a given environment, the retrieved 3D points will not correspond to the tracked 2D features and the state updates degrade.

What would settle it

Run the system in an environment where LiDAR intensity is nearly uniform on surfaces that have strong visual texture (for example, a room with printed posters on evenly painted walls) and measure the fraction of retrieved map points that reproject within a few pixels of their associated features; if the successful-association rate collapses while pure visual tracking still works, the saliency-selection assumption is falsified and the accuracy gain should disappear.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the reported accuracy holds, map-based VIO on bandwidth-limited edge devices can be built around on-demand point queries rather than downloading local maps, meshes, or images.
  • The two-stage state update—first without map-point refinement, then with registered map points—suggests that even noisy retrieved 3D points can constrain drift if the pose is already coarsely corrected.
  • Because each voxel retains at most one point, the prior map size grows with the number of occupied voxels, not the raw LiDAR density, making the approach scalable to the large outdoor KAIST sequences.
  • The multi-resolution selection reduces sensitivity to resolution choice: the ablation shows fixed single-resolution maps are consistently worse than the adaptive scheme, so the depth-dependent footprint is a real factor rather than an implementation detail.
  • With 44 bytes per feature and 5.1 ms latency at 200 features, the system stays inside frame time even at 1.66 MB/s bandwidth, supporting real-time cloud-assisted operation.

Where Pith is reading between the lines

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

  • A likely extension is to replace the one-time ground-truth alignment with a relocalization module; the paper names this as future work, but if relocalization can produce the same global pose, the bandwidth budget would be unchanged while removing the need for initial ground truth.
  • The photometric saliency choice ties map representativeness to KLT's preference for high gradients; a direct perceptual test—comparing retrieved points against triangulated stereo points on sequences with dense ground-truth depth—would quantify how much of the accuracy gain survives in low-texture or LiDAR-texture-mismatched scenes.
  • The same voxel/ray protocol could carry semantic or timestamp metadata instead of only geometric position, allowing the cloud-side map to serve richer priors (for example, dynamic-object masks) at the same bandwidth.
  • The reported per-frame transmission assumes one query per feature; batching or caching repeated queries across frames could push effective bandwidth lower, a straightforward engineering follow-up the paper does not analyze.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes MR-Voxel-SVIO, a stereo visual-inertial odometry system that uses a prior LiDAR map compressed into a multi-resolution voxel structure, where each voxel retains a single point chosen by a LiDAR-intensity-based photometric saliency score. During online operation, the edge device transmits compact ray parameters (32 bytes per feature) to a cloud server, which performs 3D-DDA ray traversal through the voxel map and returns a 3D map point (12 bytes per feature). The retrieved 2D-3D correspondences are used in a two-stage MSCKF-based estimator to refine the state. Experiments on six EuRoC and five KAIST sequences report lower ATE RMSE than several baselines, and ablations show that multi-resolution voxel selection outperforms fixed-resolution maps. The source code is released.

Significance. If the claims hold, the paper makes a useful contribution to edge-cloud collaborative VIO by demonstrating a compact prior-map representation that achieves low communication payload (44 bytes per feature) while retaining competitive pose accuracy. The release of source code and the use of two public datasets are strengths. However, the core cross-modal association assumption and several experimental rigor issues need to be addressed before the central claim can be considered fully supported.

major comments (4)
  1. [§IV-B, Eqs. (1)–(2)] The paper selects the single representative point per voxel using a LiDAR-intensity saliency score (Eq. 1) and stores the argmax (Eq. 2), with the rationale that this aligns the map with visual features trackable by KLT. This cross-modal correlation is not validated. If LiDAR intensity gradients do not predict image gradients, the retrieved 3D points will not correspond to the tracked 2D features, especially at outdoor voxel sizes of 0.15–0.20 m where the mismatch can be significant. Please provide an ablation replacing Eq. (1) with random or centroid-based selection, or a direct statistical comparison of LiDAR-selected points with visual feature locations.
  2. [§V-A, Tables II–III] The paper states that results are 'averaged over five runs' but reports no standard deviations or error bars, making it impossible to judge whether the improvements are statistically significant. Moreover, the MSCKF (w/map) baseline results are directly cited from its paper rather than obtained on the same hardware/data pipeline. Please report per-run statistics and either reproduce the baseline or clearly qualify the comparison.
  3. [§V-A, experimental setup] The method requires 'global localization poses as inputs,' and the paper uses ground-truth poses corrupted with translation noise (mean 2 cm, variance 0.25 cm) as the default input. This is a strong external assumption that is not part of the proposed system. The claimed accuracy may be contingent on this idealized input. Please analyze sensitivity to the corruption level (e.g., varying noise from 1 cm to 1 m) and discuss how such poses are obtained in practice without ground truth.
  4. [§IV-F, communication payload] The central claim of 'minimal data transmission' is supported only by an analytical computation of 44 bytes per feature and a total of 0.0084 MB per frame at 200 keypoints. No experimental comparison is made with the actual communication traffic of competing cloud-based systems (e.g., GMM-Loc [14], Bao et al. [1]) that also transmit data between edge and cloud. Please include a quantitative comparison table or a discussion of the payloads of the cited baselines to substantiate the claim.
minor comments (6)
  1. [§IV-F, first paragraph] The sentence 'Following the configuration in, the system tracks...' is incomplete; a reference or explanation is missing.
  2. [§V-C] The text refers to 'Section IV-G' but the communication payload analysis is in Section IV-F. Please correct the cross-reference.
  3. [§IV-B, Eq. (1)] The scaling parameter α is not defined or tuned; please specify its default value and any sensitivity to it.
  4. [Tables II–V and text] The method name is inconsistently typed as 'V oxel-SVIO' in several places; please fix the typo.
  5. [Fig. 2 caption] The caption contains garbled symbols (e.g., '??') and should be cleaned up.
  6. [§I and §IV-E] The claim that the two-stage estimation 'significantly prevents local minima' is not substantiated by analysis or experiments. Please either add support or soften the claim.

Circularity Check

0 steps flagged

No significant circularity: accuracy and bandwidth claims are empirical or arithmetic, not derived from fitted or self-referential inputs.

full rationale

The central accuracy results (Tables II–V and Fig. 4) are direct experiments on EuRoC MAV and KAIST against external baselines; ATE RMSE is measured, not predicted from the map construction. Offline map construction (Eqs. 1–2) selects one LiDAR point per voxel via a hand-specified intensity-gradient saliency score; no trajectory, ATE, or online estimate is used to define S_photo or p*, so there is no fitted-input-called-prediction loop. Online resolution selection (Eq. 19) chooses r* by the geometric pixel footprint d/f; it is not fitted to the reported errors, and the ablation in Tables IV–V is an empirical comparison of fixed versus adaptive resolutions. The 44-byte transmission figure is pure packet arithmetic (32 B uplink ray parameters + 12 B downlink 3D position), i.e., an accounting of the proposed protocol rather than a derived prediction. The main self-citation is [19] Voxel-SVIO for the two-stage estimator; that system is published, externally benchmarked, and is itself the baseline compared in Table II, while the MSCKF machinery is independently grounded in OpenVINS [18] and the filter equations (11)–(18) are standard. No uniqueness theorem or ansatz is imported only from the authors. Two non-circular weaknesses are noted for completeness: the cross-modal saliency assumption in Eq. (1) (LiDAR intensity gradient implying KLT-trackable image texture) is asserted without statistical validation or a random/centroid selection ablation, and Sec. IV-F contains an incomplete reference ('Following the configuration in,'). These are validation and completeness risks, not circular reductions, because neither the saliency score nor the resolution rule takes the reported pose error as input.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 0 invented entities

The paper introduces no new physical entities; its contribution is a data structure and indexing strategy. However, the central claim relies on several hand-set hyperparameters (resolution sets, per-voxel point caps, feature counts, noise covariances) and two untested domain assumptions (LiDAR saliency matching visual features, and depth-based resolution selection).

free parameters (6)
  • Voxel resolution sets = Indoor: {0.01, 0.02, 0.03} m; Outdoor: {0.05, 0.10, 0.15, 0.20} m
    Hand-chosen per scenario to align with typical pixel footprints; not derived and not optimized with a held-out set.
  • Saliency scaling parameter alpha = unspecified
    Appears in Eq. 1 to prioritize structural edges; no value is reported, so it is effectively a free tuning parameter.
  • Maximum tracked keypoints = 200 per frame
    Used for the bandwidth calculation in Sec. IV-F; the sentence referencing its configuration is missing a citation.
  • Registered-map voxel grid size and capacity = 0.3 m indoor / 1.0 m outdoor, 5 points per voxel
    Maintenance voxel parameters chosen by hand in Sec. IV-E.3; no sensitivity analysis provided.
  • Number of co-visible features in first-stage update = up to 100
    Set in Sec. IV-E.1 to bound computation; no ablation on this number.
  • Map-point measurement noise covariance R_m,k = unspecified diagonal eigenvalue
    Described as 'a certain eigenvalue' in Sec. IV-E.2; without a value, tuning is required to reproduce results.
axioms (6)
  • standard math 3D-DDA algorithm correctly traverses multi-resolution voxel grids
    All 2D-3D associations rely on this classic ray traversal; correctness is cited to Amanatides & Woo [16] and not re-derived.
  • domain assumption LiDAR intensity-gradient saliency correlates with KLT-trackable visual features
    Eqs. 1–2 select the single representative point per voxel based on LiDAR intensity; this cross-modal alignment is asserted but not validated with analysis.
  • domain assumption Initial ground-truth pose is available for one-time global alignment and is not used later
    Sec. IV-D states the local frame is aligned using the initial GT pose; if such a pose is unavailable, all map queries operate in the wrong frame.
  • ad hoc to paper Eq. 19 correctly maps the pixel footprint to the optimal voxel resolution
    The formula r* = argmin |r - d/f| is a heuristic geometric similarity rule; no derivation or error analysis is given.
  • ad hoc to paper Two-stage state estimation prevents local minima
    The paper asserts this benefit but provides no comparison against a single-stage update or any local-minima analysis.
  • ad hoc to paper Corrupted ground-truth poses (mean 2 cm) are a valid proxy for global localization inputs
    The fair-comparison protocol follows Bao et al. [1]; no sensitivity to noise magnitude is tested, so the proxy is an unverified assumption.

pith-pipeline@v1.3.0-alltime-deepseek · 12944 in / 11939 out tokens · 117011 ms · 2026-08-01T18:17:11.309565+00:00 · methodology

0 comments
read the original abstract

Incorporating prior maps significantly enhances the accuracy and robustness of pose estimation in visual-inertial odometry (VIO). However, the large data volume of such maps, combined with limited transmission bandwidth, makes it impractical to continuously load local maps onto an edge device. In this paper, we propose a multi-resolution prior map construction method and a corresponding map-based VIO system. The prior map is voxelized at multiple resolutions, with each voxel retaining only a single map point. During online VIO operation, a cone-shaped indexing strategy associates 2D features on the edge device with 3D map points. The cone's intercept is determined by the distance from the current position to the 3D points, enabling the selection of the appropriate resolution level and the retrieval of the unique map point within the corresponding voxel via a 3D digital differential analyzer (DDA) algorithm. This approach minimizes both the volume of data required for transmission and the computational load during data association. Extensive experiments on two public datasets demonstrate that our system achieves accurate pose estimation while requiring minimal data transmission.

Figures

Figures reproduced from arXiv: 2607.17332 by Chengliang Zhang, Hangtian Wang, Shuyi Pan, Xin Yang, Zhaoxing Zhang, Zikang Yuan.

Figure 1
Figure 1. Figure 1: Illustration of multi-resolution voxelized prior map [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the proposed multi-resolution voxel indexing and data association. (a) Preconstructed multi-resolution [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of our MR-Voxel-SVIO which consists of four modules: an offline processing engine, an image processing [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: (a)–(d) are the comparison results between our [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

35 extracted references · 5 linked inside Pith

  1. [1]

    Robust tightly-coupled visual-inertial odometry with pre-built maps in high latency situations,

    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

    2022

  2. [2]

    Localization of mobile robot in prior 3d LiDAR maps using stereo image sequence,

    I. V . Belkin, A. A. Abramenko, V . D. Bezuglyi, and D. A. Yudin, “Localization of mobile robot in prior 3d LiDAR maps using stereo image sequence,”Computer Optics, vol. 48, no. 3, pp. 406–417, 2024

    2024

  3. [3]

    Visual- inertial localization with prior lidar map constraints,

    X. Zuo, P. Geneva, Y . Yang, W. Ye, Y . Liu, and G. Huang, “Visual- inertial localization with prior lidar map constraints,”IEEE Robotics and Automation Letters, vol. 4, no. 4, pp. 3394–3401, 2019

    2019

  4. [4]

    Monocular direct sparse localization in a prior 3d surfel map,

    H. Ye, H. Huang, and M. Liu, “Monocular direct sparse localization in a prior 3d surfel map,” in2020 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2020, pp. 8892–8898

    2020

  5. [5]

    Sharing heterogeneous spatial knowledge: Map fusion between asynchronous monocular vision and lidar or other prior inputs,

    Y . Lu, J. Lee, S.-H. Yeh, H.-M. Cheng, B. Chen, and D. Song, “Sharing heterogeneous spatial knowledge: Map fusion between asynchronous monocular vision and lidar or other prior inputs,” inRobotics Re- search: The 18th International Symposium ISRR. Springer, 2019, pp. 727–741

    2019

  6. [6]

    ecar: Edge-assisted collaborative augmented reality framework,

    J. Jeon and W. Woo, “ecar: Edge-assisted collaborative augmented reality framework,”arXiv preprint arXiv:2405.06872, 2024

    Pith/arXiv arXiv 2024

  7. [7]

    Sr-livo: Lidar- inertial-visual odometry and mapping with sweep reconstruction,

    Z. Yuan, J. Deng, R. Ming, F. Lang, and X. Yang, “Sr-livo: Lidar- inertial-visual odometry and mapping with sweep reconstruction,” IEEE Robotics and Automation Letters, vol. 9, no. 6, pp. 5110–5117, 2024

    2024

  8. [8]

    Gemdepth: Geometry-embedded features for 3d-consistent video depth,

    Y . Liu, J. Cheng, L. Liu, W. Liao, H. Cheng, Y . Wang, and X. Yang, “Gemdepth: Geometry-embedded features for 3d-consistent video depth,”arXiv preprint arXiv:2605.10525, 2026

    Pith/arXiv arXiv 2026

  9. [9]

    Monster: Marry monodepth to stereo unleashes power,

    J. Cheng, L. Liu, G. Xu, X. Wang, Z. Zhang, Y . Deng, J. Zang, Y . Chen, Z. Cai, and X. Yang, “Monster: Marry monodepth to stereo unleashes power,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 6273–6282

    2025

  10. [10]

    D 2 vo: Monocular deep direct visual odometry,

    Q. Jia, Y . Pu, J. Chen, J. Cheng, C. Liao, and X. Yang, “D 2 vo: Monocular deep direct visual odometry,” in2020 IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems (IROS). IEEE, 2020, pp. 10 158–10 165

    2020

  11. [11]

    Monster++: Unified stereo matching, multi- view stereo, and real-time stereo with monodepth priors,

    J. Cheng, W. Liao, Z. Cai, L. Liu, G. Xu, X. Wang, Y . Wang, Z. Yuan, Y . Deng, J. Zang,et al., “Monster++: Unified stereo matching, multi- view stereo, and real-time stereo with monodepth priors,”arXiv preprint arXiv:2501.08643, 2025

    arXiv 2025

  12. [12]

    Decoupling bidirectional geometric representations of 4d cost volume with 2d convolution,

    X. Wei, C. Shu, Z. Yue, C. Huang, W. Liu, S. Yang, L. Yang, P. Gao, W. Zhang, G. Zhu,et al., “Decoupling bidirectional geometric representations of 4d cost volume with 2d convolution,”arXiv preprint arXiv:2509.02415, 2025

    Pith/arXiv arXiv 2025

  13. [13]

    A wavelet- based stereo matching framework for solving frequency convergence inconsistency,

    X. Wei, J. Liu, D. Yang, J. Cheng, C. Shu, and W. Wang, “A wavelet- based stereo matching framework for solving frequency convergence inconsistency,”arXiv preprint arXiv:2505.18024, 2025

    Pith/arXiv arXiv 2025

  14. [14]

    Geometric structure aided visual inertial localization,

    H. Huang, H. Ye, J. Jiao, Y . Sun, and M. Liu, “Geometric structure aided visual inertial localization,”arXiv preprint arXiv:2011.04173, 2020

    Pith/arXiv arXiv 2011

  15. [15]

    Gmmloc: Structure consistent visual localization with gaussian mixture models,

    H. Huang, H. Ye, Y . Sun, and M. Liu, “Gmmloc: Structure consistent visual localization with gaussian mixture models,”IEEE Robotics and Automation Letters, vol. 5, no. 4, pp. 5043–5050, 2020

    2020

  16. [16]

    A fast voxel traversal algorithm for ray tracing

    J. Amanatides, A. Woo,et al., “A fast voxel traversal algorithm for ray tracing.” inEurographics, vol. 87, no. 3, 1987, pp. 3–10

    1987

  17. [17]

    Leveraging consistent spatio-temporal correspondence for robust vi- sual odometry,

    Z. Zhang, J. Cheng, G. Xu, X. Wang, C. Zhang, and X. Yang, “Leveraging consistent spatio-temporal correspondence for robust vi- sual odometry,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 10, 2025, pp. 10 367–10 375

    2025

  18. [18]

    Openvins: A research platform for visual-inertial estimation,

    P. Geneva, K. Eckenhoff, W. Lee, Y . Yang, and G. Huang, “Openvins: A research platform for visual-inertial estimation,” in2020 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2020, pp. 4666–4672

    2020

  19. [19]

    V oxel-svio: Stereo visual-inertial odometry based on voxel map,

    Z. Yuan, F. Lang, J. Deng, H. Luo, and X. Yang, “V oxel-svio: Stereo visual-inertial odometry based on voxel map,”IEEE Robotics and Automation Letters, 2025

    2025

  20. [20]

    Coatrsnet: Fully exploiting convolution and attention for stereo matching by region separation,

    J. Cheng, G. Xu, P. Guo, and X. Yang, “Coatrsnet: Fully exploiting convolution and attention for stereo matching by region separation,” International Journal of Computer Vision, vol. 132, no. 1, pp. 56–73, 2024

    2024

  21. [21]

    Region separable stereo matching,

    J. Cheng, X. Yang, Y . Pu, and P. Guo, “Region separable stereo matching,”IEEE Transactions on Multimedia, vol. 25, pp. 4880–4893, 2022

    2022

  22. [22]

    Relocalization, global optimization and map merging for monocular visual-inertial slam,

    T. Qin, P. Li, and S. Shen, “Relocalization, global optimization and map merging for monocular visual-inertial slam,” in2018 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp. 1197–1204

    2018

  23. [23]

    Global- map-registered local visual odometry using on-the-fly pose graph updates,

    M. Yamaguchi, S. Mori, H. Saito, S. Yachida, and T. Shibata, “Global- map-registered local visual odometry using on-the-fly pose graph updates,” inInternational Conference on Augmented Reality, Virtual Reality and Computer Graphics. Springer, 2020, pp. 299–311

    2020

  24. [24]

    Collaborative augmented reality on smartphones via life-long city-scale maps,

    L. Platinsky, M. Szabados, F. Hlasek, R. Hemsley, L. Del Pero, A. Pancik, B. Baum, H. Grimmett, and P. Ondruska, “Collaborative augmented reality on smartphones via life-long city-scale maps,” in 2020 IEEE International Symposium on Mixed and Augmented Reality (ISMAR). IEEE, 2020, pp. 533–541

    2020

  25. [25]

    Monocular camera localization in 3d lidar maps,

    T. Caselitz, B. Steder, M. Ruhnke, and W. Burgard, “Monocular camera localization in 3d lidar maps,” in2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2016, pp. 1926–1931

    2016

  26. [26]

    Get out of my lab: Large-scale, real-time visual-inertial localization

    S. Lynen, T. Sattler, M. Bosse, J. A. Hesch, M. Pollefeys, and R. Siegwart, “Get out of my lab: Large-scale, real-time visual-inertial localization.” inRobotics: Science and Systems, vol. 1, no. 10.15607, 2015

    2015

  27. [27]

    Laser map aided visual inertial localization in changing environment,

    X. Ding, Y . Wang, D. Li, L. Tang, H. Yin, and R. Xiong, “Laser map aided visual inertial localization in changing environment,” in2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2018, pp. 4794–4801

    2018

  28. [28]

    Stereo camera localization in 3d lidar maps,

    Y . Kim, J. Jeong, and A. Kim, “Stereo camera localization in 3d lidar maps,” in2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2018, pp. 1–9

    2018

  29. [29]

    Adaptive fusion of single-view and multi-view depth for autonomous driving,

    J. Cheng, W. Yin, K. Wang, X. Chen, S. Wang, and X. Yang, “Adaptive fusion of single-view and multi-view depth for autonomous driving,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 10 138–10 147

    2024

  30. [30]

    An iterative image registration technique with an application to stereo vision,

    B. D. Lucas and T. Kanade, “An iterative image registration technique with an application to stereo vision,” inIJCAI’81: 7th international joint conference on Artificial intelligence, vol. 2, 1981, pp. 674–679

    1981

  31. [31]

    Direct sparse odometry,

    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

  32. [32]

    Rnin- vio: Robust neural inertial navigation aided visual-inertial odometry in challenging scenes

    D. Chen, N. Wang, R. Xu, W. Xie, H. Bao, and G. Zhang, “Rnin- vio: Robust neural inertial navigation aided visual-inertial odometry in challenging scenes.” inISMAR, 2021, pp. 275–283

    2021

  33. [33]

    Sr-lio: Lidar-inertial odometry with sweep reconstruction,

    Z. Yuan, F. Lang, T. Xu, and X. Yang, “Sr-lio: Lidar-inertial odometry with sweep reconstruction,” in2024 IEEE/RSJ International Confer- ence on Intelligent Robots and Systems (IROS). IEEE, 2024, pp. 7862–7869

    2024

  34. [34]

    Vins-mono: A robust and versatile monoc- ular visual-inertial state estimator,

    T. Qin, P. Li, and S. Shen, “Vins-mono: A robust and versatile monoc- ular visual-inertial state estimator,”IEEE transactions on robotics, vol. 34, no. 4, pp. 1004–1020, 2018

    2018

  35. [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