Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

LoD-Loc v2: Aerial Visual Localization over Low Level-of-Detail City Models using Explicit Silhouette Alignment

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read A drone can localize against low-detail LoD1 city models by aligning predicted and rendered building silhouettes.

desk verdict LoD-Loc v2 is a genuine first demonstration of LoD1-based aerial localization with useful datasets, but the headline overstates the texture-model comparison and the sensitivity to LoD1 model error is untested. read the letter →

arxiv 2507.00659 v1 pith:WIWW7PUO submitted 2025-07-01 cs.CV

classification cs.CV
keywords visuallocalizationaerialUAVLoD1citymodelssilhouettealignmentparticlefilterbuildingsegmentationposeestimationrender-and-compare
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

LoD-Loc v2 claims that aerial visual localization can work with low level-of-detail (LoD1) city models: the block-shaped 3D building footprints that are widely available and that many countries are already producing nationally. The paper's central idea is that even when a LoD1 model has no texture or detailed structure, its projected building silhouettes still overlap well with the building silhouettes segmented from a query image when the pose is correct. On that basis it builds a coarse-to-fine pose estimator: a 4-degree-of-freedom cost volume scored by silhouette intersection-over-union, followed by particle-filter refinement. The paper reports median translation errors of 0.72–0.71 m on its UA VD4L-LoDv2 dataset and 1.83–2.64 m on Swiss-EPFLv2, and says this is the first localization method to work on LoD1 models. Why this matters: if true, a drone's global localization no longer depends on scarce, expensive textured 3D maps.

What carries the argument

The central object is the explicit silhouette alignment cost of Eq. (2), c(Iq,Ihyp)=|Mq∩Mhyp|/|Mq∪Mhyp|, a pixel-wise intersection-over-union between the building mask predicted from the query image and the building mask produced by projecting the LoD1 model under a candidate pose. It carries the argument because it converts the apparently impossible problem of matching structureless LoD1 geometry into a tractable mask-overlap problem; the coarse 4-DoF pose cost volume built from this cost, and the particle-filter refinement with multi-beam tracking, are the mechanisms that make it operational.

What would settle it

Take a block where the LoD1 model has been built by extruding footprints at uniform roof heights while the real buildings have varied heights, render the model at the true pose, and run the method; if the argmax of the Eq. (2) cost volume or the refined pose moves by more than the method's reported error, the silhouette-alignment principle fails where LoD1 generalization matters.

Watch

Extended reading notes

Core claim

The paper tries to establish that explicit silhouette alignment is a sufficient signal for pose recovery over low-LoD city models. Given a query image and a prior pose, the method first extracts a building mask Mq with a SAM2-based segmentation network. It then renders the LoD1 model from uniformly sampled hypotheses in (x, y, z, yaw) around the prior, binarizes the renderings, and scores each hypothesis by the IoU cost c(Iq,Ihyp)=|Mq∩Mhyp|/|Mq∪Mhyp|. The pose with maximum cost is the coarse estimate; a particle filter that perturbs particles on SO(3)×T(3) with multi-beam tracking refines it. In the paper's own evaluation, this pipeline outperforms the previous LoD-Loc wireframe alignment by large margins, exceeds texture-model baselines such as RoMa on the LoD1 datasets, and holds up under prior translation errors up to 200 m.

Load-bearing premise

The load-bearing premise is that the LoD1 model's geometry—building footprints and heights—is accurate enough that the best silhouette overlap occurs at the true camera pose, and the paper does not stress-test this against systematically simplified or outdated LoD1 models.

Editorial extensions

If this is right

  • Localization becomes possible on the LoD1 models that many countries are already building nationwide, so drone pose estimation no longer requires textured mesh reconstructions.
  • The method also improves accuracy on high-LoD (LoD2/LoD3) models, so the same pipeline serves multiple LoD levels.
  • The convergence basin widens to prior translation errors around 200 m, so the system can operate with degraded GPS or inertial priors.
  • Since LoD1 maps are lightweight and contain no photorealistic detail, storing and sharing them raises fewer privacy concerns than textured 3D maps.

Reading between the lines

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

  • A consequence the authors leave implicit is that the silhouette cost is a general scoring function that could be reused with any segmentation backbone, so runtime and accuracy can be traded directly.
  • The paper's stated limitation—alignment degrades when building silhouettes fill the frame—implies that low-altitude or close-range flights where one building dominates the view should break the method, because the cost volume can no longer separate translation along the building face.
  • The two released LoD1 models were generated from the same sources as the ground-truth geometry, so the harder condition of an independently surveyed LoD1 map with generalized heights or merged footprints remains untested; an evaluation on such a map would stress the central assumption.
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

3 major / 6 minor

Summary. The paper proposes LoD-Loc v2, a coarse-to-fine aerial visual localization method that aligns predicted building silhouettes with projected low Level-of-Detail (LoD) city models. A SAM2-based segmentation network extracts building masks from a query image; a 4-DOF pose cost volume is constructed by evaluating the IoU between projected LoD1 silhouettes and the predicted mask (Eq. 2), and a particle filter refines the selected pose. The authors release two LoD1 datasets (UA VD4L-LoDv2 and Swiss-EPFLv2) and report large improvements over the previous LoD-Loc method as well as over texture-model-based baselines, together with an enlarged convergence basin for larger prior errors.

Significance. If the reported results hold, the paper makes a strong case that low-LoD city models, which are widely available and lightweight, can support useful aerial visual localization. The method's simplicity is appealing, and the release of two geo-referenced LoD1 datasets with real query images is a valuable contribution. The paper also demonstrates cross-city generalization (trained on UA VD4L, tested on Swiss-EPFL) and robust performance under large prior errors. However, the evaluation's reliance on a LoD1 model derived from the same mesh as the ground truth raises questions about how the method will behave on operational LoD1 data with unknown geometric errors.

major comments (3)
  1. [Sec. 4 (Datasets), Eq. (2)] The UA VD4L-LoDv2 LoD1 model is 'automatically generated from the textured mesh model of the UA VD4L by the DP modeler,' which is the same mesh that provides the ground-truth poses. As the cost in Eq. (2) directly compares predicted masks to projections of this LoD1 model, the evaluation is partially circular: any systematic simplification in the LoD1 geometry is shared by the training labels and the alignment objective, so the favorable results do not establish robustness to the geometric errors typical of operational LoD1 models (e.g., generalized roof heights, merged footprints). Please report quantitative LoD1-model error metrics (height RMSE, footprint IoU vs. the reference mesh) and add a sensitivity analysis that perturbs building heights and footprints and measures the resulting pose accuracy. Otherwise, the 'first time' claim is only demonstrated for datasets whose LoD1 models are consistent with the ground truth by construction.
  2. [Abstract, Conclusion, Sec. 4.2] The paper repeatedly claims that LoD-Loc v2 'even surpass[es] texture-model-based methods,' but Sec. 4.2 explicitly states 'the comparison with texture model-based baselines is quite unfair.' A claim presented as a headline finding is thus acknowledged in the body to rest on an unfair experimental setup. Please either remove this claim or replace it with a fair comparison (e.g., reporting both methods under the same LoD1 reference or clearly labeling the texture-model results as an upper-bound reference).
  3. [Eq. (1), Sec. 4.1] The coarse pose sampling ranges r(d) and counts n(d) are never specified, despite being essential for reproducing the method and for interpreting the convergence-basin experiments in Tab. 4. State the numerical values used in all experiments, or give a table in the appendix.
minor comments (6)
  1. [General] Several places reference 'the Appendix' (Sec. 2, Sec. 3.1, Sec. 4.1, Sec. 5), but the submitted manuscript does not include an appendix; ensure the appendix is provided.
  2. [Eq. (1)] The notation {r(d)} and {n(d)} is ambiguous: clarify whether these are per-axis ranges or vectors.
  3. [Table 3] The footnote 'same*' is confusing: state explicitly which rows are identical to Table 2.
  4. [Table 6] Table 6 reports IoU for segmentation modules, but the metric definitions are not given; specify how the IoU is computed (e.g., on the test set).
  5. [Sec. 4.1] The paper says the testing dataset 'differs in both area and viewpoints from the training sets' for the Swiss-EPFL images; specify whether Swiss-EPFL was used for validation or hyperparameter tuning.
  6. [Sec. 5 (Limitation)] The limitation statement is vague ('limitations when building silhouettes predominantly occupying the query image'); provide a quantitative characterization, e.g., a threshold on mask area fraction.

Circularity Check

1 steps flagged · score 2.0 of 10

Minor self-referential segmentation/reference coupling; central pose claim still rests on independent Swiss-EPFL and high-LoD evidence.

  1. other [Sec. 3.2 (Eq. 2), Sec. 3.4 (Eq. 4), Sec. 4.1]
    "The ground truth mask labels G, are generated by projecting the LoD model [97] onto a 2D plane with paired poses. ... c(Iq, Ihyp) = |Mq ∩ Mhyp| / |Mq ∪ Mhyp|"

    The two masks compared in the pose cost are not independent: Mhyp is a projection of the LoD model, and Mq is produced by a segmentation network whose training labels G are projections of the same LoD model at paired poses. In effect, the network is taught to output the LoD model's own silhouettes, and the pose cost then aligns that output with the same LoD model's silhouettes. This makes the coarse pose selection partly a self-consistency check of the LoD geometry rather than a comparison against an independent image-derived measurement.

full rationale

The central pose estimate is obtained by maximizing a fixed geometric objective (Eq. 2) over a cost volume and particle-filter refinement; no pose parameter is fit to test annotations and then reported as a prediction. Real query images and independent pose annotations are used for evaluation, and the Swiss-EPFLv2 experiment provides out-of-domain evidence with a semi-automated LoD1 model in a region not used for training. The main self-referential element is that segmentation supervision (Eq. 4, Sec. 4.1) uses LoD-model projections, so the segmentation output and the pose-scoring render share the same geometric source. This is a benchmark-favorability concern for the UA VD4L-LoDv2 dataset rather than a derivation-level circularity: the method could in principle fail on a genuinely inaccurate LoD1 model, but that is an unquantified robustness limitation, not an equation-level reduction of the output to the input. Self-citations to [97] provide datasets and the LoD-Loc baseline, but the central claim does not depend on an unverified self-citation chain or an imported uniqueness theorem.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

The central claim rests mainly on engineering assumptions: an accurate prior, negligible pitch and roll error, LoD1 model fidelity, and a segmentation network that transfers from synthetic UA VD4L training to unseen cities. No new physical or mathematical entities are introduced; the many hand-set hyperparameters, sampling interval, particle filter settings, and thresholds, are the effective free parameters of the system.

free parameters (8)
  • Coarse pose sampling interval = 10 m
    Grid stride in (x,y,z,yaw) around the prior pose; hand-set in Sec 4.1 and controls coarse pose resolution and cost volume size.
  • Particle filter translation perturbation sigma = 1.5
    Gaussian noise scale for perturbing candidate poses during refinement (Sec 4.1).
  • Particle filter attenuation coefficient gamma = 0.3
    Controls the decay of perturbations across iterations (Sec 4.1).
  • Particle filter iteration count N = 40
    Number of refinement iterations (Sec 4.1).
  • Particle filter beams = 2
    Number of independent optimization threads in the multi-beam tracking approach (Sec 4.1).
  • Segmentation threshold delta_s = 0.5
    Binarization threshold for predicted building masks (Sec 4.1).
  • Rendered mask threshold delta_I = 127
    Binarization threshold for rendered silhouettes (Sec 4.1).
  • Particle candidates per iteration n = 52
    Number of candidate poses generated in each particle filter iteration (Sec 4.1).
assumptions (5)
  • domain assumption Pitch and roll errors from onboard inertial sensors are negligible, so the rotation search is restricted to yaw only.
    Invoked in Sec 3.2 to fix the rotation axis (rho, r) and in Sec 3.3 to keep pitch and roll fixed; if IMU gravity direction is disturbed, the pose search misses roll and pitch degrees of freedom.
  • domain assumption The LoD1 reference model geometry is sufficiently accurate for silhouette alignment.
    The method compares query-derived masks to LoD1 projections; the paper provides no error model for LoD1 geometry and the datasets are generated from or alongside the same source models.
  • domain assumption The segmentation network's training labels, generated by projecting the LoD model (Sec 4.1), transfer to real images and unseen cities.
    Training uses only synthetic UA VD4L renders; cross-city evidence is limited to the Swiss-EPFL dataset, where out-of-place accuracy is noticeably lower.
  • domain assumption A prior pose within the coarse sampling range is available.
    The coarse stage samples uniformly around the prior pose; the paper tests prior errors up to 200 m but does not address global localization without a prior.
  • domain assumption Ground-truth poses from the UA VD4L-LoD and Swiss-EPFL datasets are accurate and in the same coordinate frame as the LoD models.
    All evaluation metrics depend on these annotations, and the paper does not independently verify their accuracy.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LoD-Loc v2: Aerial Visual Localization over Low Level-of-Detail City Models using Explicit Silhouette Alignment." pith.science (2026). https://pith.science/paper/WIWW7PUO

@misc{pith2026250700659,
  author       = {Pith},
  title        = {Pith review of: LoD-Loc v2: Aerial Visual Localization over Low Level-of-Detail City Models using Explicit Silhouette Alignment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WIWW7PUO}},
  note         = {Machine review of arXiv:2507.00659}
}
read the original abstract

We propose a novel method for aerial visual localization over low Level-of-Detail (LoD) city models. Previous wireframe-alignment-based method LoD-Loc has shown promising localization results leveraging LoD models. However, LoD-Loc mainly relies on high-LoD (LoD3 or LoD2) city models, but the majority of available models and those many countries plan to construct nationwide are low-LoD (LoD1). Consequently, enabling localization on low-LoD city models could unlock drones' potential for global urban localization. To address these issues, we introduce LoD-Loc v2, which employs a coarse-to-fine strategy using explicit silhouette alignment to achieve accurate localization over low-LoD city models in the air. Specifically, given a query image, LoD-Loc v2 first applies a building segmentation network to shape building silhouettes. Then, in the coarse pose selection stage, we construct a pose cost volume by uniformly sampling pose hypotheses around a prior pose to represent the pose probability distribution. Each cost of the volume measures the degree of alignment between the projected and predicted silhouettes. We select the pose with maximum value as the coarse pose. In the fine pose estimation stage, a particle filtering method incorporating a multi-beam tracking approach is used to efficiently explore the hypothesis space and obtain the final pose estimation. To further facilitate research in this field, we release two datasets with LoD1 city models covering 10.7 km , along with real RGB queries and ground-truth pose annotations. Experimental results show that LoD-Loc v2 improves estimation accuracy with high-LoD models and enables localization with low-LoD models for the first time. Moreover, it outperforms state-of-the-art baselines by large margins, even surpassing texture-model-based methods, and broadens the convergence basin to accommodate larger prior errors.

Figures

Figures reproduced from arXiv: 2507.00659 by the authors.

Figure 1
Figure 1. In this paper, we introduce LoD-Loc v2 to tackle aerial visual localization using [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Comparison between LoD3 and LoD1. The LoD3 model provides a highly accurate representation of the building’s geometry. The LoD1 model focuses on the building’s basic shape without intricate details. Consequently, achieving effective visual localization using low-LoD city models poses a significant chal￾lenge. More details can be found in the Appendix. this work, we explore a straightforward yet effective idea: even … view at source ↗
Figure 3
Figure 3. Overview of LoD-Loc v2. 1. LoD-Loc v2 employs a building segmentation module to extract building silhouettes Mq from the query image Iq (Sec. 3.1). 2. A 4D pose cost volume C is built for pose hypotheses {ξhyp} sampled around the prior pose ξp to select the pose ξc with the highest probability, based on the alignment between projected and predicted building silhouettes (Sec. 3.2). 3. A particle filter refinement is … view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Visualization of segmentation results and overlay of different iterations. The visualized segmentation results demonstrate the model’s excellent segmentation performance. Better alignment indicates a more accurate pose prediction. Besides, the top two rows are from the…
Figure 5
Figure 5. Figure 5: Convergence basin concerning translation and orien￾tation. The average alignment cost is computed on the UAVD4L￾LoDv2 dataset using Eq. (2). a build-in prior pose data, LoD-Loc v2 employs a coarse￾to-fine pipeline to recover the camera pose, consisting of building silh…

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. Full citation record

  1. Camera Pose Refinement via 3D Gaussian Splatting

    cs.CV 2025-08 unverdicted novelty 6.0 of 10

    GS-SMC refines camera poses without retraining by enforcing epipolar constraints between a query photo and several photos rendered from an existing 3D Gaussian Splatting model.

Reference graph

Works this paper leans on

98 extracted references · 66 canonical work pages · cited by 1 Pith paper

  1. [1]

    https://3d.bk.tudelft.nl/projects/ 3dbag/

    3d bag. https://3d.bk.tudelft.nl/projects/ 3dbag/. 2

  2. [2]

    https : / / www

    City json. https : / / www . cityjson . org / datasets/. 2

  3. [3]

    https : / / www

    Dp modeler. https : / / www . whulabs . com / DPModeler/index.aspx. 6

  4. [4]

    https : / / www

    Das 3d-modell. https : / / www . berlin . de / sen/sbw/stadtdaten/stadtwissen/digitale- innenstadt/3d-modell/. 2

  5. [5]

    https://www.bousai.go.jp/ kohou/kouhoubousai/r03/102/news_05.html

    The development and utilization of 3d city model and the introduction of disaster prevention measures in the project of data opening. https://www.bousai.go.jp/ kohou/kouhoubousai/r03/102/news_05.html . 2

  6. [6]

    https://geospatialworld.net/prime/case- study / aec / 3d - evolution - of - the - dutch - city-of-rotterdam-the-netherlands/

    3d-evolution of the dutch city of rotterdam, the netherlands. https://geospatialworld.net/prime/case- study / aec / 3d - evolution - of - the - dutch - city-of-rotterdam-the-netherlands/ . 2

  7. [7]

    http://www.openscenegraph

    Openscenegraph. http://www.openscenegraph. com/. 4, 7

  8. [8]

    https://www.smartnation.gov

    Opening address by minister vivian balakrishnan at geo con- nect asia 2021, . https://www.smartnation.gov. sg/media- hub/speeches/geo-connect-asia- 2021/. 2

Show all 98 references
  1. [9]

    https : / / www

    Integrated environmental modeller, . https : / / www . smartnation . gov . sg / initiatives / integrated-environmental-modeller/. 2

  2. [10]

    https : //www.swisstopo.admin.ch/en/vision- and- strategic-fields-of-action-2025

    Vision and strategic fields of action 2025. https : //www.swisstopo.admin.ch/en/vision- and- strategic-fields-of-action-2025 . 2

  3. [11]

    https://www.gov.cn/ xinwen/2022-03/01/content_5676226.htm

    Ministry of natural resources fully advances the construc- tion of real-scene 3d china. https://www.gov.cn/ xinwen/2022-03/01/content_5676226.htm. 2

  4. [12]

    https : / / aws

    Open city model. https : / / aws . amazon . com / marketplace / pp / prodview - q5gmfev7upw7e # resources. 2

  5. [13]

    https : / / www

    swissbuildings3d. https : / / www . swisstopo . admin . ch / en / landscape - model - swissbuildings3d-2-0. 2

  6. [14]

    All about vlad

    Relja Arandjelovic and Andrew Zisserman. All about vlad. In Proceedings of the IEEE conference on Computer Vision and Pattern Recognition, pages 1578–1585, 2013. 3

  7. [15]

    Netvlad: Cnn architecture for weakly supervised place recognition

    Relja Arandjelovic, Petr Gronat, Akihiko Torii, Tomas Pa- jdla, and Josef Sivic. Netvlad: Cnn architecture for weakly supervised place recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 5297–5307, 2016. 3

  8. [16]

    Magsac: marginalizing sample consensus

    Daniel Barath, Jiri Matas, and Jana Noskova. Magsac: marginalizing sample consensus. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10197–10205, 2019. 3

  9. [17]

    Giovanni Barbarani, Mohamad Mostafa, Hajali Bayramov, Gabriele Trivigno, Gabriele Berton, Carlo Masone, and Bar- bara Caputo. Are local features all you need for cross- domain visual place recognition? In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Re...

  10. [18]

    Review of target geo-location al- gorithms for aerial remote sensing cameras without control points

    Yiming Cai, Yao Zhou, Hongwen Zhang, Yuli Xia, Peng Qiao, and Junsuo Zhao. Review of target geo-location al- gorithms for aerial remote sensing cameras without control points. Applied Sciences, 12(24):12689, 2022. 1

  11. [19]

    Rsprompter: Learning to prompt for remote sensing instance segmenta- tion based on visual foundation model

    Keyan Chen, Chenyang Liu, Hao Chen, Haotian Zhang, Wenyuan Li, Zhengxia Zou, and Zhenwei Shi. Rsprompter: Learning to prompt for remote sensing instance segmenta- tion based on visual foundation model. IEEE Transactions on Geoscience and Remote Sensing, 2024. 3

  12. [20]

    Sdpl: Shifting-dense partition learning for uav-view geo- localization

    Quan Chen, Tingyu Wang, Zihao Yang, Haoran Li, Rongfeng Lu, Yaoqi Sun, Bolun Zheng, and Chenggang Yan. Sdpl: Shifting-dense partition learning for uav-view geo- localization. IEEE Transactions on Circuits and Systems for Video Technology, pages 1–1, 2024. 3

  13. [21]

    Real-time geo-localization using satellite imagery and topography for unmanned aerial vehicles

    Shuxiao Chen, Xiangyu Wu, Mark W Mueller, and Koushil Sreenath. Real-time geo-localization using satellite imagery and topography for unmanned aerial vehicles. In 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 2275–2281. IEEE, 2021. 1

  14. [22]

    Sam fails to segment anything?–sam-adapter: Adapt- ing sam in underperformed scenes: Camouflage, shadow, medical image segmentation, and more

    Tianrun Chen, Lanyun Zhu, Chaotao Ding, Runlong Cao, Yan Wang, Zejian Li, Lingyun Sun, Papa Mao, and Ying Zang. Sam fails to segment anything?–sam-adapter: Adapt- ing sam in underperformed scenes: Camouflage, shadow, medical image segmentation, and more. arXiv preprint arXiv:2...

  15. [23]

    Monte carlo filtering on lie groups

    Alessandro Chiuso and Stefano Soatto. Monte carlo filtering on lie groups. In Proceedings of the 39th IEEE Conference on Decision and Control (Cat. No. 00CH37187), pages 304–

  16. [24]

    Robust 3d vi- sual tracking using particle filtering on the special euclidean group: A combined approach of keypoint and edge features

    Changhyun Choi and Henrik I Christensen. Robust 3d vi- sual tracking using particle filtering on the special euclidean group: A combined approach of keypoint and edge features. The International Journal of Robotics Research, 31(4):498– 519, 2012. 5

  17. [25]

    Optimal randomized ransac

    Ond ˇrej Chum and Ji ˇr´ı Matas. Optimal randomized ransac. IEEE Transactions on Pattern Analysis and Machine Intelli- gence, 30(8):1472–1482, 2008. 3

  18. [26]

    Locally opti- mized ransac

    Ond ˇrej Chum, Ji ˇr´ı Matas, and Josef Kittler. Locally opti- mized ransac. In Pattern Recognition: 25th DAGM Sympo- sium, Magdeburg, Germany, September 10-12, 2003. Pro- ceedings 25, pages 236–243. Springer, 2003. 3

  19. [27]

    A transformer-based feature segmentation and region align- ment method for uav-view geo-localization

    Ming Dai, Jianhong Hu, Jiedong Zhuang, and Enhui Zheng. A transformer-based feature segmentation and region align- ment method for uav-view geo-localization. IEEE Transac- tions on Circuits and Systems for Video Technology , 32(7): 4376–4389, 2021. 3

  20. [28]

    Segment anything model (sam) for digital pathology: Assess zero- shot segmentation on whole slide imaging

    Ruining Deng, Can Cui, Quan Liu, Tianyuan Yao, Lu- cas W Remedios, Shunxing Bao, Bennett A Landman, Lee E Wheless, Lori A Coburn, Keith T Wilson, et al. Segment anything model (sam) for digital pathology: Assess zero- shot segmentation on whole slide imaging. arXiv preprint ar...

  21. [29]

    Superpoint: Self-supervised interest point detection and description

    Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabi- novich. Superpoint: Self-supervised interest point detection and description. In Proceedings of the IEEE conference on 9 computer vision and pattern recognition workshops , pages 224–236, 2018. 3, 6

  22. [30]

    Roma: Robust dense fea- ture matching

    Johan Edstedt, Qiyu Sun, Georg B ¨okman, M ˚arten Wadenb¨ack, and Michael Felsberg. Roma: Robust dense fea- ture matching. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19790– 19800, 2024. 6

  23. [31]

    Pranet: Parallel reverse attention network for polyp segmentation

    Deng-Ping Fan, Ge-Peng Ji, Tao Zhou, Geng Chen, Huazhu Fu, Jianbing Shen, and Ling Shao. Pranet: Parallel reverse attention network for polyp segmentation. In International conference on medical image computing and computer- assisted intervention, pages 263–273. Springer, 2020. 5

  24. [32]

    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. 3

  25. [33]

    gdls*: Gener- alized pose-and-scale estimation given scale and gravity pri- ors

    Victor Fragoso, Joseph DeGol, and Gang Hua. gdls*: Gener- alized pose-and-scale estimation given scale and gravity pri- ors. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), 2020. 3

  26. [34]

    Bags of binary words for fast place recognition in image sequences

    Dorian G ´alvez-L´opez and Juan D Tardos. Bags of binary words for fast place recognition in image sequences. IEEE Transactions on robotics, 28(5):1188–1197, 2012. 3

  27. [35]

    Self-supervising fine-grained region similarities for large-scale image localization

    Yixiao Ge, Haibo Wang, Feng Zhu, Rui Zhao, and Hong- sheng Li. Self-supervising fine-grained region similarities for large-scale image localization. In Computer Vision– ECCV 2020: 16th European Conference, Glasgow, UK, Au- gust 23–28, 2020, Proceedings, Part IV 16 , pages 369–...

  28. [36]

    Ogc city geography markup language (citygml) encoding standard

    Gerhard Gr ¨oger, Thomas H Kolbe, Claus Nagel, and Karl- Heinz H ¨afele. Ogc city geography markup language (citygml) encoding standard. Open Geospatial Consortium: Wayland, 2012. 2

  29. [37]

    Review and analysis of solutions of the three point perspective pose estimation problem

    Bert M Haralick, Chung-Nan Lee, Karsten Ottenberg, and Michael N¨olle. Review and analysis of solutions of the three point perspective pose estimation problem. International journal of computer vision, 13:331–356, 1994. 3

  30. [38]

    Patch-netvlad: Multi-scale fusion of locally-global descriptors for place recognition

    Stephen Hausler, Sourav Garg, Ming Xu, Michael Milford, and Tobias Fischer. Patch-netvlad: Multi-scale fusion of locally-global descriptors for place recognition. In Proceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14141–14152, 2021. 3

  31. [39]

    Investigating the role of image re- trieval for visual localization: An exhaustive benchmark

    Martin Humenberger, Yohann Cabon, No ´e Pion, Philippe Weinzaepfel, Donghwan Lee, Nicolas Gu ´erin, Torsten Sat- tler, and Gabriela Csurka. Investigating the role of image re- trieval for visual localization: An exhaustive benchmark. In- ternational Journal of Computer Vision,...

  32. [40]

    Particle filter networks with application to visual localization

    Peter Karkus, David Hsu, and Wee Sun Lee. Particle filter networks with application to visual localization. In Confer- ence on robot learning, pages 169–178. PMLR, 2018. 3, 4

  33. [41]

    Segment anything in high qual- ity

    Lei Ke, Mingqiao Ye, Martin Danelljan, Yu-Wing Tai, Chi- Keung Tang, Fisher Yu, et al. Segment anything in high qual- ity. Advances in Neural Information Processing Systems, 36,

  34. [42]

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

    Alex Kendall, Matthew Grimes, and Roberto Cipolla. Posenet: A convolutional network for real-time 6-dof cam- era relocalization. In Proceedings of the IEEE international conference on computer vision, pages 2938–2946, 2015. 4

  35. [43]

    Segment any- thing

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C Berg, Wan-Yen Lo, et al. Segment any- thing. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 4015–4026, 2023. 3

  36. [44]

    A novel parametrization of the perspective-three-point prob- lem for a direct computation of absolute camera position and orientation

    Laurent Kneip, Davide Scaramuzza, and Roland Siegwart. A novel parametrization of the perspective-three-point prob- lem for a direct computation of absolute camera position and orientation. In CVPR 2011, pages 2969–2976. IEEE, 2011. 3

  37. [45]

    Ogc city geography markup language (citygml) version 3.0 part 2: Gml encoding standard

    Tatjana Kutzner, Carl Smyth, Claus Nagel, V olker Coors, Diego Vinasco-Alvarez, Nobuhiro Ishimaru, Zhihang Yao, Charles Heazel, and Thomas H Kolbe. Ogc city geography markup language (citygml) version 3.0 part 2: Gml encoding standard. https://docs.ogc.org/is/21-006r2/21-006r2.html.,

  38. [46]

    Visual tracking via par- ticle filtering on the affine group

    Junghyun Kwon and Frank C Park. Visual tracking via par- ticle filtering on the affine group. The International Journal of Robotics Research, 29(2-3):198–217, 2010. 5

  39. [47]

    Particle filtering on the euclidean group: framework and applications

    Junghyun Kwon, Minseok Choi, Frank Chongwoo Park, and Changmook Chun. Particle filtering on the euclidean group: framework and applications. Robotica, 25(6):725– 737, 2007. 2

  40. [48]

    Monocular model-based 3d tracking of rigid objects: A survey

    Vincent Lepetit, Pascal Fua, et al. Monocular model-based 3d tracking of rigid objects: A survey. Foundations and Trends® in Computer Graphics and Vision, 1(1):1–89, 2005. 5

  41. [49]

    Parallel inversion of neural radiance fields for robust pose estimation

    Yunzhi Lin, Thomas M ¨uller, Jonathan Tremblay, Bowen Wen, Stephen Tyree, Alex Evans, Patricio A Vela, and Stan Birchfield. Parallel inversion of neural radiance fields for robust pose estimation. In 2023 IEEE International Confer- ence on Robotics and Automation (ICRA), pages...

  42. [50]

    Receptive field block net for accurate and fast object detection

    Songtao Liu, Di Huang, et al. Receptive field block net for accurate and fast object detection. In Proceedings of the Eu- ropean conference on computer vision (ECCV) , pages 385– 400, 2018. 3

  43. [51]

    Distinctive image features from scale- invariant keypoints

    David G Lowe. Distinctive image features from scale- invariant keypoints. International journal of computer vi- sion, 60:91–110, 2004. 3, 6

  44. [52]

    Towards seamless adapta- tion of pre-trained models for visual place recognition.arXiv preprint arXiv:2402.14505, 2024

    Feng Lu, Lijun Zhang, Xiangyuan Lan, Shuting Dong, Yaowei Wang, and Chun Yuan. Towards seamless adapta- tion of pre-trained models for visual place recognition.arXiv preprint arXiv:2402.14505, 2024. 3

  45. [53]

    A survey on vision-based uav navigation

    Yuncheng Lu, Zhucun Xue, Gui-Song Xia, and Liangpei Zhang. A survey on vision-based uav navigation. Geo- spatial information science, 21(1):21–32, 2018. 1

  46. [54]

    Large-scale, real-time visual–inertial localization re- visited

    Simon Lynen, Bernhard Zeisl, Dror Aiger, Michael Bosse, Joel Hesch, Marc Pollefeys, Roland Siegwart, and Torsten Sattler. Large-scale, real-time visual–inertial localization re- visited. The International Journal of Robotics Research, 39 (9):1061–1084, 2020. 3

  47. [55]

    Segment anything in medical images

    Jun Ma, Yuting He, Feifei Li, Lin Han, Chenyu You, and Bo Wang. Segment anything in medical images. Nature Communications, 15(1):654, 2024. 3 10

  48. [56]

    Loc-nerf: Monte carlo local- ization using neural radiance fields

    Dominic Maggio, Marcus Abate, Jingnan Shi, Courtney Mario, and Luca Carlone. Loc-nerf: Monte carlo local- ization using neural radiance fields. In 2023 IEEE Inter- national Conference on Robotics and Automation (ICRA) , pages 4018–4025. IEEE, 2023. 4

  49. [57]

    Segment anything model for medical image analysis: an experimental study

    Maciej A Mazurowski, Haoyu Dong, Hanxue Gu, Jichen Yang, Nicholas Konz, and Yixin Zhang. Segment anything model for medical image analysis: an experimental study. Medical Image Analysis, 89:102918, 2023. 3

  50. [58]

    Visual place recognition for aerial imagery: A survey

    Ivan Moskalenko, Anastasiia Kornilova, and Gonzalo Ferrer. Visual place recognition for aerial imagery: A survey. arXiv preprint arXiv:2406.00885, 2024. 3

  51. [59]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 6

  52. [60]

    Meshloc: Mesh-based visual localization

    V ojtech Panek, Zuzana Kukelova, and Torsten Sattler. Meshloc: Mesh-based visual localization. In European Con- ference on Computer Vision, pages 589–609. Springer, 2022. 2, 3

  53. [61]

    Visual localization using imperfect 3d models from the internet

    V ojtech Panek, Zuzana Kukelova, and Torsten Sattler. Visual localization using imperfect 3d models from the internet. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 13175–13186, 2023. 2, 5, 6

  54. [62]

    Automated 3d reconstruction of lod2 and lod1 models for all 10 million buildings of the nether- lands

    Ravi Peters, Bal ´azs Dukai, Stelios Vitalis, Jordi van Liempt, and Jantien Stoter. Automated 3d reconstruction of lod2 and lod1 models for all 10 million buildings of the nether- lands. Photogrammetric Engineering & Remote Sensing, 88 (3):165–170, 2022. 2

  55. [63]

    Segloc: Learning segmentation-based representations for privacy-preserving visual localization

    Maxime Pietrantoni, Martin Humenberger, Torsten Sattler, and Gabriela Csurka. Segloc: Learning segmentation-based representations for privacy-preserving visual localization. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 15380–153...

  56. [64]

    Sam 2: Segment anything in images and videos

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman R¨adle, Chloe Rolland, Laura Gustafson, et al. Sam 2: Segment anything in images and videos. arXiv preprint arXiv:2408.00714, 2024. 3

  57. [65]

    Segment anything, from space? In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 8355–8365, 2024

    Simiao Ren, Francesco Luzi, Saad Lahrichi, Kaleb Kas- saw, Leslie M Collins, Kyle Bradbury, and Jordan M Malof. Segment anything, from space? In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 8355–8365, 2024. 3

  58. [66]

    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. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 12716–12725, 2019. 3

  59. [67]

    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. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4938–4947, 2020. 3, 6

  60. [68]

    Orienternet: Visual localization in 2d public maps with neural matching

    Paul-Edouard Sarlin, Daniel DeTone, Tsun-Yi Yang, Armen Avetisyan, Julian Straub, Tomasz Malisiewicz, Samuel Rota Bulo, Richard Newcombe, Peter Kontschieder, and Vasileios Balntas. Orienternet: Visual localization in 2d public maps with neural matching. In Proceedings of the I...

  61. [69]

    Snap: Self-supervised neural maps for visual positioning and semantic understanding.Ad- vances in Neural Information Processing Systems, 36, 2024

    Paul-Edouard Sarlin, Eduard Trulls, Marc Pollefeys, Jan Hosang, and Simon Lynen. Snap: Self-supervised neural maps for visual positioning and semantic understanding.Ad- vances in Neural Information Processing Systems, 36, 2024. 3

  62. [70]

    Structure- from-motion revisited

    Johannes L Schonberger and Jan-Michael Frahm. Structure- from-motion revisited. In Proceedings of the IEEE con- ference on computer vision and pattern recognition , pages 4104–4113, 2016. 2, 3

  63. [71]

    Role of 3d city model data as open digital commons: a case study of openness in japan’s digital twin” project plateau”

    T Seto, T Furuhashi, and Y Uchiyama. Role of 3d city model data as open digital commons: a case study of openness in japan’s digital twin” project plateau”. The International Archives of the Photogrammetry, Remote Sensing and Spa- tial Information Sciences, 48:201–208, 2023. 2

  64. [72]

    The last puzzle of global building footprints—mapping 280 million build- ings in east asia based on vhr images

    Qian Shi, Jiajun Zhu, Zhengyu Liu, Haonan Guo, Song Gao, Mengxi Liu, Zihong Liu, and Xiaoping Liu. The last puzzle of global building footprints—mapping 280 million build- ings in east asia based on vhr images. Journal of Remote Sensing, 4:0138, 2024. 2

  65. [73]

    Beyond cross-view image retrieval: Highly accurate vehicle localization using satel- lite image

    Yujiao Shi and Hongdong Li. Beyond cross-view image retrieval: Highly accurate vehicle localization using satel- lite image. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17010– 17020, 2022. 3

  66. [74]

    Where am i looking at? joint location and orientation es- timation by cross-view matching

    Yujiao Shi, Xin Yu, Dylan Campbell, and Hongdong Li. Where am i looking at? joint location and orientation es- timation by cross-view matching. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4064–4072, 2020. 3

  67. [75]

    Loftr: Detector-free local feature matching with transformers

    Jiaming Sun, Zehong Shen, Yuang Wang, Hujun Bao, and Xiaowei Zhou. Loftr: Detector-free local feature matching with transformers. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 8922–8931, 2021. 3, 6

  68. [76]

    Gable: A first fine-grained 3d building model of china on a national scale from very high resolu- tion satellite imagery

    Xian Sun, Xingliang Huang, Yongqiang Mao, Taowei Sheng, Jihao Li, Zhirui Wang, Xue Lu, Xiaoliang Ma, Deke Tang, and Kaiqiang Chen. Gable: A first fine-grained 3d building model of china on a national scale from very high resolu- tion satellite imagery. Remote Sensing of Enviro...

  69. [77]

    Can sam segment any- thing? when sam meets camouflaged object detection

    Lv Tang, Haoke Xiao, and Bo Li. Can sam segment any- thing? when sam meets camouflaged object detection. arXiv preprint arXiv:2304.04709, 2023. 3

  70. [78]

    Probabilistic robotics

    Sebastian Thrun. Probabilistic robotics. Communications of the ACM, 45(3):52–57, 2002. 2

  71. [79]

    Long-term visual localization revisited

    Carl Toft, Will Maddern, Akihiko Torii, Lars Hammarstrand, Erik Stenborg, Daniel Safari, Masatoshi Okutomi, Marc Pollefeys, Josef Sivic, Tomas Pajdla, et al. Long-term visual localization revisited. IEEE Transactions on Pattern Analy- sis and Machine Intelligence, 44(4):2074–2...

  72. [80]

    The unreasonable effectiveness of pre- trained features for camera pose refinement

    Gabriele Trivigno, Carlo Masone, Barbara Caputo, and Torsten Sattler. The unreasonable effectiveness of pre- trained features for camera pose refinement. In Proceedings 11 of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 12786–12798, 2024. 2, 4, 5, 6

  73. [81]

    A survey on load transportation using mul- tirotor uavs

    Daniel KD Villa, Alexandre S Brandao, and M ´ario Sarcinelli-Filho. A survey on load transportation using mul- tirotor uavs. Journal of Intelligent & Robotic Systems , 98: 267–296, 2020. 1

  74. [82]

    Efficient loftr: Semi-dense local feature matching with sparse-like speed

    Yifan Wang, Xingyi He, Sida Peng, Dongli Tan, and Xiaowei Zhou. Efficient loftr: Semi-dense local feature matching with sparse-like speed. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 21666–21675, 2024. 6

  75. [83]

    F3net: fusion, feedback and focus for salient object detection

    Jun Wei, Shuhui Wang, and Qingming Huang. F3net: fusion, feedback and focus for salient object detection. In Proceed- ings of the AAAI conference on artificial intelligence , pages 12321–12328, 2020. 5

  76. [84]

    Maplocnet: Coarse-to- fine feature registration for visual re-localization in naviga- tion maps

    Hang Wu, Zhenghao Zhang, Siyuan Lin, Xiangru Mu, Qiang Zhao, Ming Yang, and Tong Qin. Maplocnet: Coarse-to- fine feature registration for visual re-localization in naviga- tion maps. arXiv preprint arXiv:2407.08561, 2024. 3

  77. [85]

    Medical sam adapter: Adapting seg- ment anything model for medical image segmentation.arXiv preprint arXiv:2304.12620, 2023

    Junde Wu, Wei Ji, Yuanpei Liu, Huazhu Fu, Min Xu, Yanwu Xu, and Yueming Jin. Medical sam adapter: Adapting seg- ment anything model for medical image segmentation.arXiv preprint arXiv:2304.12620, 2023. 3

  78. [86]

    Uavd4l: A large-scale dataset for uav 6-dof localization

    Rouwan Wu, Xiaoya Cheng, Juelin Zhu, Yuxiang Liu, Mao- jun Zhang, and Shen Yan. Uavd4l: A large-scale dataset for uav 6-dof localization. In 2024 International Conference on 3D Vision (3DV), pages 1574–1583. IEEE, 2024. 1, 2, 3, 5, 6, 7

  79. [87]

    Reviewing open data seman- tic 3d city models to develop novel 3d reconstruction meth- ods

    Olaf Wysocki, Benedikt Schwab, Christof Beil, Christoph Holst, and Thomas H Kolbe. Reviewing open data seman- tic 3d city models to develop novel 3d reconstruction meth- ods. The International Archives of the Photogrammetry, Re- mote Sensing and Spatial Information Sciences, 4...

  80. [88]

    Visual cross-view metric localization with dense un- certainty estimates

    Zimin Xia, Olaf Booij, Marco Manfredi, and Julian FP Kooij. Visual cross-view metric localization with dense un- certainty estimates. In European Conference on Computer Vision, pages 90–106. Springer, 2022. 3

  81. [89]

    Moving object segmentation: All you need is sam (and flow)

    Junyu Xie, Charig Yang, Weidi Xie, and Andrew Zisserman. Moving object segmentation: All you need is sam (and flow). arXiv preprint arXiv:2404.12389, 2024. 3

  82. [90]

    Sam2-unet: Segment anything 2 makes strong encoder for natural and medical image segmentation

    Xinyu Xiong, Zihuang Wu, Shuangyi Tan, Wenxue Li, Fei- long Tang, Ying Chen, Siying Li, Jie Ma, and Guanbin Li. Sam2-unet: Segment anything 2 makes strong encoder for natural and medical image segmentation. arXiv preprint arXiv:2408.08870, 2024. 3

  83. [91]

    Efficientsam: Leveraged masked image pretraining for efficient segment anything

    Yunyang Xiong, Bala Varadarajan, Lemeng Wu, Xiaoyu Xi- ang, Fanyi Xiao, Chenchen Zhu, Xiaoliang Dai, Dilin Wang, Fei Sun, Forrest Iandola, et al. Efficientsam: Leveraged masked image pretraining for efficient segment anything. In Proceedings of the IEEE/CVF Conference on Compu...

  84. [92]

    Render-and-compare: Cross-view 6-dof localization from noisy prior

    Shen Yan, Xiaoya Cheng, Yuxiang Liu, Juelin Zhu, Rouwan Wu, Yu Liu, and Maojun Zhang. Render-and-compare: Cross-view 6-dof localization from noisy prior. In 2023 IEEE International Conference on Multimedia and Expo (ICME), pages 2171–2176. IEEE, 2023. 2

  85. [93]

    Long-term visual localization with mobile sensors

    Shen Yan, Yu Liu, Long Wang, Zehong Shen, Zhen Peng, Haomin Liu, Maojun Zhang, Guofeng Zhang, and Xiaowei Zhou. Long-term visual localization with mobile sensors. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17245–17255, 2023. 3

  86. [94]

    Faster segment anything: Towards lightweight sam for mo- bile applications

    Chaoning Zhang, Dongshen Han, Yu Qiao, Jung Uk Kim, Sung-Ho Bae, Seungkyu Lee, and Choong Seon Hong. Faster segment anything: Towards lightweight sam for mo- bile applications. arXiv preprint arXiv:2306.14289 , 2023. 3

  87. [95]

    Fast segment any- thing

    Xu Zhao, Wenchao Ding, Yongqi An, Yinglong Du, Tao Yu, Min Li, Ming Tang, and Jinqiao Wang. Fast segment any- thing. arXiv preprint arXiv:2306.12156, 2023. 3, 8

  88. [96]

    University- 1652: A multi-view multi-source benchmark for drone- based geo-localization

    Zhedong Zheng, Yunchao Wei, and Yi Yang. University- 1652: A multi-view multi-source benchmark for drone- based geo-localization. In Proceedings of the 28th ACM international conference on Multimedia , pages 1395–1403,

  89. [97]

    Lod-loc: Aerial visual localization using lod 3d map with neural wireframe alignment

    Juelin Zhu, Shen Yan, Long Wang, zhang shengYue, Yu Liu, and Maojun Zhang. Lod-loc: Aerial visual localization using lod 3d map with neural wireframe alignment. In Advances in Neural Information Processing Systems , pages 119063– 119098, 2024. 1, 2, 3, 5, 6, 7

  90. [98]

    R2former: Unified retrieval and reranking transformer for place recognition

    Sijie Zhu, Linjie Yang, Chen Chen, Mubarak Shah, Xiao- hui Shen, and Heng Wang. R2former: Unified retrieval and reranking transformer for place recognition. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 19370–19380, 2023. 3 12

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.