Pith. sign in

REVIEW 4 major objections 8 minor 81 references

SaliencyI2PLoc: saliency-guided image-point cloud localization using contrastive learning

T0 review · 4 major / 8 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read SaliencyI2PLoc claims image-to-point-cloud localization with 78.92% Recall@1 on urban KITTI-360, 37.35 points above a retrained AE-Spherical baseline.

desk verdict Sound architectural idea and honest ablations, but headline gains against an unreproduced retrained baseline means the exact margins are not robust; still worth a serious referee. read the letter →

arxiv 2412.15577 v1 pith:UQ3J6AYB submitted 2024-12-20 cs.CV cs.LGcs.RO

classification cs.CVcs.LGcs.RO
keywords cross-modalitylocalizationglobalcontrastivelearningsaliency-guidedaggregationNetVLADpointcloudretrievalplacerecognitiondual-transformer
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to settle a practical problem: given a single photo and a pre-built LiDAR point-cloud map, find the submap where the photo was taken without GPS. The authors argue the right way to bridge the image-point-cloud gap is a dual-transformer network trained end-to-end with contrastive learning, so no triplet or quadruplet mining is needed, and the attention maps from the final transformer layer should be folded into the NetVLAD aggregation so that stationary landmarks dominate the global descriptor. They report Recall@1 of 78.92% and Recall@20 of 97.59% on a KITTI-360 urban evaluation set, improving on their retrained AE-Spherical baseline by 37.35 and 18.07 points, and outperform the modality-transformation baseline LIP-Loc by 14.46 and 6.32 points. If the result holds, robot and vehicle localization in GNSS-denied urban canyons can be done with a camera against a LiDAR map rather than a second LiDAR unit.

What carries the argument

The load-bearing mechanism is the saliency-weighted NetVLAD layer, an otherwise standard VLAD descriptor in which the soft-assignment residual $\sum_i a_k(f_i)(f_i(j)-c_k(j))$ is multiplied by the attention vector $\mathbf{Attn}$ taken from the final transformer block, so highly attended local patches dominate the global descriptor. The second mechanism is the multi-manifold feature-relation consistency loss, which supervises that pairwise distances between image features match pairwise distances between the corresponding point-cloud features, measured with Euclidean distance and with the hyperbolic distance of the Poincaré ball model, and added to the InfoNCE loss. Together they let the same Transformer backbone process 2D image patches and 3D PointNet-tokenized patches while projecting both modalities into one embedding space.

What would settle it

Run the released code on the same KITTI-360 splits and retrain the AE-Spherical baseline until it reaches the recall values originally published in [10]; if a properly tuned baseline reaches or exceeds the 78.92% Recall@1 reported here, the claimed architecture advantage is disproven.

Watch

Extended reading notes

Core claim

The central claim is that cross-modal image-to-point-cloud localization reduces to learning a shared embedding in which a query image and its matching point-cloud submap are nearest neighbours, and that this embedding is best learned by a dual-transformer architecture trained with InfoNCE contrastive loss together with a pair-distance consistency penalty computed in both Euclidean and hyperbolic space. The saliency weight from the last transformer block is multiplied into the NetVLAD residual aggregation, so patches attending to buildings, signs, and other stationary objects contribute more to the global feature than sky, ground, or moving vehicles. On the urban KITTI-360 evaluation subset the method reaches 78.92% Recall@1 and 97.59% Recall@20 at a 20-meter tolerance, gains of 37.35 and 18.07 points over the retrained AE-Spherical baseline, and the authors report comparable parameter counts with slower inference (24.15 FPS versus 84.49 FPS for the baseline).

Load-bearing premise

The 37.35-point margin is measured against the authors' own retrained version of AE-Spherical, whose published results they say they could not reproduce; if that retrained baseline is weaker than the original, the gap is inflated.

Editorial extensions

If this is right

  • Contrastive training removes the need to mine hard triplets or quadruplets, so adding new places to the map can be done by retraining on image-point-cloud pairs instead of rebuilding negative sets.
  • A model trained on a quarter of the data still reaches 69.28% Recall@1 on urban KITTI-360, suggesting the approach scales with data volume without engineered pair construction.
  • At a 20-meter tolerance, the dense urban scenario is effectively solved at 97.59% Recall@20, so the remaining research problem is the highway/open-road case where Recall@1 drops to 30.94%.
  • On KITTI's sparse raw LiDAR frames and perspective images, Recall@20 improves to 72.59% versus 63.00% for VXP, indicating some robustness to point-cloud density and field-of-view differences.
  • The higher inference cost (24.15 versus 84.49 FPS) means deployment on mobile robots will likely need the proposed flash-attention or distillation speedups.

Reading between the lines

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

  • If the ablation table is the right lens, the single biggest internal jump comes from the contrastive framework replacing triplet mining (SCNN-Contrast at 60.24% versus AE-Spherical at 41.57% Recall@1), while adding saliency weighting and the two relation-consistency terms contributes a further 9.04 points; that ordering suggests the loss function, not the hyperbolic geometry, carries most of the i
  • The hyperbolic-consistency term alone adds little in the ablation (65.96% with hyperbolic, 66.27% with Euclidean), so a testable extension is to replace it with a cheaper regularizer, such as distance-distribution alignment, and measure whether Recall@1 holds.
  • Because the authors could not reproduce the original AE-Spherical numbers, the 37.35-point headline gap should be read as a comparison to a re-trained baseline; a community benchmark with a fixed public split and released model weights would settle whether the margin persists.
  • The same saliency-weighted aggregation could be applied to radar or sonar place recognition, where stationary, reflective structures likewise dominate reliable localization, as a direct transfer of the mechanism beyond images and LiDAR.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 8 minor

Summary. This paper addresses cross-modal global localization from a 2D image to a 3D point-cloud map. The authors propose SaliencyI2PLoc, a dual-tower architecture in which a Vision Transformer (ViT) encodes the image and a PointNet-plus-Transformer tokenizer encodes the point cloud. The local patch features are aggregated by a NetVLAD layer that is weighted by self-attention-derived saliency scores. The network is trained end-to-end with an InfoNCE contrastive loss plus a multi-manifold feature relation consistency loss that enforces agreement between image-pair and point-cloud-pair relations in Euclidean (dot-product) and hyperbolic spaces. Experiments on KITTI-360 (urban and highway) and KITTI report Recall@N and F1 metrics, claiming a Recall@1 of 78.92% on the urban evaluation set, an improvement of 37.35 percentage points over a retrained AE-Spherical baseline. Ablation studies show that each proposed component contributes to performance on a quarter of the training data.

Significance. The paper proposes a clean, end-to-end contrastive framework that avoids explicit triplet mining, integrates saliency-aware aggregation into NetVLAD, and augments the loss with relation consistency in multiple manifolds. The ablations and qualitative visualizations provide initial evidence that each component helps. The reported gains on KITTI-360 are substantial if the comparison baseline is properly tuned. However, the headline improvement is measured against a retrained AE-Spherical baseline that the authors themselves could not reproduce from the original settings, so the exact magnitude of the improvement remains uncertain. The paper also promises to release code and the dataset generation pipeline, which would significantly enhance reproducibility. The method's applicability to highway scenes is weaker, but the open-scenario discussion is balanced.

major comments (4)
  1. [Section 4.3, Table 2] The central claim of a 37.35 pp Recall@1 improvement on urban scenes is calculated against the retrained AE-Spherical baseline (R1 41.57), but the authors' own estimate of the published AE-Spherical result is R1 46.23 (AE-Spherical*). The paper states that the original results "could not be reproduced following their training settings," yet it provides no analysis of this discrepancy, no variance over multiple training runs, and no release of baseline checkpoints. This makes the magnitude of the claimed improvement unsupported. Please report the comparison against the published numbers (or the AE-Spherical* estimates), provide training details for the retrained baseline (epochs, LR schedule, data splits, augmentations), and include error bars over at least three runs on the 332-query evaluation set.
  2. [Section 3.3, Eq. (16)] The text in Section 3.3 says that D_euc is the Euclidean distance between features, but the sentence after Eq. (16) states "D_euc represent the dot product." These are contradictory. The dot product is a similarity, not a distance, and the MSE between dot products and hyperbolic distances (which are metrics with different scales) is not a standard consistency measure. Please specify exactly which quantity is used in the implementation, and adjust the terminology accordingly (e.g., "cosine similarity consistency" instead of "Euclidean distance").
  3. [Section 4.5, Table 5] The ablation study is run on a 1/4 subset of the training pairs, while the main results are obtained on the full dataset. The contribution of the saliency and relation-consistency components may change with dataset scale (the full model gains ~10 pp R1 over the 1/4 model). To support the claim that the proposed components are effective, please run the ablation on the full dataset or explicitly justify the proxy.
  4. [Section 3.2] The saliency score Attn used in Eq. (8) is not uniquely defined. For images, ViT produces a multi-head attention matrix among N+1 tokens (including the class token), and for point clouds there is no class token. Please provide the exact formula for computing the per-token scalar saliency (e.g., CLS attention averaged over heads, or mean of attention rows), so that the method is reproducible.
minor comments (8)
  1. [Abstract and Section 4.4.1] "Improvement of 37.35%" should be rephrased as "37.35 percentage points" to avoid confusion with a relative improvement.
  2. [Table 2] For AE-Spherical*, the Max F1 column contains "***"; replace with "n/a" or a value, as the current notation is unexplained.
  3. [Section 4.1.1] The statement that "there are no publicly available datasets for cross-modality global localization tasks" is overstated. Consider citing existing cross-modal place recognition benchmarks or limiting the claim to the specific image-panorama-to-submap setting.
  4. [Section 4.4.1] The highway-scenario results show that LIP-Loc (an image-projection method) outperforms the proposed method. The paper discusses this as a limitation of feature-based methods in structureless areas, but the reader would benefit from a quantitative analysis (e.g., how the proposed method's point cloud branch fails in these scenes).
  5. [Section 4.6] The reported inference speed (24.15 FPS vs 84.49 FPS for AE-Spherical) is a substantial deployment concern. Please mention this trade-off in the abstract or conclusions.
  6. [Eq. (8)] The multiplication "Attn·a_k(f_i)" should be indexed as Attn_i to clarify that it is a per-local-feature scalar.
  7. [Section 3.1.1] The equations for the Transformer blocks (Eqs. (2)-(3)) do not explicitly handle the class token; add a sentence clarifying how the class token is processed and whether it is used in the saliency computation.
  8. [Section 4.2] The statement "we froze the Transformer blocks except for the last 4 blocks" refers to the image branch; clarify whether the point cloud Transformer is trained from scratch and whether any other components are frozen.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the method's recall gains are measured on held-out data against external and retrained baselines, and no prediction reduces by construction to a fitted input or self-citation chain.

full rationale

SaliencyI2PLoc's derivation chain is self-contained. The global features are produced by Eqs. (8)-(10) as a saliency-weighted NetVLAD aggregation, with the attention weights taken from the last Transformer block rather than from the evaluation labels; the saliency map is an internal, learned weighting, not a fitted copy of the Recall@N targets. The training objective in Eq. (17) combines InfoNCE (Eq. 11) with the relation-consistency loss (Eq. 16), and the evaluation metric Recall@N (Eq. 19) is computed on a held-out query set (Table 1) using a 20 m Euclidean threshold. None of these quantities is defined in terms of the Recall@N numbers reported in Section 4.4.1, so the central claim does not reduce to its own inputs. The paper's own limitation statements concern the retrained AE-Spherical baseline (Section 4.3: 'the results listed in the paper could not be reproduced following their training settings') and higher inference time (Section 4.6); these bear on the magnitude and cost of the reported improvement, not on circularity. Self-citations such as references [2] and [33] are contextual and are not used to justify the uniqueness or correctness of the proposed loss or architecture. The comparison also includes the independent LIP-Loc baseline and VXP numbers on KITTI, so the method is not validated solely against a baseline produced by the same authors.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claim rests on standard deep-learning components and a handful of hand-set hyperparameters. The most assumption-heavy steps are the treatment of transformer attention as saliency and the use of hyperbolic distance without a described projection procedure. The retrained baseline is a significant unverifiable input.

free parameters (6)
  • lambda (λ) = 1
    Weight for Euclidean relation consistency loss in Eq. (16); set by hand to 1 with no sensitivity analysis.
  • beta (β) = 2.0
    Weight for Hyperbolic relation consistency loss in Eq. (16); set to 2.0 with no sensitivity analysis.
  • InfoNCE temperature (τ) = 0.07
    Default temperature for the InfoNCE loss (Section 4.2).
  • Number of 3D tokens = 3072
    Selected because recall saturates at 3072 points in Table 6 ablation; larger counts increase training time.
  • VLAD cluster count K = 64
    Architecture choice in the saliency-guided NetVLAD layer (Section 3.2).
  • Global descriptor dimension D_g = 256
    Output dimension of the global feature for both modalities (Section 3.2).
assumptions (5)
  • domain assumption Self-attention weights from the last Transformer block are a valid saliency signal for localization-relevant scene content.
    Section 3.2 and Figure 4 assume that attention scores indicate which local patches (buildings, road signs) should be weighted higher in NetVLAD; no external saliency supervision is used.
  • domain assumption Relative distances between samples should be consistent across image and point cloud modalities in both Euclidean and Hyperbolic spaces.
    Section 3.3 defines the relation consistency loss under this premise; the paper does not prove that this constraint is theoretically justified, but it is a plausible regularization.
  • domain assumption Features can be mapped into the Poincaré ball via the exponential map without explicit norm clipping or curvature tuning.
    Equation (15) gives the exponential map, but the paper does not describe how the global features are normalized or constrained to satisfy the ball condition; implementation details are omitted.
  • domain assumption The retrained AE-Spherical baseline is a fair and faithful implementation of Zhao et al. [10].
    Section 4.3 states that the original results could not be reproduced; the authors retrained the model themselves, and the reported improvements are relative to that retrained version.
  • domain assumption KITTI-360 fisheye stitching using FFMPEG produces panoramic images that are correctly aligned with the point cloud submaps.
    Section 4.1.1 describes stitching 185-degree fisheye images into 512x1024 panoramas; the correctness of this alignment is assumed and not independently verified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SaliencyI2PLoc: saliency-guided image-point cloud localization using contrastive learning." pith.science (2026). https://pith.science/paper/UQ3J6AYB

@misc{pith2026241215577,
  author       = {Pith},
  title        = {Pith review of: SaliencyI2PLoc: saliency-guided image-point cloud localization using contrastive learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UQ3J6AYB}},
  note         = {Machine review of arXiv:2412.15577}
}
read the original abstract

Image to point cloud global localization is crucial for robot navigation in GNSS-denied environments and has become increasingly important for multi-robot map fusion and urban asset management. The modality gap between images and point clouds poses significant challenges for cross-modality fusion. Current cross-modality global localization solutions either require modality unification, which leads to information loss, or rely on engineered training schemes to encode multi-modality features, which often lack feature alignment and relation consistency. To address these limitations, we propose, SaliencyI2PLoc, a novel contrastive learning based architecture that fuses the saliency map into feature aggregation and maintains the feature relation consistency on multi-manifold spaces. To alleviate the pre-process of data mining, the contrastive learning framework is applied which efficiently achieves cross-modality feature mapping. The context saliency-guided local feature aggregation module is designed, which fully leverages the contribution of the stationary information in the scene generating a more representative global feature. Furthermore, to enhance the cross-modality feature alignment during contrastive learning, the consistency of relative relationships between samples in different manifold spaces is also taken into account. Experiments conducted on urban and highway scenario datasets demonstrate the effectiveness and robustness of our method. Specifically, our method achieves a Recall@1 of 78.92% and a Recall@20 of 97.59% on the urban scenario evaluation dataset, showing an improvement of 37.35% and 18.07%, compared to the baseline method. This demonstrates that our architecture efficiently fuses images and point clouds and represents a significant step forward in cross-modality global localization. The project page and code will be released.

Figures

Figures reproduced from arXiv: 2412.15577 by the authors.

Figure 1
Figure 1. Overview of general cross-modality global localization. Given a point cloud map and a query image, the cross-modality localization task aims to retrieve the most closet or similar pre-built point cloud submaps. strikes a balance between cost and performance, offering a more flexible and feasible approach to positioning [6]. However, the modality gaps present significant challenges to cross-modality fusion and locali… view at source ↗
Figure 2
Figure 2. The architecture of SaliencyI2PLoc. SaliencyI2PLoc encodes the input image-point cloud pairs into a high-dimensional feature embedding space using a feature encoder (ViT for images, mini-PointNet combined with Transformer for point clouds) and feature aggregator (saliency map boosted NetVLAD layer). It then achieves feature fusion and alignment through the contrastive learning loss function that incorporates cross-m… view at source ↗
Figure 3
Figure 3. Visualization of farthest point sampling (FPS) and group￾ing process, and the architecture of 3d tokenizer. Numbers in the bracket are layer sizes for the Multi-layer perceptron (MLP). Batchnorm is used for all layers with ReLU. 3D tokenizer takes in local patch point clouds and returns a 𝐷3𝑑 -dimensional feature. local feature interaction indicating the relative relations. We then fuse the saliency map into the fea… view at source ↗
Figures from the paper (12 more)
Figure 5
Figure 5. Figure 5: The pipeline of saliency-guided NetVLAD layer. The red arrow indicates the position where the saliency score is applied, while the purple blocks are the vanilla NetVLAD layer. point cloud modalities, the attention from the final layer of Transformer blocks in different…
Figure 6
Figure 6. Figure 6: The demonstration of cross-modality feature relation con￾sistency. From Euclidean space to Hyperbolic space, the exponen￾tial mapping is required. For some fixed base point 𝐮 ∈ 𝔻𝑛 𝑐 , usually set to 𝟎, the exponential mapping exp𝑐 𝐪 ∶ ℝ𝑛 → 𝔻𝑛 𝑐 is defined as: exp𝑐 𝐮 (𝐯…
Figure 7
Figure 7. Figure 7: Image-point cloud pair dataset for cross-modality global localization. This dataset covers different scenarios, such as urban which contains buildings and vehicles, and the highway scene which contains feature-less structures. (a) displays the panoramic image at the le…
Figure 8
Figure 8. Figure 8: The significant margin of recall compared to the baseline method on KITTI-360 urban evaluation datasets. 0.0 0.2 0.4 0.6 0.8 1.0 Recall 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Precision SaliencyI2PLoc(Ours) SaliencyI2PLoc-1/4(Ours) LIP-Loc AE-Spherical* [PITH_FULL_IMAGE:figures/f…
Figure 9
Figure 9. Figure 9: The precision-recall (PR) curve on KITTI-360 urban eval￾uation datasets. absence of significant landmarks presents a major challenge for feature extraction from 3D point clouds, leading to the failure of feature-based localization methodology. The flat￾ten structure-le…
Figure 10
Figure 10. Figure 10: Visualization of the cosine similarity between image and point clouds of KITTI-360 the evaluation dataset. The color gradient from green to yellow in the figure indicates a gradual increase in similarity, where red boxes highlight the bright areas. (a) AE-Spherical (b…
Figure 11
Figure 11. Figure 11: Visualization of feature descriptors on Euclidean space using t-SNE [81]. The data index is rendered from blue to red in ascending order. ▴ and ⋅ represent the feature from the query and database respectively. (a) AE-Spherical (b) SaliencyI2PLoc (ours) [PITH_FULL_IMA…
Figure 12
Figure 12. Figure 12: Visualization of feature descriptors on Hyperbolic space using UMAP. The data index is rendered from blue to red in ascending order. ▴ and ⋅ represent the feature from the query and database respectively. 4.4.3. Generalization ability We further evaluate the generaliz…
Figure 13
Figure 13. Figure 13: Visualization of the retrieval results on the KITTI-360 urban scenario evaluation dataset. The green boxes indicate the corrected results, while the red represents the failed ones. The images of each fetched point cloud are displayed for better visualization. 4.5. Abl…
Figure 14
Figure 14. Figure 14: The visualization of saliency maps of the query images. During training, we notice that the saliency map shifts to the scene layout and stationary buildings. 4.5.3. The spherical-CNN based image tokenizer The baseline model [10], employs spherical convolution to addre…
Figure 15
Figure 15. Figure 15: The VLAD cluster assignment of the query images and the Top-1 point cloud from the database. The auxiliary point clouds/images are listed for better visualization. The reference and Top-1 point clouds are rendered by the relative height, and whole point clouds are vie…
Figure 16
Figure 16. Figure 16: The failure case of our method. : Preprint submitted to Elsevier Page 13 of 17 [PITH_FULL_IMAGE:figures/full_fig_p014_16.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

81 extracted references · 37 canonical work pages

  1. [1]

    Multi-gnss ppp/ins/vision/lidartightlyintegratedsystemforprecisenavigationin urban environments

    Li, S., Li, X., Wang, H., Zhou, Y., Shen, Z., 2023a. Multi-gnss ppp/ins/vision/lidartightlyintegratedsystemforprecisenavigationin urban environments. Information Fusion 90, 218–232. doi:10.1016/ j.inffus.2022.09.018

  2. [2]

    Mucograph: A multi-scale constraint enhanced pose-graph framework for mls point cloud inconsistency correction

    Li, Y., Zou, X., Li, T., Sun, S., Wang, Y., Liang, F., Li, J., Yang, B., Dong, Z., 2023b. Mucograph: A multi-scale constraint enhanced pose-graph framework for mls point cloud inconsistency correction. ISPRS Journal of Photogrammetry and Remote Sensing 204, 421–

  3. [3]

    Information Fusion 37, 22–36

    Yuan,J.,Zhang,J.,Ding,S.,Dong,X.,2017.Cooperativelocalization for disconnected sensor networks and a mobile robot in friendly environments. Information Fusion 37, 22–36. doi:10.1016/j.inffus. 2017.01.001

  4. [4]

    Fail-safe multi-modal localization frame- workusingheterogeneousmap-matchingsources

    Lee, S., Seo, S.W., 2022. Fail-safe multi-modal localization frame- workusingheterogeneousmap-matchingsources. IEEETransactions on Intelligent Transportation Systems 23, 4008–4020. doi:10.1109/ TITS.2020.3038441

  5. [5]

    Automaticmap update using dashcam videos

    Zhanabatyrova,A.,SouzaLeite,C.F.,Xiao,Y.,2023. Automaticmap update using dashcam videos. IEEE Internet of Things Journal 10, 11825–11843. doi:10.1109/JIOT.2023.3244693

  6. [6]

    A survey on monocular re-localization: From the perspective of scene map representation

    Miao, J., Jiang, K., Wen, T., Wang, Y., Jia, P., Wijaya, B., Zhao, X., Cheng, Q., Xiao, Z., Huang, J., Zhong, Z., Yang, D., 2024. A survey on monocular re-localization: From the perspective of scene map representation. IEEE Transactions on Intelligent Vehicles , 1– 33doi:10.1109/TIV.2024.3378716

  7. [7]

    Brain-inspired multimodal hybrid neural network for robot place recognition

    Yu, F., Wu, Y., Ma, S., Xu, M., Li, H., Qu, H., Song, C., Wang, T., Zhao, R., Shi, L., 2023. Brain-inspired multimodal hybrid neural network for robot place recognition. Science Robotics 8, eabm6996. doi:10.1126/scirobotics.abm6996

  8. [8]

    Information Fusion 98, 101834

    Huang,Z.,Sun,S.,Zhao,J.,Mao,L.,2023.Multi-modalpolicyfusion for end-to-end autonomous driving. Information Fusion 98, 101834. doi:10.1016/j.inffus.2023.101834

Show all 81 references
  1. [9]

    Hcto: Optimality-awarelidarinertialodometrywithhybridcontinuoustime optimization for compact wearable mapping system

    Li,J.,Yuan,S.,Cao,M.,Nguyen,T.M.,Cao,K.,Xie,L.,2024. Hcto: Optimality-awarelidarinertialodometrywithhybridcontinuoustime optimization for compact wearable mapping system. ISPRS Journal of Photogrammetry and Remote Sensing 211, 228–243. doi:10.1016/ j.isprsjprs.2024.04.004

  2. [10]

    Attention- Enhanced Cross-modal Localization Between Spherical Images and Point Clouds

    Zhao, Z., Yu, H., Lyu, C., Yang, W., Scherer, S., 2023. Attention- Enhanced Cross-modal Localization Between Spherical Images and Point Clouds. IEEE Sensors Journal , 1–1doi: 10.1109/JSEN.2023. 3306377

  3. [12]

    Yin,P.,Xu,L.,Zhang,J.,Choset,H.,Scherer,S.,2021.i3dloc:Image- to-rangecross-domainlocalizationrobusttoinconsistentenvironmen- talconditions,in:ProceedingsofRobotics:ScienceandSystems(RSS ’21), Robotics: Science and Systems 2021

  4. [13]

    Zheng, S., Li, Y., Yu, Z., Yu, B., Cao, S.Y., Wang, M., Xu, J., Ai, R., Gu, W., Luo, L., Shen, H.L., 2023. I2p-rec: Recognizing images on large-scale point cloud maps through bird’s eye view projections, in: 2023 IEEE/RSJ International Conference on Intelligent Robots and Syst...

  5. [14]

    (lc)2: Lidar- camera loop constraints for cross-modal place recognition

    Lee, A.J., Song, S., Lim, H., Lee, W., Myung, H., 2023. (lc)2: Lidar- camera loop constraints for cross-modal place recognition. IEEE Robotics and Automation Letters , 1–8doi:10.1109/LRA.2023.3268848

  6. [15]

    Modalink: Unifying modalities for efficient image-to-pointcloud place recognition

    Xie, W., Luo, L., Ye, N., Ren, Y., Du, S., Wang, M., Xu, J., Ai, R., Gu, W., Chen, X., 2024. Modalink: Unifying modalities for efficient image-to-pointcloud place recognition. doi:10.48550/arXiv. 2403.18762, arXiv:2403.18762

  7. [16]

    Cattaneo,D.,Vaghi,M.,Fontana,S.,Ballardini,A.L.,Sorrenti,D.G.,

  8. [17]

    Vxp: Voxel-cross-pixel large-scale image-lidar place recognition

    Li, Y.J., Gladkova, M., Xia, Y., Wang, R., Cremers, D., 2024. Vxp: Voxel-cross-pixel large-scale image-lidar place recognition. doi:10. 48550/arXiv.2403.14594, arXiv:2403.14594

  9. [18]

    Allaboutvlad,in:2013IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Arandjelovic,R.,Zisserman,A.,2013. Allaboutvlad,in:2013IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 1578–1585. doi:10.1109/CVPR.2013.207

  10. [19]

    Netvlad:Cnnarchitectureforweaklysupervisedplacerecognition,in: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Arandjelovic, R., Gronat, P., Torii, A., Pajdla, T., Sivic, J., 2016. Netvlad:Cnnarchitectureforweaklysupervisedplacerecognition,in: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 5297–5307. doi:10.1109/CVPR.2016.572

  11. [20]

    Fine-tuning cnn image retrieval with no human annotation

    Radenovic, F., Tolias, G., Chum, O., 2019. Fine-tuning cnn image retrieval with no human annotation. IEEE Transactions on Pattern Analysis and Machine Intelligence 41, 1655–1668. doi: 10.1109/ TPAMI.2018.2846566

  12. [21]

    Hausler, S., Garg, S., Xu, M., Milford, M., Fischer, T., 2021. Patch- netvlad: Multi-scale fusion of locally-global descriptors for place recognition,in:2021IEEE/CVFConferenceonComputerVisionand PatternRecognition(CVPR),IEEE,Nashville,TN,USA.pp.14136– 14147. doi:10.1109/CVPR46...

  13. [23]

    IEEE Transactions on Circuits and Systems for Video Technology 33, 1109–1122

    Wang,Y.,Qiu,Y.,Cheng,P.,Zhang,J.,2023.Hybridcnn-transformer features for visual place recognition. IEEE Transactions on Circuits and Systems for Video Technology 33, 1109–1122. doi:10.1109/ TCSVT.2022.3212434

  14. [24]

    Imagepatch-matchingwithgraph-based learninginstreetscenes

    She, R., Kang, Q., Wang, S., Tay, W.P., Guan, Y.L., Navarro, D.N., Hartmannsgruber,A.,2023. Imagepatch-matchingwithgraph-based learninginstreetscenes. IEEETransactionsonImageProcessing32, 3465–3480. doi:10.1109/TIP.2023.3281171

  15. [26]

    Information fusion in content based image retrieval: A comprehensive overview

    Piras, L., Giacinto, G., 2017. Information fusion in content based image retrieval: A comprehensive overview. Information Fusion 37, 50–60. doi:10.1016/j.inffus.2017.01.003

  16. [27]

    19370–19380

    Zhu,S.,Yang,L.,Chen,C.,Shah,M.,Shen,X.,Wang,H.,.R2former: Unified retrieval and reranking transformer for place recognition, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 19370–19380. doi:10.1109/CVPR52729.2023. 01856. : Preprint su...

  17. [28]

    Placeformer: Transformer-based visual place recognition using multi-scale patch selection and fusion

    Kannan, S.S., Min, B.C., 2024. Placeformer: Transformer-based visual place recognition using multi-scale patch selection and fusion. IEEE Robotics and Automation Letters 9, 6552–6559. doi:10.1109/ LRA.2024.3408075

  18. [29]

    Global features are all you need for image retrieval and reranking, in: 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pp

    Shao, S., Chen, K., Karpur, A., Cui, Q., Araujo, A., Cao, B., 2023. Global features are all you need for image retrieval and reranking, in: 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pp. 11002–11012. doi:10.1109/ICCV51070.2023.01013

  19. [30]

    Optimal transport aggregation for visual place recognition, in: 2024 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pp

    Izquierdo, S., Civera, J., 2024. Optimal transport aggregation for visual place recognition, in: 2024 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pp. 17658–17668. doi:10.1109/CVPR52733.2024.01672

  20. [31]

    Towards seamless adaptation of pre-trained models for visual place recognition, in: The Twelfth International Conference on Learning Representations

    Lu, F., Zhang, L., Lan, X., Dong, S., Wang, Y., Yuan, C., 2024. Towards seamless adaptation of pre-trained models for visual place recognition, in: The Twelfth International Conference on Learning Representations

  21. [32]

    Anyloc: Towards universal visual place recognition

    Keetha,N.,Mishra,A.,Karhade,J.,Jatavallabhula,K.M.,Scherer,S., Krishna, M., Garg, S., 2024. Anyloc: Towards universal visual place recognition. IEEE Robotics and Automation Letters 9, 1286–1293. doi:10.1109/LRA.2023.3343602

  22. [33]

    Wang, H., Liu, Y., Wang, B., Sun, Y., Dong, Z., Wang, W., Yang, B., 2023. Freereg: Image-to-point cloud registration leveraging pretrained diffusion models and monocular depth estimators, in: The Twelfth International Conference on Learning Representations

  23. [34]

    Kim, G., Kim, A., 2018. Scan context: Egocentric spatial descriptor for place recognition within 3d point cloud map, in: 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, Madrid. pp. 4802–4809. doi:10.1109/IROS.2018.8593953

  24. [35]

    Pointnetvlad: Deep point cloud based retrieval for large-scale place recognition, in: 2018 IEEE/CVF Con- ferenceonComputerVisionandPatternRecognition,pp.4470–4479

    Uy, M.A., Lee, G.H., 2018. Pointnetvlad: Deep point cloud based retrieval for large-scale place recognition, in: 2018 IEEE/CVF Con- ferenceonComputerVisionandPatternRecognition,pp.4470–4479. doi:10.1109/CVPR.2018.00470

  25. [36]

    PointNet: DeepLearningonPointSetsfor3DClassificationandSegmentation, in:2017IEEEConferenceonComputerVisionandPatternRecogni- tion (CVPR), pp

    Charles, R.Q., Su, H., Kaichun, M., Guibas, L.J., 2017. PointNet: DeepLearningonPointSetsfor3DClassificationandSegmentation, in:2017IEEEConferenceonComputerVisionandPatternRecogni- tion (CVPR), pp. 77–85. doi:10.1109/CVPR.2017.16

  26. [37]

    Zhou, Z., Zhao, C., Adolfsson, D., Su, S., Gao, Y., Duckett, T., Sun, L., 2021. Ndt-transformer: Large-scale 3d point cloud localisation usingthenormaldistributiontransformrepresentation,in:2021IEEE International Conference on Robotics and Automation (ICRA), pp. 5654–5660. doi...

  27. [38]

    Lpd-net: 3d point cloud learning for large-scale place recognition and environment analysis, in: 2019 IEEE/CVF International Conference on Computer Vision (ICCV), pp

    Liu, Z., Zhou, S., Suo, C., Yin, P., Chen, W., Wang, H., Li, H., Liu, Y., 2019. Lpd-net: 3d point cloud learning for large-scale place recognition and environment analysis, in: 2019 IEEE/CVF International Conference on Computer Vision (ICCV), pp. 2831–

  28. [39]

    Zhang, W., Xiao, C., 2019. Pcan: 3d attention map learning us- ing contextual information for point cloud based retrieval, in: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, Long Beach, CA, USA. pp. 12428–12437. doi:10. 1109/CVPR.2019.01272

  29. [40]

    Hui, L., Yang, H., Cheng, M., Xie, J., Yang, J., 2021. Pyramid point cloud transformer for large-scale place recognition, in: 2021 IEEE/CVF International Conference on Computer Vision (ICCV), IEEE,Montreal,QC,Canada.pp.6078–6087.doi: 10.1109/ICCV48922. 2021.00604

  30. [41]

    HiTPR:HierarchicalTrans- former for Place Recognition in Point Cloud, in: 2022 International Conference on Robotics and Automation (ICRA), pp

    Hou,Z.,Yan,Y.,Xu,C.,Kong,H.,2022. HiTPR:HierarchicalTrans- former for Place Recognition in Point Cloud, in: 2022 International Conference on Robotics and Automation (ICRA), pp. 2612–2618. doi:10.1109/ICRA46639.2022.9811737

  31. [42]

    Spectral Geometric Verification: Re-Ranking Point Cloud Retrieval for Metric Localization

    Vidanapathirana, K., Moghadam, P.,Sridharan, S., Fookes, C., 2023. Spectral Geometric Verification: Re-Ranking Point Cloud Retrieval for Metric Localization. IEEE Robotics and Automation Letters 8, 2494–2501. doi:10.1109/LRA.2023.3255560

  32. [43]

    Rank- pointretrieval: Reranking point cloud retrieval via a visually consis- tent registration evaluation

    Zhang, W., Zhou, H., Dong, Z., Yan, Q., Xiao, C., 2023. Rank- pointretrieval: Reranking point cloud retrieval via a visually consis- tent registration evaluation. IEEE Transactions on Visualization and Computer Graphics 29, 3840–3854. doi:10.1109/TVCG.2022.3170695

  33. [44]

    Minkloc3d: Point cloud based large-scale place recognition, in: 2021 IEEE Winter Conference on Applica- tions of Computer Vision (WACV), pp

    Komorowski, J., 2021. Minkloc3d: Point cloud based large-scale place recognition, in: 2021 IEEE Winter Conference on Applica- tions of Computer Vision (WACV), pp. 1789–1798. doi:10.1109/ WACV48630.2021.00183

  34. [45]

    CVTNet: A Cross- View Transformer Network for LiDAR-Based Place Recognition in AutonomousDrivingEnvironments

    Ma, J., Xiong, G., Xu, J., Chen, X., 2023. CVTNet: A Cross- View Transformer Network for LiDAR-Based Place Recognition in AutonomousDrivingEnvironments. IEEETransactionsonIndustrial Informatics , 1–10doi:10.1109/TII.2023.3313635

  35. [46]

    Spherevlad++:Attention- based and signal-enhanced viewpoint invariant descriptor

    Zhao,S.,Yin,P.,Yi,G.,Scherer,S.,2023. Spherevlad++:Attention- based and signal-enhanced viewpoint invariant descriptor. IEEE Robotics and Automation Letters 8, 256–263. doi:10.1109/LRA.2022. 3223555

  36. [47]

    PIC-Net: Point cloud and imagecollaborationnetworkforlarge-scaleplacerecognition

    Lu, Y., Yang, F., Chen, F., Xie, D., 2020. PIC-Net: Point cloud and imagecollaborationnetworkforlarge-scaleplacerecognition. CoRR abs/2008.00658. arXiv:2008.00658

  37. [48]

    DeepResidualLearningfor Image Recognition, in: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp

    He,K.,Zhang,X.,Ren,S.,Sun,J.,2016. DeepResidualLearningfor Image Recognition, in: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778. doi:10.1109/CVPR. 2016.90, arXiv:1512.03385

  38. [49]

    Pan, Y., Xu, X., Li, W., Cui, Y., Wang, Y., Xiong, R., 2021. Coral: Colored structural representation for bi-modal place recognition, in: 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, Prague, Czech Republic. pp. 2084–2091. doi:10.1109/I...

  39. [50]

    AdaFusion:Visual-LiDARFusion With Adaptive Weights for Place Recognition

    Lai,H.,Yin,P.,Scherer,S.,2022. AdaFusion:Visual-LiDARFusion With Adaptive Weights for Place Recognition. IEEE Robotics and Automation Letters 7, 12038–12045. doi:10.1109/LRA.2022.3210880

  40. [51]

    MMDF:Multi- Modal Deep Feature Based Place Recognition of Mobile Robots With Applications on Cross-Scene Navigation

    Yu,X.,Zhou,B.,Chang,Z.,Qian,K.,Fang,F.,2022. MMDF:Multi- Modal Deep Feature Based Place Recognition of Mobile Robots With Applications on Cross-Scene Navigation. IEEE Robotics and Automation Letters 7, 6742–6749. doi:10.1109/LRA.2022.3176731

  41. [52]

    Mff-pr: Point cloud and image multi- modal feature fusion for place recognition, in: 2022 IEEE Interna- tional Symposium on Mixed and Augmented Reality (ISMAR), pp

    Liu, W., Fei, J., Zhu, Z., 2022. Mff-pr: Point cloud and image multi- modal feature fusion for place recognition, in: 2022 IEEE Interna- tional Symposium on Mixed and Augmented Reality (ISMAR), pp. 647–655. doi:10.1109/ISMAR55827.2022.00082

  42. [53]

    LCPR: A Multi-Scale Attention-Based LiDAR-Camera Fusion Network for Place Recogni- tion

    Zhou, Z., Xu, J., Xiong, G., Ma, J., 2023. LCPR: A Multi-Scale Attention-Based LiDAR-Camera Fusion Network for Place Recogni- tion. IEEE Robotics and Automation Letters 9, 1342–1349. doi:10. 1109/LRA.2023.3346753

  43. [54]

    Camera-lidar fusion with la- tentcorrelationforcross-sceneplacerecognition

    Pan, Y., Xie, J., Wu, J., Zhou, B., 2024. Camera-lidar fusion with la- tentcorrelationforcross-sceneplacerecognition. IEEETransactions on Industrial Electronics , 1–9doi:10.1109/TIE.2024.3440470

  44. [55]

    Cross fusion of point cloud and learned image for loop closure detection

    Yue, H., Cao, D., Liu, Z., Wang, T., Chen, W., 2024. Cross fusion of point cloud and learned image for loop closure detection. IEEE Robotics and Automation Letters 9, 2965–2972. doi:10.1109/LRA. 2024.3362681

  45. [56]

    Distilvpr: Cross-modal knowledge distillation for visual placerecognition

    Wang, S., She, R., Kang, Q., Jian, X., Zhao, K., Song, Y., Tay, W.P., 2024. Distilvpr: Cross-modal knowledge distillation for visual placerecognition. ProceedingsoftheAAAIConferenceonArtificial Intelligence 38, 10377–10385. doi:10.1609/aaai.v38i9.28905

  46. [57]

    Lever- aging BEV Representation for 360-degree Visual Place Recognition

    Xu, X., Jiao, Y., Lu, S., Ding, X., Xiong, R., Wang, Y., 2023. Lever- aging BEV Representation for 360-degree Visual Place Recognition. doi:10.48550/arXiv.2305.13814, arXiv:2305.13814

  47. [58]

    Learning Transferable Visual Models From Nat- ural Language Supervision, in: Proceedings of the 38th International Conference on Machine Learning, PMLR

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agar- wal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., Sutskever, I., 2021. Learning Transferable Visual Models From Nat- ural Language Supervision, in: Proceedings of the 38th International Confere...

  48. [59]

    Ermolov,A.,Mirvakhabova,L.,Khrulkov,V.,Sebe,N.,Oseledets,I.,

  49. [60]

    Hyper- bolic deep neural networks: A survey

    Peng, W., Varanka, T., Mostafa, A., Shi, H., Zhao, G., 2022. Hyper- bolic deep neural networks: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence 44, 10023–10044. doi:10.1109/ TPAMI.2021.3136921. : Preprint submitted to Elsevier Page 16 of 17

  50. [62]

    Attention is all you need, in: Proceedings of the 31st International Conference on Neural Infor- mation Processing Systems, Curran Associates Inc

    Vaswani,A.,Shazeer,N.,Parmar,N.,Uszkoreit,J.,Jones,L.,Gomez, A.N., Kaiser, Å., Polosukhin, I., . Attention is all you need, in: Proceedings of the 31st International Conference on Neural Infor- mation Processing Systems, Curran Associates Inc.. pp. 6000–6010. doi:10.5555/32952...

  51. [63]

    Self-supervised image denoising for real-world images with context-aware transformer

    Zhang, D., Zhou, F., 2023. Self-supervised image denoising for real-world images with context-aware transformer. IEEE Access 11, 14340–14349. doi:10.1109/ACCESS.2023.3243829

  52. [64]

    Highdynamic rangeimaging with context-aware transformer, in: 2023 International Joint Conference onNeuralNetworks(IJCNN),pp.1–8

    Zhou, F.,Fu, Z.,Zhang, D.,2023. Highdynamic rangeimaging with context-aware transformer, in: 2023 International Joint Conference onNeuralNetworks(IJCNN),pp.1–8. doi: 10.1109/IJCNN54540.2023. 10191491

  53. [65]

    Cross-modality attentive feature fusion for object detection in multispectral remote sensing imagery

    Fang, Q., Wang, Z., 2022. Cross-modality attentive feature fusion for object detection in multispectral remote sensing imagery. Pattern Recognition , 108786doi:10.1016/j.patcog.2022.108786

  54. [66]

    Hidformer: Hierarchical dual-tower transformer using multi-scale mergence for long-term time series forecasting

    Liu, Z., Cao, Y., Xu, H., Huang, Y., He, Q., Chen, X., Tang, X., Liu, X., 2024. Hidformer: Hierarchical dual-tower transformer using multi-scale mergence for long-term time series forecasting. Expert Systems with Applications 239, 122412. doi:10.1016/j.eswa.2023. 122412

  55. [68]

    Multimodal Earth observation data fusion: Graph-based approach in shared latent space

    Arun,P.,Sadeh,R.,Avneri,A.,Tubul,Y.,Camino,C.,Buddhiraju,K., Porwal, A., Lati, R., Zarco-Tejada, P., Peleg, Z., Herrmann, I., 2022. Multimodal Earth observation data fusion: Graph-based approach in shared latent space. Information Fusion 78, 20–39. doi:10.1016/j. inffus.2021.09.004

  56. [69]

    Facenet: A unified embedding for face recognition and clustering, in: 2015 IEEE Con- ference on Computer Vision and Pattern Recognition (CVPR), pp

    Schroff, F., Kalenichenko, D., Philbin, J., 2015. Facenet: A unified embedding for face recognition and clustering, in: 2015 IEEE Con- ference on Computer Vision and Pattern Recognition (CVPR), pp. 815–823. doi:10.1109/CVPR.2015.7298682

  57. [70]

    Kppr: Ex- ploiting momentum contrast for point cloud-based place recognition

    Wiesmann, L., Nunes, L., Behley, J., Stachniss, C., 2023. Kppr: Ex- ploiting momentum contrast for point cloud-based place recognition. IEEERoboticsandAutomationLetters8,592–599. doi: 10.1109/LRA. 2022.3228174

  58. [71]

    Deep visual geo-localization benchmark, in: 2022IEEE/CVFConferenceonComputerVisionandPatternRecog- nition (CVPR), pp

    Berton, G., Mereu, R., Trivigno, G., Masone, C., Csurka, G., Sattler, T., Caputo, B., 2022. Deep visual geo-localization benchmark, in: 2022IEEE/CVFConferenceonComputerVisionandPatternRecog- nition (CVPR), pp. 5386–5397. doi:10.1109/CVPR52688.2022.00532

  59. [72]

    Decoupled-and-coupled networks: Self-supervised hyperspectral im- age super-resolution with subpixel fusion

    Hong, D., Yao, J., Li, C., Meng, D., Yokoya, N., Chanussot, J., 2023. Decoupled-and-coupled networks: Self-supervised hyperspectral im- age super-resolution with subpixel fusion. IEEE Transactions on Geoscience and Remote Sensing 61, 1–12. doi:10.1109/TGRS.2023. 3324497

  60. [73]

    Lip-loc: Lidar image pretraining for cross-modal localization, in: ProceedingsoftheIEEE/CVFWinterConferenceonApplicationsof Computer Vision (WACV) Workshops, pp

    Shubodh,S.,Omama,M.,Zaidi,H.,Parihar,U.S.,Krishna,M.,2024. Lip-loc: Lidar image pretraining for cross-modal localization, in: ProceedingsoftheIEEE/CVFWinterConferenceonApplicationsof Computer Vision (WACV) Workshops, pp. 948–957

  61. [74]

    An image is worth 16x16 words: Transformers for image recognition at scale, in: International Conference on Learning Representations

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., Houlsby, N., 2023. An image is worth 16x16 words: Transformers for image recognition at scale, in: International Conferenc...

  62. [75]

    Representation learning with contrastive predictive coding

    van den Oord, A., Li, Y., Vinyals, O., 2019. Representation learning with contrastive predictive coding. doi:10.48550/arXiv.1807.03748, arXiv:1807.03748

  63. [76]

    Poincaréembeddingsforlearninghierar- chicalrepresentations,in:AdvancesinNeuralInformationProcessing Systems, Curran Associates, Inc

    Nickel,M.,Kiela,D.,2017. Poincaréembeddingsforlearninghierar- chicalrepresentations,in:AdvancesinNeuralInformationProcessing Systems, Curran Associates, Inc

  64. [77]

    Kitti-360: A novel dataset and benchmarks for urban scene understanding in 2d and 3d

    Liao, Y., Xie, J., Geiger, A., 2023. Kitti-360: A novel dataset and benchmarks for urban scene understanding in 2d and 3d. IEEE TransactionsonPatternAnalysisandMachineIntelligence45,3292–

  65. [78]

    Arewereadyforautonomous driving? the kitti vision benchmark suite, in: 2012 IEEE Conference onComputerVisionandPatternRecognition,pp.3354–3361

    Geiger,A.,Lenz,P.,Urtasun,R.,2012. Arewereadyforautonomous driving? the kitti vision benchmark suite, in: 2012 IEEE Conference onComputerVisionandPatternRecognition,pp.3354–3361. doi: 10. 1109/CVPR.2012.6248074

  66. [79]

    Mapillary street-level sequences: A dataset for lifelong place recognition, in: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Warburg,F.,Hauberg,S.,Lopez-Antequera,M.,Gargallo,P.,Kuang, Y., Civera, J., 2020. Mapillary street-level sequences: A dataset for lifelong place recognition, in: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2623–2632. doi:10.1109/CVPR42600.2020.00270

  67. [80]

    Evaluation of object proposals and convnet features for landmark-based visual place recognition

    Hou, Y., Zhang, H., Zhou, S., 2018. Evaluation of object proposals and convnet features for landmark-based visual place recognition. Journal of Intelligent & Robotic Systems 92, 505–520. doi:10.1007/ s10846-017-0735-y

  68. [81]

    Visualizing data using t-sne

    van der Maaten, L., Hinton, G., 2008. Visualizing data using t-sne. Journal of Machine Learning Research 9, 2579–2605

  69. [82]

    Flashatten- tion:Fastandmemory-efficientexactattentionwithio-awareness,in: Koyejo,S.,Mohamed,S.,Agarwal,A.,Belgrave,D.,Cho,K.,Oh,A

    Dao, T., Fu, D., Ermon, S., Rudra, A., Ré, C., 2022. Flashatten- tion:Fastandmemory-efficientexactattentionwithio-awareness,in: Koyejo,S.,Mohamed,S.,Agarwal,A.,Belgrave,D.,Cho,K.,Oh,A. (Eds.), Advances in Neural Information Processing Systems, Curran Associates, Inc.. pp. 1634...

  70. [441]

    doi:10.1016/j.isprsjprs.2023.09.018

  71. [2020]

    4365–4371

    Global visual localization in lidar-maps through shared 2d- 3d embedding space, in: 2020 IEEE International Conference on Robotics and Automation (ICRA), pp. 4365–4371. doi: 10.1109/ ICRA40945.2020.9196859

  72. [2840]

    doi:10.1109/ICCV.2019.00292

  73. [3310]

    doi:10.1109/TPAMI.2022.3179507

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.