Pith. sign in

REVIEW 4 major objections 6 minor 26 references

GeoUniPR: A Geometry-Consistent Unified Framework for Cross-Modal Place Recognition

T0 review · 4 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Projecting LiDAR into the camera view tops cross-modal place recall by aligning geometry before feature learning.

desk verdict Simple and credible CMPR framework with a spatially-aware contrastive loss, but the SOTA claim is undercut by uncontrolled comparisons in Table 1. read the letter →

arxiv 2608.11263 v1 pith:I3GR5BMT submitted 2026-08-10 cs.CV

Wonbong Kim , Jiatong Xiao , Rui Li , Xufei Wang , Qiwen Gu , Junqiao Zhao , Chen Ye , Guang Chen This is my paper · ORCID
classification cs.CV
keywords cross-modalplacerecognitionLiDAR-RGBretrievaldepthimageviewgeometry-consistentrepresentationcontrastivelearningparameter-efficientfine-tuningvisiontransformerfalsenegatives
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 argues that cross-modal place recognition — determining whether a camera image and a LiDAR scan come from the same location — can be solved by aligning the two sensors in the input representation before any feature learning happens. The authors project LiDAR points onto the camera image plane to build a dense, multi-channel depth view (metric depth, intensity, and surface-normal structure) that is pixel-aligned with the RGB image, and train two architecturally identical vision-transformer encoders with a contrastive loss that ignores spatially nearby false negatives. The result is reported top recall-at-1 on KITTI and KITTI-360 for both retrieval directions, with strong cross-dataset generalization, while fine-tuning far fewer parameters than full-backbone methods. If the claim holds, the auxiliary alignment modules and multi-stage training used by prior cross-modal systems become unnecessary engineering, and geometric consistency becomes the primary design principle for cross-modal descriptors.

What carries the argument

The load-bearing objects are the multi-channel Depth Image View (DIV) and the Spatially-Consistent InfoNCE (SC-InfoNCE) loss. DIV is a camera-perspective image built by projecting LiDAR points with the calibration matrices, keeping the closest point per pixel via z-buffering, densifying the sparse depth with a training-free completion step, and stacking metric depth, LiDAR intensity, and a normal-ratio structural descriptor as three channels, so the LiDAR input is pixel-aligned with the RGB frame. SC-InfoNCE modifies the standard contrastive denominator with a distance-dependent weight (either a hard mask or a smooth hybrid curve) so that samples within a small radius of the anchor are not penalized as negatives, which removes the false-hard-negative problem created by spatial continuity of trajectories. Together, DIV supplies the geometry-consistent representation and SC-InfoNCE supplies the consistent optimization; no auxiliary alignment module, staged training, or full backbone fine-tuning is used.

What would settle it

A decisive experiment is to take the trained model and compare GeoUniPR against a strong feature-level alignment baseline on the same inputs with the LiDAR-to-camera rotation perturbed by ±5 degrees at inference, exactly the setting of Table 6 in the paper. The central claim predicts that representation-level alignment should remain superior even under such miscalibration; if the baseline degrades less and overtakes GeoUniPR, then the reported gains come from perfect-calibration assumptions rather than from geometric consistency itself.

Watch

Extended reading notes

Core claim

The central discovery is that geometric consistency at the input level, rather than feature-level alignment after encoding, closes most of the RGB-to-LiDAR gap. By projecting the LiDAR point cloud through the camera intrinsics and extrinsics into a z-buffered, densified depth image view (DIV) that also carries intensity and a surface-normal-ratio channel, every pixel of the LiDAR input corresponds to a pixel in the RGB image. Two modality-specific ViT encoders with identical architectures then map RGB and DIV into a shared embedding space, and the Spatially-Consistent InfoNCE objective down-weights or removes nearby samples from the negative set, suppressing distance-induced false negatives. The paper reports that this recipe reaches recall-at-1 of 97.20% and 96.97% for 2D-to-3D and 3D-to-2D retrieval on KITTI, and 97.41% and 97.49% on KITTI-360, the best figures in its comparisons, while fine-tuning only a small fraction of the pretrained backbone's parameters.

Load-bearing premise

The framework assumes that the camera-to-LiDAR calibration used when constructing the depth image view remains accurate at inference time; if the extrinsic rotation or translation drifts, pixel-wise RGB-LiDAR correspondence breaks and retrieval accuracy drops sharply.

Editorial extensions

If this is right

  • A single representation and training recipe now serves all four retrieval settings — 2D-to-2D, 3D-to-3D, 2D-to-3D, and 3D-to-2D — with simultaneous gains in each.
  • Cross-modal place recognition can be trained with parameter-efficient fine-tuning, sharply cutting the number of trainable parameters relative to full-backbone methods while improving accuracy.
  • Camera-perspective depth views generalize across datasets better than spherical range views because they are not tied to the LiDAR sensor's spherical projection parameters.
  • The normal-ratio structural channel is the largest single contributor among the LiDAR-native cues, suggesting that surface-structure information, not just depth, carries place-discriminative signal.
  • SC-InfoNCE generalizes InfoNCE: with a constant weighting function it reduces to the standard objective, so the same loss family can be used whether or not spatial continuity is a concern.

Reading between the lines

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

  • The same distance-weighted negative handling should transfer to other spatially or temporally ordered retrieval problems, such as vehicle re-identification or video clip matching, where nearby samples are correlated rather than independent negatives.
  • Because the representation is built from calibration at input time, training with synthetic extrinsic perturbations is a natural robustness augmentation; the paper identifies calibration sensitivity as a limitation but does not test this remedy.
  • The 3D-to-2D direction is the deployment bottleneck: it requires online DIV construction (normal-ratio computation plus densification costs roughly 70 ms per scan), so efficiency work should target the LiDAR query path specifically.
  • Reported failures on near-duplicate places suggest the learned geometry cannot disambiguate appearance-identical structures; a lightweight re-ranking step using local patch correspondences would likely close this residual gap.
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 / 6 minor

Summary. GeoUniPR proposes a cross-modal place recognition framework that projects LiDAR point clouds into camera-perspective multi-channel depth image views (DIV) with depth, intensity, and normal-ratio channels, then learns a unified embedding space via two modality-specific ViT encoders with identical architectures using parameter-efficient adaptation. The paper introduces Spatially-Consistent InfoNCE (SC-InfoNCE), a contrastive objective that reweights or masks spatially adjacent negatives to mitigate false negatives under trajectory continuity. Evaluations on KITTI and KITTI-360 report state-of-the-art claims in both cross-modal directions, with internal ablations on DIV channels, aggregation heads, loss variants, and backbones, plus a controlled LiP-Loc comparison in Appendix C.

Significance. If the SOTA claim were established under a uniform protocol, the paper would make a useful contribution: the representation-first idea (reducing modality discrepancy before feature extraction via camera-perspective DIV) is coherent, the parameter-efficient training design is practical, and the controlled RIV-vs-DIV comparison in Appendix C isolates the view-parameterization effect in a way that is informative and well-executed. The internal ablations (Tables 2–5) are internally consistent and the training-dynamics/stability analyses in Appendix B add genuine value. However, the central SOTA claim is currently supported only by an uncontrolled comparison across methods with different thresholds, training protocols, and omitted baselines, and the test-set selection of hyperparameters and backbones further weakens the claim. The framework's dependence on accurate camera-LiDAR calibration is acknowledged in Section 6 and quantified in Appendix D.1, which is appropriate but should temper the strength of the headline results.

major comments (4)
  1. [§5.2, Table 1 and footnote] The SOTA comparison is not controlled because UniLoc's results are reported at a 20 m threshold while all other rows, including GeoUniPR, use a 10 m threshold. Since Recall@K is monotonically non-decreasing in the spatial threshold, UniLoc's numbers are not directly comparable, and the claim that GeoUniPR outperforms UniLoc on KITTI-360 is therefore unsupported as stated. The authors should re-evaluate UniLoc under the same 10 m protocol, or clearly separate the thresholds in the table and refrain from claiming superiority over UniLoc.
  2. [§5.2, Table 1 and §2.2] The most directly relevant DIV-based CMPR method, InsCMPR [8], is omitted from Table 1. Since GeoUniPR's core design is camera-perspective DIV, a comparison against InsCMPR under the same protocol is essential to support the SOTA claim. Additionally, Cross-PRNet is trained separately on each dataset while GeoUniPR trains on KITTI-360 and tests on KITTI, so the KITTI row for Cross-PRNet is not a cross-dataset comparison; the table mixes different training/evaluation protocols without a clear statement of what each row represents.
  3. [§5.2 and Appendix A] The VXP baseline is said to be re-evaluated under the same cross-dataset protocol as GeoUniPR, but the exact protocol used (e.g., which sequences, which training split, which injection of DIV vs. RIV) is not specified in the main text or appendix. This makes the reported VXP numbers unverifiable. The authors should provide the precise re-evaluation setup, ideally with code or a reproducible description.
  4. [§5.3, Tables 3 and 5] The SC-InfoNCE hyperparameters (Rn=20 for SC-Hard, p=2 for SC-Hybrid) and the backbone choice (DINOv3) are selected by comparing performance on the same evaluation sequences used to produce the headline results. This constitutes test-set selection and inflates the reported SOTA figures. The authors should either perform model selection on the validation sequences (e.g., KITTI-360 sequence 10) and report only the final test-set numbers, or disclose the selection process and discuss its impact on the claimed generalization.
minor comments (6)
  1. [§4.3, Eq. (4)] The piecewise definition of ω_hyb is typeset compactly and is difficult to parse; for example, the cases for d_ij=0, 0<d_ij≤R_e, R_e<d_ij<R_s, and d_ij≥R_s should be written consistently with clear commas and spacing, and the behavior at the exact thresholds R_e and R_s should be explicitly stated.
  2. [§5.1] The dataset names are used inconsistently: KITTI Odometry sequences are referred to as 'KITTI' throughout, which can confuse readers with KITTI-360. The text should explicitly state that cross-dataset evaluation uses KITTI Odometry sequences 00, 02, 07, and 08, and clarify that these are held out from training.
  3. [Table 1 caption] The table caption should state the spatial threshold for each row group rather than relying on a footnote, since the threshold mismatch is a central issue for interpreting the results.
  4. [Figure 2 caption] The labels for the negative pairs (e.g., INP1, I1P1, I1P2) are not explained in the caption or the text; adding a brief explanation of the notation would improve readability.
  5. [§5.3, Table 3] The choice of Rn=20 as the default for SC-Hard is not uniformly the best in Table 3: on KITTI 3D→2D, Rn=3 yields higher R@1 (95.10 vs. 94.75), and on KITTI-360 2D→3D, Rn=10 is slightly better. The selection rationale should be stated explicitly, for example by emphasizing stability across the sweep rather than peak performance.
  6. [§5.4] The t-SNE visualization in Figure 3 is on KITTI sequence 00, but the model was trained on KITTI-360; the caption should note that this is a cross-dataset qualitative example to avoid implying in-domain evaluation.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: DIV construction and SC-InfoNCE are defined independently of the benchmark metrics, and the reported accuracies are empirical measurements on external datasets.

full rationale

GeoUniPR's derivation chain is not circular. The DIV representation is constructed by explicit camera-LiDAR projection with K and T(v to c), z-buffering, and IP-Basic densification (Section 4.2), and SC-InfoNCE is defined as a distance-weighted generalization of InfoNCE (Section 4.3, Eqs. 2-4). These formulas do not contain the KITTI/KITTI-360 R@1 numbers as inputs, so no reported metric reduces to an equation by construction. Baselines (LiP-Loc, VXP, UniLoc, Cross-PRNet) and borrowed components (DINO, SALAD, IP-Basic, MultiConv adapters) come from independent prior work, and no load-bearing self-citation or uniqueness theorem is used to force the framework. Section 6 and Appendix D candidly state limitations (sensor generalization, online cost, calibration sensitivity), which are robustness caveats rather than circular moves. The skeptical concerns about Table 1 (mixed 10 m/20 m thresholds, omitted InsCMPR, re-evaluated VXP, and hyperparameter/backbone selection on evaluation sequences in Tables 3 and 5) are experimental-validity and test-set-selection issues that may weaken the SOTA claim, but they do not make any derivation equivalent to its own inputs. Under the hard rule requiring a quotable reduction, no circular step is present.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

These parameters are selected on the evaluation datasets themselves, so the reported gains of SC-InfoNCE are partly a product of benchmark-specific tuning rather than a parameter-free prediction.

free parameters (4)
  • Rn (SC-Hard ambiguous-region radius) = 20 m
    Default used in main experiments; chosen via sweep in Table 3 and stability analysis in Appendix B.
  • Re (SC-Hybrid exclusion radius) = 3 m
    Fixed from the empirical range suggested by the SC-Hard sweep (Section 5.3).
  • Rs (SC-Hybrid saturation radius) = 20 m
    Default with Re=3 (Section 5.3).
  • p (SC-Hybrid curvature exponent) = 2
    Selected via sweep in Table 3; higher p did not improve consistently.
assumptions (4)
  • domain assumption Camera and LiDAR are accurately calibrated and synchronized, so z-buffered projection of LiDAR into the camera plane yields true pixel-level correspondence with RGB.
    Assumed throughout Section 4.2; Section 6 states sensitivity to large extrinsic perturbations, making this premise load-bearing.
  • domain assumption IP-Basic densification fills the sparse DIV without introducing harmful artifacts that break geometric alignment.
    Applied in Section 4.2 and Appendix A; no evaluation of densification accuracy is provided.
  • domain assumption Spatially adjacent samples on the same trajectory are usually false negatives, so downweighting them improves contrastive learning.
    This is the core premise of SC-InfoNCE in Section 4.3; if nearby places are sometimes true matches, the weighting could suppress hard positives.
  • domain assumption Pre-trained DINO features are transferable to depth-like image inputs after light adaptation.
    Backbone choice in Section 4.1; the paper does not validate this transfer beyond the reported experiments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GeoUniPR: A Geometry-Consistent Unified Framework for Cross-Modal Place Recognition." pith.science (2026). https://pith.science/paper/I3GR5BMT

@misc{pith2026260811263,
  author       = {Pith},
  title        = {Pith review of: GeoUniPR: A Geometry-Consistent Unified Framework for Cross-Modal Place Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I3GR5BMT}},
  note         = {Machine review of arXiv:2608.11263}
}
read the original abstract

Cross-modal place recognition (CMPR) aims to identify the same location across heterogeneous sensing modalities, such as vision and LiDAR. Existing methods commonly bridge the modality gap using complex alignment modules, multi-stage training, or full fine-tuning of pretrained backbones. In this work, we revisit CMPR from the perspective of geometric consistency and propose GeoUniPR, a unified and concise geometry-consistent framework. GeoUniPR reduces cross-modal discrepancy at the representation level by projecting LiDAR point clouds into the camera perspective to construct Geometry-Consistent depth image views (DIV), which establish direct RGB-LiDAR correspondence. We further augment DIV with native LiDAR cues, including intensity and surface-normal information, yielding a multi-channel geometric representation that improves structural consistency. Based on this representation, GeoUniPR learns a unified embedding space using two modality-specific ViT-based encoders with identical architectures, trained through parameter-efficient adaptation without auxiliary alignment modules, multi-stage training, or full backbone fine-tuning. In addition, we introduce Spatially-Consistent InfoNCE (SC-InfoNCE), a CMPR-specific contrastive objective that suppresses distance-induced false negatives under spatial continuity. Extensive experiments on KITTI and KITTI-360 demonstrate that GeoUniPR achieves state-of-the-art (SOTA) performance in both same-modal and cross-modal place recognition, with strong cross-dataset generalization.

Figures

Figures reproduced from arXiv: 2608.11263 by the authors.

Figure 1
Figure 1. GeoUniPR achieves SOTA performance on KITTI and KITTI-360. Note that Cross-PRNet reports only 2D→3D results, and UniLoc reports re￾sults only on KITTI-360, so the remain￾ing entries are unavailable. These observations motivate a different perspective: in￾stead of compensating for modality discrepancy after fea￾ture extraction, CMPR should first establish geometry￾consistent representations before learning modality￾i… view at source ↗
Figure 2
Figure 2. Overview of GeoUniPR framework. LiDAR scans are projected and densified into a DIV [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. t-SNE visualization of 2D→3D embed￾dings on KITTI sequence 00 under different loss functions. Query and its positive match are high￾lighted in color, while negatives are shown in gray. t-SNE visualization. We qualitatively inspect the learned embedding geometry using t-SNE [26] on KITTI sequence 00 [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Validation loss curves across various combinations of image-view representations and [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: Cross-modal R@1 (mean±std over five runs) on KITTI and KITTI-360; SC-Hard yields lower variance, while SC-Hybrid attains higher mean accuracy under cross-dataset evaluation. Across both datasets and retrieval directions, SC-Hard consistently shows the smallest standard…
Figure 6
Figure 6. Figure 6: Cross-modal retrieval accuracy (R@1, %) of LiP-Loc under two [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: KITTI qualitative top-3 retrievals. Each pair of rows corresponds to one sequence frame: the first row is 2D→3D and the second row is 3D→2D. From top to bottom, the four blocks correspond to KITTI sequences 00, 02, 07, and 08. Each row shows the query (left) and the to…
Figure 8
Figure 8. Figure 8: KITTI-360 qualitative top-3 retrievals. Each pair of rows corresponds to one sequence frame: the first row is 2D→3D and the second row is 3D→2D. From top to bottom, the three blocks correspond to KITTI-360 sequences 0003, 0005, and 0009. Each row shows the query (left)…
Figure 9
Figure 9. Figure 9: KITTI qualitative top-3 retrievals with errors. Layout follows [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]
Figure 10
Figure 10. Figure 10: KITTI-360 qualitative top-3 retrievals with errors. Layout follows [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 20 canonical work pages

  1. [8]

    InsCMPR: Efficient cross-modal place recognition via instance-aware hybrid mamba-transformer

    Sheng Jiao, Zhenyu Su, Luyang Luo, Haoyu Yu, Zihan Zhou, Huimin Lu, and Xieyuanli Chen. InsCMPR: Efficient cross-modal place recognition via instance-aware hybrid mamba-transformer. InProceedings of the IEEE International Conference on Robotics and Automation (ICRA), pages 2212–2218. IEEE, 2025

    work page 2025

  2. [1]

    General place recognition survey: Towards real-world autonomy.IEEE Transactions on Robotics, 41:3019–3038, 2025

    Peng Yin, Jianhao Jiao, Shiqi Zhao, Lingyun Xu, Guoquan Huang, Howie Choset, Sebastian Scherer, and Jianda Han. General place recognition survey: Towards real-world autonomy.IEEE Transactions on Robotics, 41:3019–3038, 2025. doi: 10.1109/TRO.2025.3550771

    arXiv 2025

  3. [2]

    Reid, and John J

    César Cadena, Luca Carlone, Henry Carrillo, Yasir Latif, Davide Scaramuzza, José Neira, Ian D. Reid, and John J. Leonard. Past, present, and future of simultaneous localization and mapping: Toward the robust- perception age.IEEE Transactions on Robotics, 32(6):1309–1332, 2016. doi: 10.1109/TRO.2016.2624754

    arXiv 2016

  4. [3]

    Embedded-physics machine learning for coarse-graining and collective variable discovery without data

    Paul M. Newman, John J. Leonard, Juan D. Tardós, and José Neira. Explore and return: Experimental validation of real-time concurrent mapping and localization. InProceedings of the IEEE International Conference on Robotics and Automation (ICRA), pages 1802–1809. IEEE, 2002. doi: 10.1109/ROBOT. 2002.1014803

    work page Pith review arXiv 2002

  5. [4]

    OneShot global localization: Instant LiDAR-visual pose estimation

    Sebastian Ratz, Marcin Dymczyk, Roland Siegwart, and Renaud Dubé. OneShot global localization: Instant LiDAR-visual pose estimation. InProceedings of the IEEE International Conference on Robotics and Automation (ICRA), pages 5415–5421. IEEE, 2020. doi: 10.1109/ICRA40945.2020.9197458

    arXiv 2020

  6. [5]

    From coarse to fine: Robust hierarchical localization at large scale

    Paul-Edouard Sarlin, César Cadena, Roland Siegwart, and Marcin Dymczyk. From coarse to fine: Robust hierarchical localization at large scale. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 12716–12725, 2019

    work page 2019

  7. [6]

    Tiago Barros, Ricardo Pereira, Luís Garrote, Cristiano Premebida, and Urbano J. Nunes. Place recognition survey: An update on deep learning approaches.CoRR, abs/2106.10458, 2021. URL https://arxiv. org/abs/2106.10458

    arXiv 2021

  8. [7]

    Yifan Xia, Ziyuan Li, Y . J. Li, Lei Shi, H. Cao, Joao F. Henriques, and Daniel Cremers. UniLoc: Towards universal place recognition using any single modality.arXiv preprint, 2024

    work page 2024

Show all 26 references
  1. [9]

    A. J. Lee, Seunghyeon Song, Hyungtae Lim, Wooseok Lee, and Hyun Myung. LiDAR-camera loop constraints for cross-modal place recognition.IEEE Robotics and Automation Letters, 8(6):3589–3596, 2023

    2023

  2. [10]

    S. S. Puligilla, M. Omama, H. Zaidi, U. S. Parihar, and M. Krishna. LIP-Loc: LiDAR image pretraining for cross-modal localization. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision Workshops (WACVW), pages 939–948. IEEE, 2024

    2024

  3. [11]

    Y . J. Li, Maria Gladkova, Yifan Xia, Rui Wang, and Daniel Cremers. VXP: V oxel-cross-pixel large-scale camera-LiDAR place recognition. InProceedings of the International Conference on 3D Vision (3DV), pages 1233–1242. IEEE, 2025

    2025

  4. [12]

    Wang, et al

    Shen Zheng, Yiming Li, Zhihao Yu, Bowen Yu, Shuangyuan Cao, M. Wang, et al. I2P-Rec: Recognizing images on large-scale point cloud maps through bird’s eye view projections. InProceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 1...

    2023

  5. [13]

    Emerging properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Herve Jegou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging properties in self-supervised vision transformers. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 9650–9660, 2021

    2021

  6. [14]

    DINOv2: Learning robust visual features without supervision.arXiv preprint, 2023

    Maxime Oquab, Timothee Darcet, Theo Moutakanni, Huy V o, Marcin Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. DINOv2: Learning robust visual features without supervision.arXiv preprint, 2023

    2023

  7. [15]

    V o, Maximilian Seitzer, Federico Baldassarre, Maxime Oquab, Cesar Jose, et al

    Oriane Simeoni, Huy V . V o, Maximilian Seitzer, Federico Baldassarre, Maxime Oquab, Cesar Jose, et al. DINOv3.arXiv preprint, 2025

    2025

  8. [16]

    CricA VPR: Cross-image correlation-aware representation learning for visual place recognition

    Feng Lu, Xiaoyan Lan, Lei Zhang, Dong Jiang, Yao Wang, and Chun Yuan. CricA VPR: Cross-image correlation-aware representation learning for visual place recognition. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 16772–16782, 2024

    2024

  9. [17]

    SelaVPR++: Towards seamless adaptation of foundation models for efficient place recognition.arXiv preprint, 2025

    Feng Lu, Tian Jin, Xiaoyan Lan, Lei Zhang, Yuting Liu, Yao Wang, and Chun Yuan. SelaVPR++: Towards seamless adaptation of foundation models for efficient place recognition.arXiv preprint, 2025

    2025

  10. [18]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision. InProceedings of the...

    2021

  11. [19]

    Ballardini, and Domenico G

    Davide Cattaneo, Marco Vaghi, Simone Fontana, Alessandro L. Ballardini, and Domenico G. Sorrenti. Global visual localization in LiDAR-maps through shared 2D-3D embedding space. InProceedings of the IEEE International Conference on Robotics and Automation (ICRA), pages 4365–437...

    2020

  12. [20]

    S. Meng, Y . Wang, H. Xu, and Lap-Pui Chau. Contrastive learning-based place descriptor representation for cross-modality place recognition.Information Fusion, page 103351, 2025

    2025

  13. [21]

    Optimal transport aggregation for visual place recognition

    Sergio Izquierdo and Javier Civera. Optimal transport aggregation for visual place recognition. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 17658–17668, 2024

    2024

  14. [22]

    ImLPR: Image-based LiDAR place recognition using vision foundation models

    Minwoo Jung, Lanke Frank Tarimo Fu, Maurice Fallon, and Ayoung Kim. ImLPR: Image-based LiDAR place recognition using vision foundation models. InProceedings of the Conference on Robot Learning (CoRL), Seoul, September 2025

    2025

  15. [23]

    Waslander

    Jason Ku, Ali Harakeh, and Steven L. Waslander. In defense of classical image processing: Fast depth completion on the CPU. InProceedings of the 15th Conference on Computer and Robot Vision (CRV), pages 16–22. IEEE, 2018

    2018

  16. [24]

    KITTI-360: A novel dataset and benchmarks for urban scene understanding in 2D and 3D.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(3): 3292–3310, 2023

    Yiyi Liao, Jun Xie, and Andreas Geiger. KITTI-360: A novel dataset and benchmarks for urban scene understanding in 2D and 3D.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(3): 3292–3310, 2023. doi: 10.1109/TPAMI.2022.3179507

    2023

  17. [25]

    Are we ready for autonomous driving? the KITTI vision benchmark suite

    Andreas Geiger, Philip Lenz, and Raquel Urtasun. Are we ready for autonomous driving? the KITTI vision benchmark suite. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 3354–3361. IEEE, 2012

    2012

  18. [26]

    Visualizing data using t-SNE.Journal of Machine Learning Research, 9:2579–2605, 2008

    Laurens van der Maaten and Geoffrey Hinton. Visualizing data using t-SNE.Journal of Machine Learning Research, 9:2579–2605, 2008. 11 Appendix Overview.This appendix is organized as follows: • Appendix A details the implementation setup, including the network architecture, DIV ...

    2008

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.