Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

EmbodiedPlace: Learning Mixture-of-Features with Embodied Constraints for Visual Place Recognition

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A tiny re-ranking module lifts visual place recognition by mixing features drawn from GPS- or timestamp-selected neighbors, the paper claims.

desk verdict New idea—embodied-constraint neighbor selection for global-feature refinement—but the paper omits the equation that defines the learned weights, so the core result is unreproducible as submitted. read the letter →

arxiv 2506.13133 v1 pith:TJRXV67U submitted 2025-06-16 cs.CV

Bingxi Liu , Hao Chen , Shiyi Guo , Yihong Wu , Jinqiang Cui , Hong Zhang This is my paper · ORCID
classification cs.CV
keywords visualplacerecognitionloopclosuredetectionglobalfeaturere-rankingmixture-of-featuresembodiedconstraintsGPStagsself-similaritymatrixmetriclearning
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

EmbodiedPlace is a re-ranking method for visual place recognition that claims accuracy can be improved with almost no extra cost by blending the global features of neighboring database images, as long as those neighbors are chosen by embodied constraints rather than by visual similarity alone. The paper identifies four usable constraints in existing datasets—GPS tags, sequence timestamps, local feature-matching correspondences, and self-similarity matrices—and learns a small set of mixture weights with metric losses. On the standard Pittsburgh test set, the method raises recall at rank 1 from 91.8% to 92.7% over a DINOv2-GeM baseline while adding only 25 KB of parameters and about 10 microseconds per frame. The same plug-in also improves MSLS and Nordland results, and it transfers to a second feature extractor, BoQ. The broader point is that spatial-temporal metadata already present in place-recognition databases can replace both dedicated local features and motion-sequence conditioning as a cheap source of verification.

What carries the argument

The central mechanism is a Mixture-of-Features (MoF) layer: for each candidate feature $f_{c_i}$, the refined feature is a weighted sum over $L$ neighbors, $f_{c_i}' = \sum_{j=1}^{L} w_j f_{n_j}$, with the candidate itself as the first neighbor. The neighbors are selected by one of four embodied constraints—GPS tags, sequential timestamps, feature-matching correspondences, or a self-similarity matrix—instead of by approximate nearest-neighbor search. The weights come from a learnable matrix optimized with a combination of a direct refinement loss, which pulls refined positives toward the query and pushes negatives away, and an intra-class refinement loss, which pulls same-label refined candidates together and pushes different-label candidates apart. This lightweight matrix (25 KB) is what keeps the re-ranking overhead at about 10 microseconds per frame while adapting to whichever embodied constraint a dataset provides.

What would settle it

Run EmbodiedPlace on a VPR dataset whose GPS tags and timestamps have been randomly permuted so the embodied constraints carry no signal; if recall at rank 1 still improves over the baseline by roughly the same amount as with intact constraints, the paper's claim that embodied constraints drive the gains is wrong.

Watch

Extended reading notes

Core claim

EmbodiedPlace claims that visual place recognition can be improved without dedicated local features or motion-sequence conditioning by re-weighting mixtures of existing global features, where the mixture members are chosen by any available embodied constraint. Concretely, after a standard KNN retrieval, each candidate image is replaced with a weighted sum of itself and a small set of neighboring database images selected by GPS proximity, timestamp adjacency, feature-matching correspondences, or a self-similarity threshold. The weights are learned by a small matrix layer optimized with two metric losses: one pulling refined positives toward the query and pushing negatives away, and one pulling same-label refined candidates together. The paper reports that this plug-in, with 25 KB of extra parameters and roughly 10 microseconds per frame, raises R@1 by 0.9 points on Pitts-30k (91.8 to 92.7), by 1.0 point on MSLS-val, and by 1.5 points on Nordland over the DINOv2-GeM baseline, and that it also transfers to BoQ features with smaller gains. It further argues that unconstrained KNN-based reweighting (SuperGlobal) hurts VPR performance, so the embodied constraint is what makes global-feature mixing safe and effective.

Load-bearing premise

The benefit depends on the database carrying a usable embodied constraint—GPS tags, timestamps, match correspondences, or a meaningful self-similarity structure—so that neighbor selection is not just unconstrained nearest-neighbor search.

Editorial extensions

If this is right

  • Any existing VPR model can accept EmbodiedPlace as a plug-in module without retraining the feature extractor, so accuracy gains are additive to the underlying global retrieval system.
  • With 25 KB of parameters and a per-frame cost around 10 microseconds, the re-ranking step is cheap enough for real-time loop-closure detection on embedded or on-robot hardware, where local-feature verification at hundreds of milliseconds is impractical.
  • GPS tags, timestamps, feature-matching correspondences, and self-similarity matrices can all serve as the neighbor selector, so the same method applies to street-view, long-term, seasonal, and visual-localization datasets.
  • Strong embodied constraints such as GPS and sequence timestamps give larger gains, while weaker pseudo-embodied constraints such as self-similarity give positive but smaller gains, suggesting that systems should prefer any available spatial or temporal metadata over purely visual neighbor selection.
  • Because the paper reports that unconstrained KNN-based mixing (SuperGlobal) degrades recall by 1.3 to 6.6 points, a correct implication is that embodied constraints are not a minor detail but the load-bearing reason why global-feature mixing succeeds in VPR.

Reading between the lines

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

  • A natural extension: if exact GPS is absent but odometry, IMU, or a scene graph is available at inference, those constitute embodied constraints in the same sense, so the method could transfer to indoor, underwater, or GPS-denied environments where the database still carries some data association.
  • The self-similarity-matrix variant suggests a zero-metadata fallback: compute pairwise similarity among database features once offline and threshold it to define neighbors; the paper's results on MSLS show this weak constraint works, though with smaller gains than GPS.
  • One testable consequence: combining EmbodiedPlace's global re-ranking over the top 5–10 candidates with a local-feature geometric check on that reduced set could push accuracy higher than either alone, since global mixing fixes false negatives while local verification filters false positives.
  • The learned weights are a simple matrix layer, so a single shared weight model might be trained across multiple datasets with different embodied-constraint types by encoding the constraint type as an input; the paper does not explore this.
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 / 5 minor

Summary. The paper proposes EmbodiedPlace, a plug-and-play global-feature re-ranking method for visual place recognition (VPR). The idea is to refine each retrieved candidate feature by mixing it with features of neighboring database images, where the neighbors are selected using four forms of 'embodied constraints': GPS tags, temporal sequence timestamps, local-feature matching, and self-similarity matrices. The mixing weights are claimed to be learned via a multi-metric loss. Experiments on Pitts-30k/250k, MSLS, Nordland, and Aachen v1.1 show small but consistent improvements over a DINOv2-GeM baseline, with reported additional parameters of 25 KB and re-ranking latency of 0.01 ms. The paper also compares against QE, DBA, and SuperGlobal, and reports that SuperGlobal degrades performance on these datasets.

Significance. If the method is reproducible, it is a potentially useful and lightweight re-ranking module for VPR that leverages metadata already present in many benchmarks. The taxonomy of embodied constraints (GPS, timestamps, feature matching, self-similarity) is a helpful organizational contribution, and the comparative study of global-feature re-ranking baselines on modern VPR benchmarks is valuable. However, the central learning mechanism is presented incompletely: the mapping from the learnable parameter matrix to the mixture weights is never specified, and the reported parameter count is inconsistent with the stated matrix dimensions. These issues currently prevent verification of the core claim.

major comments (3)
  1. [§4.3, Eq. (6)] The paper defines the refined candidate feature as f'_ci = sum_j w_j f_nj but never states how the learnable matrix W produces the weights w_j. Section 4.3 only says that W is 'introduced as the weight model' and immediately defines losses that depend on f'_ci. Without a rule such as w_j = softmax(W^T [f_q; f_nj]) or an equivalent, the gradient path from W to w_j is absent, making the training procedure undefined and the results unreproducible. Please provide the exact computation of w_j from W and the inputs, including any normalization or temperature.
  2. [§4.3, parameter-count claim] The abstract and Section 5 claim 'only 25 KB of additional parameters', but Section 4.3 defines W in R^{N×D} with N the database size (as defined in Section 3) and D the feature dimension (768 for DINOv2-GeM). For a database of 10k images, this matrix would contain ~7.7M parameters (~30 MB in float32), contradicting the 25 KB claim. Please clarify what N denotes in W (e.g., number of classes, a hidden dimension, or a fixed small constant) and reconcile the parameter count.
  3. [§4.3, Eqs. (8)-(9)] The losses L_Direct and L_Intra are unbounded sums of Euclidean distances without margins, normalization, or weighting between the positive and negative terms. Minimizing Eq. (8) could push refined features arbitrarily far from the query for negatives (driving the loss to -∞), while the positive term alone could collapse all refined features to the query point. Please specify whether the weights w_j are normalized (e.g., via softmax), whether the distances are bounded, and how the training is stabilized; a simple triplet/hinge formulation with a margin or a normalized contrastive form would address this concern.
minor comments (5)
  1. [§5.2] The text says 'as summarized in Sec. 5.2' but the dataset overview is presented in Table 2; please correct the reference.
  2. [Table 4] The header of Table 4 appears to place 'Time(ms)' under the Nordland column; clarify that the time column applies to all reranking methods and is measured on MSLS-val, as stated in the caption.
  3. [Table 3] The column header 'Vene Feat.' appears to be a typo; it should likely read 'Venue' or 'Method'. Please fix.
  4. [§4.2] The rule for handling insufficient neighbors ('we duplicate f_n1 to maintain a consistent list length') is vague; specify whether duplication applies only when fewer than L neighbors exist and whether duplicates receive the same weight.
  5. [Supplementary §7.4] The ranking loss in Eq. (11) uses softmax on both distributions but the description says 'KL-divergence-based ranking loss'; please clarify whether the KL divergence is computed as KL(softmax(s) || softmax(r)) or in the reverse direction, and define how r is constructed from ground-truth rankings.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the learned MoF re-ranking is a standard supervised metric-learning pipeline whose core claim is evaluated on external benchmarks; the one self-citation and the self-similarity pseudo-constraint are not load-bearing.

full rationale

The paper's central derivation is a supervised learning pipeline: a learnable matrix W is optimized by metric losses (Eqs. 7-10) so that MoF-refined candidate features f'_ci = sum_j w_j f_nj (Eq. 6) become closer to the query for positive neighbors and farther for negatives, with the refined features then used to re-rank candidates. The performance claim (e.g., +0.9% R@1 on Pitts-30k over DINOv2-GeM, Tab. 3) is tested on held-out public benchmarks against GPS/sequence ground truth, so it is not a fitted input renamed as a prediction. The self-similarity matrix (Eq. 4) is computed from the same global features being refined, making it mildly self-referential as a 'pseudo-embodied constraint,' but it only selects neighbors; the learned weights and final ranking are still evaluated against external labels (GPS for Pitts, timestamps for Nordland, GPS-based MSLS protocol), so the claim does not reduce to its input by construction. The timestamp similarity in Eq. 2 cites the first author's prior ICRA paper [21], but this is a definitional borrowing, not a load-bearing uniqueness theorem or a forced alternative, so it does not constitute circularity. Separately, the paper never specifies how W produces the weights w_j (the W-to-w mapping is absent), and the stated 25 KB parameter count is hard to reconcile with W in R^{N x D}; these are reproducibility/consistency concerns, not circularity, and are noted here to keep the circularity verdict distinct from correctness risk.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the availability and validity of embodied constraints at inference, and on the as-yet-undefined learned weight model W. The listed thresholds and loss hyperparameters are chosen by hand or tuned but not reported, making the method underspecified.

free parameters (7)
  • K (re-ranking scope) = 5 or 10 (tuned; Tables 9-10)
    Number of candidate features re-ranked; tuned on validation, values 3-50 explored, best around 8-10.
  • L (number of neighbors) = 8 (tuned; Tables 9-10)
    Number of neighboring features mixed; values 2-16 explored, best around 8.
  • λ1, λ2 (loss weights) = not reported
    Weights of direct and intra-class refinement losses in Eq. (10); values omitted from the paper.
  • margin α (Eq. 7) = not reported
    Margin hyperparameter in the base metric loss; not specified.
  • threshold δ (self-similarity) = not reported
    Threshold in Eq. (5) for filtering self-similarity pairs; value not given.
  • threshold σ (feature matching) = not reported
    Inlier ratio threshold in Eq. (3); value not given.
  • time interval t, t_m = not reported
    Similarity and dissimilarity intervals for timestamp constraint in Eq. (2); values not given.
assumptions (4)
  • domain assumption Database images at inference are accompanied by a usable embodied constraint (GPS, timestamps, or matchable local features) for neighbor selection.
    Sec. 3 and 4.2; if absent, neighbor selection reduces to unconstrained KNN, which Table 4 shows can hurt performance.
  • domain assumption GPS tags in the benchmark databases are accurate enough that 25-meter proximity is the ground truth and a valid neighbor selector.
    Eq. (1) and the evaluation protocol in Sec. 5.2.
  • domain assumption The self-similarity matrix computed from the global features themselves is a valid pseudo-embodied constraint.
    Eq. (4)-(5) in Sec. 3; used for MSLS. This is feature-derived, not an external constraint.
  • domain assumption The learned weight matrix W transfers across the training distribution (MSLS/Pitts-30k) to test domains (Nordland, Aachen) without adaptation.
    Sec. 5.3-5.4; the same W is applied to all datasets; no per-dataset fine-tuning is described.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EmbodiedPlace: Learning Mixture-of-Features with Embodied Constraints for Visual Place Recognition." pith.science (2026). https://pith.science/paper/TJRXV67U

@misc{pith2026250613133,
  author       = {Pith},
  title        = {Pith review of: EmbodiedPlace: Learning Mixture-of-Features with Embodied Constraints for Visual Place Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TJRXV67U}},
  note         = {Machine review of arXiv:2506.13133}
}
read the original abstract

Visual Place Recognition (VPR) is a scene-oriented image retrieval problem in computer vision in which re-ranking based on local features is commonly employed to improve performance. In robotics, VPR is also referred to as Loop Closure Detection, which emphasizes spatial-temporal verification within a sequence. However, designing local features specifically for VPR is impractical, and relying on motion sequences imposes limitations. Inspired by these observations, we propose a novel, simple re-ranking method that refines global features through a Mixture-of-Features (MoF) approach under embodied constraints. First, we analyze the practical feasibility of embodied constraints in VPR and categorize them according to existing datasets, which include GPS tags, sequential timestamps, local feature matching, and self-similarity matrices. We then propose a learning-based MoF weight-computation approach, utilizing a multi-metric loss function. Experiments demonstrate that our method improves the state-of-the-art (SOTA) performance on public datasets with minimal additional computational overhead. For instance, with only 25 KB of additional parameters and a processing time of 10 microseconds per frame, our method achieves a 0.9\% improvement over a DINOv2-based baseline performance on the Pitts-30k test set.

Figures

Figures reproduced from arXiv: 2506.13133 by the authors.

Figure 1
Figure 1. Mixture-of-Features with Embodied Constraints. We illustrate an example of how EmbodiedPlace works. In the mid￾dle subfigure, the candidate rankings do not correspond correctly to the geographic locations. By mixing features that satisfy em￾bodied constraints, EmbodiedPlace refines the feature represen￾tation by reducing false-negative errors and converting them into true positives, while amplifying false-positive e… view at source ↗
Figure 2
Figure 2. Illustrative Diagram of the Methodology. This figure includes three components: the four forms of embodied constraints, the general VPR inference pipeline, and the inference pipeline of EmbodiedPlace. tamps can serve as a measure of image similarity. Consider￾ing nearly uniform motion as a simple example, we define a fixed similarity interval [−t, t] and a dissimilarity interval (−∞, −t−tm]∪[t+tm, ∞), where tm helps… view at source ↗
Figure 3
Figure 3. Losses for Feature Refinement. This figure illustrates two metric loss functions designed for feature refinement. The first aims to minimize the distance to the query, while the second focuses on minimizing the distance between positive samples of the same class. maximizing the distance for samples with a label of 0. The loss function is defined as: LDirect = X N i=1 I(yi = 1) · d(f ′ ci , fq)− I(yi = 0) · d(f ′ ci … view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Differences between SuperGlobal and Embodied￾Place. 7.5. Differences from SuperGlobal As illustrated in [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: The global feature similarity matrix of the Pittsburgh-30k database. It presents periodic similar blocks, with small blocks corresponding to the segmentation mode of panoramic camera images and large blocks corresponding to spa￾tial correlations. 7.6. Which Embodied Co…
Figure 6
Figure 6. Figure 6: Visualization results of EmbodiedPlace on the Pittsburgh dataset. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Visualization results of EmbodiedPlace on the Nordland dataset. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 8
Figure 8. Figure 8: Visualization results of EmbodiedPlace on the MSLS dataset. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]

Discussion (0). Continue with ORCID 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. SAGE: Spatial-visual Adaptive Graph Exploration for Efficient Visual Place Recognition

    cs.CV 2025-09 conditional novelty 5.0 of 10

    SAGE is a training pipeline that dynamically rebuilds a geo-visual graph and uses greedy clique sampling and soft local-feature weighting to reach state-of-the-art visual place recognition on eight benchmarks.

Reference graph

Works this paper leans on

48 extracted references · 38 canonical work pages · cited by 1 Pith paper

  1. [1]

    Mixvpr: Feature mixing for visual place recognition

    Amar Ali-bey, Brahim Chaib-draa, and Philippe Gigu `ere. Mixvpr: Feature mixing for visual place recognition. InPro- ceedings of the IEEE/CVF Winter Conference on Applica- tions of Computer Vision, pages 2998–3007, 2023. 2, 6, 12

    work page 2023

  2. [2]

    Boq: A place is worth a bag of learnable queries

    Amar Ali-bey, Brahim Chaib-draa, and Philippe Gigu `ere. Boq: A place is worth a bag of learnable queries. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17794–17803, 2024. 2, 6

    work page 2024

  3. [3]

    Fast and incremental loop closure detec- tion using proximity graphs

    Shan An, Guangfu Che, Fangru Zhou, Xianglong Liu, Xin Ma, and Yu Chen. Fast and incremental loop closure detec- tion using proximity graphs. In2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 378–385. IEEE, 2019. 3

    work page 2019

  4. [4]

    Fast and incremental method for loop-closure detection using bags of visual words.IEEE transactions on robotics, 24(5):1027–1037, 2008

    Adrien Angeli, David Filliat, St ´ephane Doncieux, and Jean- Arcady Meyer. Fast and incremental method for loop-closure detection using bags of visual words.IEEE transactions on robotics, 24(5):1027–1037, 2008. 1, 2

    work page 2008

  5. [5]

    Arandjelovic and A

    R. Arandjelovic and A. Zisserman. Three things everyone should know to improve object retrieval. In2012 IEEE Con- ference on Computer Vision and Pattern Recognition, 2012. 3, 7

    work page 2012

  6. [6]

    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. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 5297–5307, 2016. 1, 2, 3, 6

    work page 2016

  7. [7]

    Hpatches: A benchmark and evaluation of handcrafted and learned local descriptors

    Vassileios Balntas, Karel Lenc, Andrea Vedaldi, and Krys- tian Mikolajczyk. Hpatches: A benchmark and evaluation of handcrafted and learned local descriptors. InProceedings of the IEEE conference on computer vision and pattern recog- nition, pages 5173–5182, 2017. 2

    work page 2017

  8. [8]

    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? InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6154–6164, 2023. 3

    work page 2023

Show all 48 references
  1. [9]

    Speeded-up robust features (surf).Computer vi- sion and image understanding, 110(3):346–359, 2008

    Herbert Bay, Andreas Ess, Tinne Tuytelaars, and Luc Van Gool. Speeded-up robust features (surf).Computer vi- sion and image understanding, 110(3):346–359, 2008. 2

    2008

  2. [10]

    Deep visual geo-localization benchmark

    Gabriele Berton, Riccardo Mereu, Gabriele Trivigno, Carlo Masone, Gabriela Csurka, Torsten Sattler, and Barbara Ca- puto. Deep visual geo-localization benchmark. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5396–5407, 2022. 1, 6

    2022

  3. [11]

    Eigenplaces: Training viewpoint robust models for visual place recognition

    Gabriele Berton, Gabriele Trivigno, Barbara Caputo, and Carlo Masone. Eigenplaces: Training viewpoint robust models for visual place recognition. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 11080–11090, 2023. 6

    2023

  4. [12]

    Total recall: Automatic query expan- sion with a generative feature model for object retrieval

    Ondrej Chum, James Philbin, Josef Sivic, Michael Isard, and Andrew Zisserman. Total recall: Automatic query expan- sion with a generative feature model for object retrieval. In 2007 IEEE 11th International Conference on Computer Vi- sion, 2007. 3, 5, 7

    2007

  5. [13]

    Fisher vectors: Be- yond bag-of-visual-words image representations

    Gabriela Csurka and Florent Perronnin. Fisher vectors: Be- yond bag-of-visual-words image representations. InInterna- tional conference on computer vision, imaging and computer graphics, pages 28–42. Springer, 2010. 2

    2010

  6. [14]

    Superpoint: Self-supervised interest point detection and description

    Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabi- novich. Superpoint: Self-supervised interest point detection and description. InProceedings of the IEEE conference on computer vision and pattern recognition workshops, pages 224–236, 2018. 2

    2018

  7. [15]

    The faiss library

    Matthijs Douze, Alexandr Guzhva, Chengqi Deng, Jeff Johnson, Gergely Szilvasy, Pierre-Emmanuel Mazar´e, Maria Lomeli, Lucas Hosseini, and Herv´e J´egou. The faiss library. arXiv preprint arXiv:2401.08281, 2024. 4

    2024 arXiv

  8. [16]

    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. InComputer Vision– ECCV 2020: 16th European Conference, Glasgow, UK, Au- gust 23–28, 2020, Proceedings, Part IV 16, pages 369–38...

    2020

  9. [17]

    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. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14141–14152, 2021. 2, 3, 6, 7

    2021

  10. [18]

    Optimal transport ag- gregation for visual place recognition

    Sergio Izquierdo and Javier Civera. Optimal transport ag- gregation for visual place recognition. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024. 2, 6, 12

    2024

  11. [19]

    Aggregating local descriptors into a compact image representation

    Herv ´e J´egou, Matthijs Douze, Cordelia Schmid, and Patrick P´erez. Aggregating local descriptors into a compact image representation. In2010 IEEE computer society conference on computer vision and pattern recognition, pages 3304–

  12. [20]

    Correlation verification for image retrieval

    Seongwon Lee, Hongje Seong, Suhyeon Lee, and Euntai Kim. Correlation verification for image retrieval. InPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5374–5384, 2022. 3

    2022

  13. [21]

    A flexible and efficient loop closure detection based on motion knowledge

    Bingxi Liu, Fulin Tang, Yujie Fu, Yanqun Yang, and Yihong Wu. A flexible and efficient loop closure detection based on motion knowledge. In2021 IEEE International Conference on Robotics and Automation (ICRA), pages 11241–11247. IEEE, 2021. 2, 3, 4 9

    2021

  14. [22]

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

    David G Lowe. Distinctive image features from scale- invariant keypoints.International journal of computer vi- sion, 60:91–110, 2004. 2

    2004

  15. [23]

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

    Feng Lu, Xiangyuan Lan, Lijun Zhang, Dongmei Jiang, Yaowei Wang, and Chun Yuan. Cricavpr: Cross-image correlation-aware representation learning for visual place recognition. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 16772– 16782...

    2024

  16. [24]

    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. 6, 7

    2024 arXiv

  17. [25]

    1 year, 1000 km: The oxford robotcar dataset.The International Journal of Robotics Research, 36:3–15, 2017

    Will Maddern, Geoffrey Pascoe, Chris Linegar, and Paul Newman. 1 year, 1000 km: The oxford robotcar dataset.The International Journal of Robotics Research, 36:3–15, 2017. 1

    2017

  18. [26]

    Seqslam: Visual route-based navigation for sunny summer days and stormy winter nights

    Michael J Milford and Gordon F Wyeth. Seqslam: Visual route-based navigation for sunny summer days and stormy winter nights. In2012 IEEE international conference on robotics and automation, pages 1643–1649. IEEE, 2012. 2

    2012

  19. [27]

    Orb-slam: A versatile and accurate monocular slam system.IEEE transactions on robotics, 31(5):1147–1163,

    Raul Mur-Artal, Jose Maria Martinez Montiel, and Juan D Tardos. Orb-slam: A versatile and accurate monocular slam system.IEEE transactions on robotics, 31(5):1147–1163,

  20. [28]

    Dinov2: Learning robust visual features without super- vision.Transactions on Machine Learning Research, 2023

    Maxime Oquab, Timoth ´ee Darcet, Th´eo Moutakanni, Huy V V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel HAZIZA, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without super- vision.Transactions on Machine Learning Research, ...

    2023

  21. [29]

    Fine- tuning cnn image retrieval with no human annotation.IEEE transactions on pattern analysis and machine intelligence, 41(7):1655–1668, 2018

    Filip Radenovi ´c, Giorgos Tolias, and Ond ˇrej Chum. Fine- tuning cnn image retrieval with no human annotation.IEEE transactions on pattern analysis and machine intelligence, 41(7):1655–1668, 2018. 2, 6, 12

    2018

  22. [30]

    Scaling vision with sparse mix- ture of experts.Advances in Neural Information Processing Systems, 34:8583–8595, 2021

    Carlos Riquelme, Joan Puigcerver, Basil Mustafa, Maxim Neumann, Rodolphe Jenatton, Andr ´e Susano Pinto, Daniel Keysers, and Neil Houlsby. Scaling vision with sparse mix- ture of experts.Advances in Neural Information Processing Systems, 34:8583–8595, 2021. 3

    2021

  23. [31]

    Orb: An efficient alternative to sift or surf

    Ethan Rublee, Vincent Rabaud, Kurt Konolige, and Gary Bradski. Orb: An efficient alternative to sift or surf. In2011 International conference on computer vision, pages 2564–

  24. [32]

    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. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12716–12725, 2019. 2, 6

    2019

  25. [33]

    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. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4938–4947, 2020. 2

    2020

  26. [34]

    Benchmarking 6dof outdoor visual localization in changing conditions

    Torsten Sattler, Will Maddern, Carl Toft, Akihiko Torii, Lars Hammarstrand, Erik Stenborg, Daniel Safari, Masatoshi Okutomi, Marc Pollefeys, Josef Sivic, et al. Benchmarking 6dof outdoor visual localization in changing conditions. In Proceedings of the IEEE conference on compu...

    2018

  27. [35]

    Global features are all you need for image retrieval and reranking

    Shihao Shao, Kaifeng Chen, Arjun Karpur, Qinghua Cui, Andr´e Araujo, and Bingyi Cao. Global features are all you need for image retrieval and reranking. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 11036–11046, 2023. 3, 5, 7

    2023

  28. [36]

    Are we there yet? challenging seqslam on a 3000 km journey across all four seasons

    Niko S ¨underhauf, Peer Neubert, and Peter Protzel. Are we there yet? challenging seqslam on a 3000 km journey across all four seasons. InProc. of workshop on long-term auton- omy, IEEE international conference on robotics and automa- tion (ICRA), page 2013, 2013. 6

    2013

  29. [37]

    Long-term visual localization revisited.IEEE Transactions on Pattern Analy- sis and Machine Intelligence, 44:2074–2088, 2020

    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, 44:2074–2088, 2020. 1

    2020

  30. [38]

    Ef- fovpr: Effective foundation model utilization for visual place recognition.arXiv preprint arXiv:2405.18065, 2024

    Issar Tzachor, Boaz Lerner, Matan Levy, Michael Green, Tal Berkovitz Shalev, Gavriel Habib, Dvir Samuel, Noam Korngut Zailer, Or Shimshi, Nir Darshan, et al. Ef- fovpr: Effective foundation model utilization for visual place recognition.arXiv preprint arXiv:2405.18065, 2024. 3, 7

    2024

  31. [39]

    Attention is all you need.Advances in neural information processing systems, 30, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017. 5

    2017

  32. [40]

    Cosface: Large margin cosine loss for deep face recognition

    Hao Wang, Yitong Wang, Zheng Zhou, Xing Ji, Dihong Gong, Jingchao Zhou, Zhifeng Li, and Wei Liu. Cosface: Large margin cosine loss for deep face recognition. InPro- ceedings of the IEEE conference on computer vision and pat- tern recognition, pages 5265–5274, 2018. 1, 6

    2018

  33. [41]

    Transvpr: Transformer-based place recognition with multi-level attention aggregation

    Ruotong Wang, Yanqing Shen, Weiliang Zuo, Sanping Zhou, and Nanning Zheng. Transvpr: Transformer-based place recognition with multi-level attention aggregation. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13648–13657, 2022. 3, 6, 7

    2022

  34. [42]

    Multi-similarity loss with general pair weighting for deep metric learning

    Xun Wang, Xintong Han, Weilin Huang, Dengke Dong, and Matthew R Scott. Multi-similarity loss with general pair weighting for deep metric learning. InProceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion, pages 5022–5030, 2019. 1, 2, 12

    2019

  35. [43]

    Mapillary street-level sequences: A dataset for lifelong place recognition

    Frederik Warburg, Soren Hauberg, Manuel Lopez- Antequera, Pau Gargallo, Yubin Kuang, and Javier Civera. Mapillary street-level sequences: A dataset for lifelong place recognition. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2626–2...

    2020

  36. [44]

    Asymmetric feature fusion for image retrieval

    Hui Wu, Min Wang, Wengang Zhou, Zhenbo Lu, and Houqiang Li. Asymmetric feature fusion for image retrieval. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11082–11092, 2023. 3

    2023

  37. [45]

    mixup: Beyond empirical risk minimiza- tion.arXiv preprint arXiv:1710.09412, 2017

    Hongyi Zhang, Moustapha Cisse, Yann N Dauphin, and David Lopez-Paz. mixup: Beyond empirical risk minimiza- tion.arXiv preprint arXiv:1710.09412, 2017. 3

    2017 arXiv

  38. [46]

    Etr: An efficient transformer for re- 10 ranking in visual place recognition

    Hao Zhang, Xin Chen, Heming Jing, Yingbin Zheng, Yuan Wu, and Cheng Jin. Etr: An efficient transformer for re- 10 ranking in visual place recognition. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 5665–5674, 2023. 3

    2023

  39. [47]

    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. InProceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 19370–19380, 2023. 3, 6, 7, 8 11

    2023

  40. [48]

    Supplementary Materials 7.1. Implementation Details of Our Baseline Inspired by the comparative experiments in SALAD [18], we fine-tune DINOv2-GeM [28, 29] on the GSV-Cities [1] dataset by using the MS-Loss function [42]. The key dif- ference is that we set the output dimensio...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.