Pith. sign in

REVIEW 3 major objections 5 minor 185 references

GeoMFD: Continual Drone-View Geo-Localization with Geometry-Aware Adapter and Margin-Field Distillation

T0 review · 3 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read One continually updated model outperforms separately trained drone-view geo-localizers, preserving cross-view embedding geometry as it adapts to new environments.

desk verdict A genuine continual-learning contribution whose headline claim of matching per-environment specialists is undercut by uncontrolled baseline backbones and an average that hides a large DenseUAV deficit. read the letter →

arxiv 2607.25788 v1 pith:WPK2NRHF submitted 2026-07-28 cs.CV

Zhongwei Chen , Hai-jun Rong , Tao Zhang , Xianfeng Nie , Xiangbao Zhang , Guoqi Li , Zhao-Xu Yang This is my paper · ORCID
classification cs.CV
keywords continuallearningdrone-viewgeo-localizationcross-viewretrievalcatastrophicforgettingmargin-fielddistillationgeometry-awareadaptercold-startbootstrappingembeddinggeometry
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

The paper introduces continual drone-view geo-localization (C-DVGL), where a single model must keep learning new environments without forgetting older ones. It proposes GeoMFD, which combines a cold-start phase that warms up the backbone before a small adapter is switched on, a geometry-aware adapter that corrects features on the unit sphere through reflection and tangent residuals, and margin-field distillation that preserves the relative margins between correct matches and hard negatives from previous stages. The authors report that one continuously updated GeoMFD model achieves average R@1/AP of 89.44/87.48 for Drone→Satellite and 79.40/77.14 for Satellite→Drone across five benchmarks, exceeding the best environment-specific model trained individually by 1.52/0.35 and 2.90/3.84 percentage points, while using far less storage. The central claim is that preserving cross-view embedding geometry—not just category or identity discrimination—is what makes continual retrieval work.

What carries the argument

The Geometry-Aware Adapter (Geo-Adapter) operates on ℓ2-normalized embeddings: it computes group-wise spectral statistics from backbone features, predicts two correction directions—a Householder reflection residual and a tangent residual on the unit hypersphere—and fuses them with a gating weight before re-normalizing. Margin-Field Distillation (MFD) then copies the old model as a frozen teacher and forces the student to preserve the margins between positive pairs and the top-K hard negatives selected by the teacher, weighted by confidence. Cold-Start Bootstrapping (CBS) first warms up the backbone with the adapter off, then activates it, so the adapter starts as a near-identity correction.

What would settle it

Retrain each individually trained baseline (e.g., DAC, CAMP, Sample4Geo) using the same backbone (DINOv3 ViT-B/16) and the same 384×384 input resolution that GeoMFD uses; if the average R@1/AP of those individually trained models then meets or exceeds GeoMFD's 89.44/87.48 (Drone→Satellite) and 79.40/77.14 (Satellite→Drone), the paper's central claim of superiority over environment-specific models would be overturned.

Watch

Extended reading notes

Core claim

GeoMFD establishes that a single model, updated sequentially across five diverse environments (infrared night, dense urban, multi-height, RGB-T, and large-scale multi-view), can match or beat models trained separately for each environment. The method works by freezing the previous model as a teacher and distilling only the similarity margins between positive pairs and their hard negatives, rather than the full affinity matrix; this protects the local retrieval boundary while leaving room to adapt. The adapter is geometry-aware: it starts as an identity mapping and applies bounded angular corrections on the normalized embedding sphere. On the five benchmarks, the single GeoMFD model reaches h

Load-bearing premise

The reported lead over individually trained methods assumes those baselines were run with backbones and input resolutions comparable to GeoMFD's; if a baseline used a weaker encoder or lower resolution, its numbers would understate that method's true capability.

Editorial extensions

If this is right

  • Drones can be deployed in new areas without saving a per-environment model; storage drops from roughly 730–2100 MB for separate models to a single 345.9 MB checkpoint.
  • A continuously updated model can retain performance on earlier environments while learning new ones, so geo-localization systems can be field-updated over time.
  • Margin-field distillation alone prevents geometry forgetting without storing old images or features, enabling exemplar-free continual retrieval.
  • The method is robust to the order in which environments arrive (average R@1 varies by less than 2.5% across five permutations), so operators do not need a favorable curriculum.
  • GeoMFD works across different backbone families (ConvNeXt and ViT), indicating the design is not tied to a specific encoder.

Reading between the lines

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

  • The margin-field idea may generalize beyond drone-view retrieval: any embedding-based continual learning task (person re-identification, place recognition, metric learning) could preserve decision boundaries by distilling positive-versus-hard-negative margins instead of full similarity matrices.
  • Because the paper frames forgetting as distortion of unit-sphere geometry, it suggests a geometric interpretation of catastrophic forgetting in retrieval: not loss of identity discrimination but shrinkage of the margin between matches and near-misses.
  • A testable extension: apply GeoMFD to a sequence of environments with increasing dissimilarity (e.g., city → desert → night thermal → snow) and measure whether margin-field preservation degrades gracefully; a stress test on an adversarial ordering would isolate the geometry assumption.
  • The cold-start bootstrapping (backbone warm-up then adapter activation) implies that adapting a strong pretrained backbone with a residual adapter works better than joint fine-tuning from scratch—worth checking whether CBS offers the same benefit in other continual metric-learning settings.
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 formalizes continual drone-view geo-localization (C-DVGL), where a single model is sequentially adapted to new environments while retaining the cross-view embedding geometry needed for retrieval. It proposes GeoMFD, combining a cold-start bootstrapping strategy (CBS), a geometry-aware adapter (Geo-Adapter), and margin-field distillation (MFD). CBS warms up the backbone before activating the adapter; Geo-Adapter performs gated reflection/tangent corrections on the unit hypersphere; MFD distills positive-versus-hard-negative margins from a frozen teacher. The paper reports that a single continually updated GeoMFD model outperforms individually trained DVGL baselines and existing continual retrieval methods on five datasets, while requiring only 345.9 MB of storage at deployment.

Significance. If the empirical claims hold, this is a useful contribution: C-DVGL is a practical setting, and the proposed geometry-aware adapter and margin-field distillation are principled and clearly explained. The supplementary geometric derivations (reflection norm preservation, tangent-space projection, first-order normalization expansion, invariance of margin matching to row-wise shifts) are correct and are a strength of the paper. However, the headline comparison to environment-specific DVGL methods is not yet controlled for backbone and input resolution, and several experimental choices (test-set hyperparameter selection, single runs) weaken the evidence. The method itself is plausible and the central idea is defensible, but the current experimental support does not fully establish the strongest claims in the abstract and Section 'Overall Comparison'.

major comments (3)
  1. [Table 1 and 'Overall Comparison'] The central claim that GeoMFD exceeds individually trained DVGL methods is not supported by a controlled comparison. GeoMFD uses a DINOv3 ViT-B/16 backbone at 384×384 (Implementation Details and supplementary 'Backbone Architecture'), while Table 1 does not state the backbones, input resolutions, or training budgets of the individual baselines (Sample4Geo, MEAN, MFRGN, CAMP, DAC, SURFNet). Table 5 shows that backbone choice strongly affects results even under the same method: sequential fine-tuning improves from 69.80% R@1 with ConvNeXt-S to 85.22% with ViT-L/16 for Drone→Satellite. If the individual rows in Table 1 are original-paper numbers obtained with weaker encoders, the reported gains of 1.52/2.90 R@1 points (D→S and S→D) could be entirely explained by the backbone/resolution upgrade. Please report exact backbone/resolution for every baseline, or, preferably, re-run all baselines
  2. [Figure 5 and Implementation Details] Hyperparameters are selected on the test set. Figure 5(a-c) reports average R@1/AP over the five datasets (which are test sets) for different K, λ1, and λ2, and the text chooses K=32, λ1=0.1, λ2=0.3 based on these curves. This is test-set model selection and biases the headline numbers in Tables 1–4. The authors should either fix hyperparameters a priori and present Figure 5 only as sensitivity, or perform selection on a held-out validation split (e.g., a subset of training locations or a separate environment order) and report the resulting numbers. This is particularly important because the margin-field distillation weight λ2 has a 3.6-point effect on Drone→Satellite R@1 (85.84% at λ2=0.1 vs 89.44% at λ2=0.3).
  3. [Tables 1–5] No variance or significance information is reported. All results appear to come from a single training run. Many per-dataset differences are in the few-point range, and retrieval metrics on small test sets (e.g., IR-VL328, C-RGBT) can be noisy. For example, Satellite→Drone DenseUAV R@1 is 17.68% for GeoMFD versus 20.45% for Sample4Geo, while the averaged S→D claim rests on a 2.90-point margin. Reporting mean±std over at least three seeds, or a paired significance test, would materially strengthen the claim that GeoMFD's advantages are not run-to-run noise.
minor comments (5)
  1. [Methodology, after Eq. (13)] There is a duplicated phrase: 'Fort >1, Fort >1, the contrastive loss...' Please remove the duplicate.
  2. [Table 4] Several numbers are run together, e.g., '91.3393.00' and '80.1277.85'. Ensure proper spacing/column alignment in the final version.
  3. [Figure 3] The caption does not define how 'margin' and 'overlap area' are computed. Please state the exact definitions (e.g., threshold used, whether distributions are over all gallery negatives) so the visualization is reproducible.
  4. [Supplementary, Storage Accounting] The paper correctly states that the frozen teacher is a transient training copy and is not counted in the 345.9 MB deployment storage. It would be helpful to also state the peak training-time memory (student + teacher + optimizer), since onboard continual adaptation may be memory-constrained during training, not only at inference.
  5. [Related Work] The distinction from HCL-Geo and the lifelong aerial VPR setting is made in the supplementary; a one-sentence version in the main text would help readers understand why these existing continual-localization settings do not already cover C-DVGL.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: GeoMFD's training objectives are defined on current-stage data and a frozen teacher, and the headline comparison is an empirical benchmark result rather than a derivation from fitted parameters.

full rationale

GeoMFD's claimed contributions (CBS, Geo-Adapter, MFD) are implemented as concrete losses: contrastive loss J(S) in Eq. (5), view-gap closure in Eq. (7), and margin-field distillation in Eq. (11). None of these losses defines the target metric (R@1/AP) in terms of itself, and no parameter is fitted to the reported benchmark numbers and then renamed as a prediction. The margin-field distillation explicitly matches student margins to the frozen teacher's margins; Fig. 3 then measures margin/overlap on test data, which is a direct diagnostic of the optimization objective rather than an independent confirmation, but this does not carry the paper's central claim, which rests on retrieval accuracy in Tables 1 and 6. The only self-citations (MEAN, Chen et al. 2026) are baselines or related-work context, not load-bearing theoretical support; no uniqueness theorem or prior ansatz is imported to force the design. The main experimental concern is that Table 1 compares GeoMFD (DINOv3 ViT-B/16 at 384x384) with published DVGL baselines whose backbones and resolutions are not stated in the paper, so the reported margins of improvement could partly reflect encoder strength. That is a benchmark-validity/confounding risk, not a circular-derivation issue. Hyperparameter choices (K=32, lambda1=0.1, lambda2=0.3) are selected with sensitivity analyses on the benchmark, which is a form of test-set selection, but it does not make the method's outputs equal to its inputs by construction. Overall, the derivation chain is self-contained and no circular step is identifiable.

Assumptions & free parameters 13 free parameters · 7 assumptions · 0 invented entities

The method introduces several hyperparameters that are either hand-chosen or tuned on the test data (Figure 5). The geometric claims rely on standard sphere geometry. The domain assumptions are the representativeness of the five benchmark datasets and the sufficiency of margin distillation for geometry preservation.

free parameters (13)
  • alpha_t (residual scale) = 0.03
    Learnable scalar controlling adapter correction strength; initialized to 0.03 and tuned via test-set sensitivity.
  • eta (view-gap compensation ratio) = 0.75
    Targets a gap of 0.25*d_b; chosen by hand.
  • lambda_1 (view-gap closure weight) = 0.1
    Selected via test-set sensitivity (Figure 5b).
  • lambda_2 (MFD weight) = 0.3
    Selected via test-set sensitivity (Figure 5c).
  • K (hard negatives) = 32
    Selected via test-set sensitivity (Figure 5a).
  • tau_m (margin temperature) = 1.0
    Set in implementation; not sensitivity-tested.
  • tau_d (teacher softmax temperature) = 4.0
    Set in implementation.
  • omega_min (confidence floor) = 0.1
    Set in implementation to avoid discarding supervision.
  • G (channel groups in spectral calibration) = 12
    Set in implementation.
  • bottleneck width = 256
    Set in implementation.
  • epochs per phase/stage = 5 (warm-up), 5 (adapter), 5 per later stage
    Chosen by hand; total budget allocated to ablations.
  • learning rates = backbone 1e-5, adapter 5e-5
    Set in implementation.
  • input resolution = 384x384
    Set in implementation.
assumptions (7)
  • standard math Householder reflection preserves the L2 norm of a unit vector
    Used in Eq. (2)/(20)-(21) to guarantee the reflection residual stays on the unit sphere.
  • standard math The tangent projection (I - f̄ f̄^T) q keeps the residual in the tangent space of the unit sphere
    Eq. (22)-(23) and first-order expansion Eq. (27) rely on this property.
  • standard math Softmax gating yields a convex combination of residuals
    Eq. (25) ensures both residual weights are non-negative and sum to one.
  • domain assumption The frozen teacher's affinity margins are a faithful target for preserving retrieval geometry
    MFD assumes matching margins to top-K negatives is sufficient; no old data are stored.
  • domain assumption The five benchmark datasets used in sequence form a representative continual DVGL environment set
    Experimental protocol; no held-out validation of this representativeness.
  • ad hoc to paper A two-phase cold-start (backbone warm-up then adapter activation) provides a stable initialization
    CBS is introduced as a heuristic; no theoretical guarantee is provided.
  • ad hoc to paper The view-gap closure target c_b = -η(f_d - f_s) with η=0.75 preserves useful discriminative variation
    Eq. (6) and supplementary interpretation; the value of η is a heuristic choice.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GeoMFD: Continual Drone-View Geo-Localization with Geometry-Aware Adapter and Margin-Field Distillation." pith.science (2026). https://pith.science/paper/WPK2NRHF

@misc{pith2026260725788,
  author       = {Pith},
  title        = {Pith review of: GeoMFD: Continual Drone-View Geo-Localization with Geometry-Aware Adapter and Margin-Field Distillation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WPK2NRHF}},
  note         = {Machine review of arXiv:2607.25788}
}
read the original abstract

Existing drone-view geo-localization (DVGL) methods are mainly developed under a static training paradigm, where models are optimized for fixed environments with all training data available in advance. However, this paradigm is difficult to extend to real-world deployment, where drones may encounter diverse environments and require multiple environment-specific models, resulting in additional storage and model-selection costs. Directly adapting a single model to new environments also risks distorting previously learned cross-view embedding geometry and causing forgetting. To address these challenges, we formalize the continual drone-view geo-localization (C-DVGL) setting and propose GeoMFD, a geometry-aware continual adaptation method for DVGL. GeoMFD combines a cold-start bootstrapping strategy (CBS), a geometry-aware adapter (Geo-Adapter), and margin-field distillation (MFD) to balance adaptation and cross-view geometry preservation. CBS initializes a stable embedding space, Geo-Adapter enables environment adaptation through controlled residual corrections, and MFD preserves similarity margins between positive pairs and hard negatives to alleviate cross-view geometry forgetting. Extensive experiments demonstrate that GeoMFD effectively mitigates forgetting and achieves competitive performance with environment-specific DVGL methods using a single continuously updated model.

Figures

Figures reproduced from arXiv: 2607.25788 by the authors.

Figure 1
Figure 1. Pipeline Overview. Our method contains three key components. (a) At the initial stage, cold-start bootstrapping [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. This figure complements Table 1 by comparing GeoMFD with state-of-the-art DVGL methods under sequential [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Visualization of cross-view similarity distributions for DAC-FT, DASK, and GeoMFD across [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Motivation of GeoMFD. (a) Conventional deployment may require multiple environment-specific models as the [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: Hyperparameter sensitivity of GeoMFD under the [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

185 extracted references · 7 linked inside Pith

  1. [1]

    Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education

    Clancey, William J. Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education. Proceedings of the Eighth International Joint Conference on Artificial Intelligence (IJCAI-83)

  2. [2]

    Classification Problem Solving

    Clancey, William J. Classification Problem Solving. Proceedings of the Fourth National Conference on Artificial Intelligence

  3. [3]

    , title =

    Robinson, Arthur L. , title =. 1980 , doi =. https://science.sciencemag.org/content/208/4447/1019.full.pdf , journal =

    1980

  4. [4]

    New Ways to Make Microcircuits Smaller---Duplicate Entry

    Robinson, Arthur L. New Ways to Make Microcircuits Smaller---Duplicate Entry. Science

  5. [5]

    Clancey and Glenn Rennels , abstract =

    Diane Warner Hasling and William J. Clancey and Glenn Rennels , abstract =. Strategic explanations for a diagnostic consultation system , journal =. 1984 , issn =. doi:https://doi.org/10.1016/S0020-7373(84)80003-6 , url =

    doi:10.1016/s0020-7373(84)80003-6 1984

  6. [6]

    and Rennels, Glenn R

    Hasling, Diane Warner and Clancey, William J. and Rennels, Glenn R. and Test, Thomas. Strategic Explanations in Consultation---Duplicate. The International Journal of Man-Machine Studies

  7. [7]

    Poligon: A System for Parallel Problem Solving

    Rice, James. Poligon: A System for Parallel Problem Solving

  8. [8]

    Transfer of Rule-Based Expertise through a Tutorial Dialogue

    Clancey, William J. Transfer of Rule-Based Expertise through a Tutorial Dialogue

Show all 185 references
  1. [9]

    The Engineering of Qualitative Models

    Clancey, William J. The Engineering of Qualitative Models

  2. [10]

    2023 , eprint=

    Attention Is All You Need , author=. 2023 , eprint=

    2023

  3. [11]

    Pluto: The 'Other' Red Planet

    NASA. Pluto: The 'Other' Red Planet

  4. [12]

    FirstName LastName , title =

  5. [13]

    FirstName Alpher , title =

  6. [14]

    Journal of Foo , volume = 13, number = 1, pages =

    FirstName Alpher and FirstName Fotheringham-Smythe , title =. Journal of Foo , volume = 13, number = 1, pages =

  7. [15]

    Journal of Foo , volume = 14, number = 1, pages =

    FirstName Alpher and FirstName Fotheringham-Smythe and FirstName Gamow , title =. Journal of Foo , volume = 14, number = 1, pages =

  8. [16]

    FirstName Alpher and FirstName Gamow , title =

  9. [17]

    2024 , publisher=

    Zhang, Xiaohan and Li, Xingyu and Sultani, Waqas and Chen, Chen and Wshah, Safwan , journal=. 2024 , publisher=

    2024

  10. [18]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages=

    Sample4geo: Hard negative sampling for cross-view geo-localisation , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages=

  11. [19]

    Advances in Neural Information Processing Systems , volume=

    Dark experience for general continual learning: a strong, simple baseline , author=. Advances in Neural Information Processing Systems , volume=

  12. [20]

    2023 , publisher=

    Shen, Tianrui and Wei, Yingmei and Kang, Lai and Wan, Shanshan and Yang, Yee-Hong , journal=. 2023 , publisher=

    2023

  13. [21]

    2024 , publisher=

    Zhao, Hu and Ren, Keyan and Yue, Tianyi and Zhang, Chun and Yuan, Shuai , journal=. 2024 , publisher=

    2024

  14. [22]

    2024 , publisher=

    Chen, Quan and Wang, Tingyu and Yang, Zihao and Li, Haoran and Lu, Rongfeng and Sun, Yaoqi and Zheng, Bolun and Yan, Chenggang , journal=. 2024 , publisher=

    2024

  15. [23]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    Multibranch joint representation learning based on information fusion strategy for cross-view geo-localization , author=. IEEE Transactions on Geoscience and Remote Sensing , volume=. 2024 , publisher=

    2024

  16. [24]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    Accurate object localization in remote sensing images based on convolutional neural networks , author=. IEEE Transactions on Geoscience and Remote Sensing , volume=. 2017 , publisher=

    2017

  17. [25]

    Investigating the role of image retrieval for visual localization , author=. Int. J. Comput. Vision , volume=. 2022 , publisher=

    2022

  18. [26]

    A survey on the convergence of edge computing and

    McEnroe, Patrick and Wang, Shen and Liyanage, Madhusanka , journal=. A survey on the convergence of edge computing and. 2022 , publisher=

    2022

  19. [27]

    IEEE Transactions on Image Processing , volume=

    Joint representation learning and keypoint detection for cross-view geo-localization , author=. IEEE Transactions on Image Processing , volume=. 2022 , publisher=

    2022

  20. [28]

    A practical cross-view image matching method between

    Ding, Lirong and Zhou, Ji and Meng, Lingxuan and Long, Zhiyong , journal=. A practical cross-view image matching method between. 2020 , publisher=

    2020

  21. [29]

    Each part matters:

    Wang, Tingyu and Zheng, Zhedong and Yan, Chenggang and Zhang, Jiyong and Sun, Yaoqi and Zheng, Bolun and Yang, Yi , journal=. Each part matters:. 2021 , publisher=

    2021

  22. [30]

    Beyond geo-localization:

    Hu, Wenmiao and Zhang, Yichen and Liang, Yuxuan and Yin, Yifang and Georgescu, Andrei and Tran, An and Kruppa, Hannes and Ng, See-Kiong and Zimmermann, Roger , booktitle=. Beyond geo-localization:

  23. [31]

    2022 , pages =

    Zhu, Sijie and Shah, Mubarak and Chen, Chen , booktitle =. 2022 , pages =

    2022

  24. [32]

    Yang, Hongji and Lu, Xiufan and Zhu, Yingying , journal=

  25. [33]

    Proceedings of the Conference on Artificial Intelligence , volume=

    Optimal feature transport for cross-view image geo-localization , author=. Proceedings of the Conference on Artificial Intelligence , volume=

  26. [34]

    IEEE Transactions on Circuits and Systems for Video Technology , year=

    Enhancing cross-view geo-localization with domain alignment and scene consistency , author=. IEEE Transactions on Circuits and Systems for Video Technology , year=

  27. [35]

    2021 , publisher=

    Tian, Xiaoyang and Shao, Jie and Ouyang, Deqiang and Shen, Heng Tao , journal=. 2021 , publisher=

    2021

  28. [36]

    A transformer-based feature segmentation and region alignment method for

    Dai, Ming and Hu, Jianhong and Zhuang, Jiedong and Zheng, Enhui , journal=. A transformer-based feature segmentation and region alignment method for. 2021 , publisher=

    2021

  29. [37]

    Sun, Bo and Liu, Ganchao and Yuan, Yuan , journal=. F3-. 2023 , publisher=

    2023

  30. [38]

    A convNet for the 2020s , author=. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , year=

  31. [39]

    2024 , volume=

    Du, Haolin and He, Jingfei and Zhao, Yuanqing , journal=. 2024 , volume=

    2024

  32. [40]

    2024 , volume=

    Wu, Qiong and Wan, Yi and Zheng, Zhi and Zhang, Yongjun and Wang, Guangshuai and Zhao, Zhenyang , journal=. 2024 , volume=

    2024

  33. [41]

    IEEE Transactions on Geoscience and Remote Sensing , year=

    Multilevel feedback joint representation learning network based on adaptive area elimination for cross-view geo-localization , author=. IEEE Transactions on Geoscience and Remote Sensing , year=

  34. [42]

    IEEE Transactions on Geoscience and Remote Sensing , year=

    Direction-guided multi-scale feature fusion network for geo-localization , author=. IEEE Transactions on Geoscience and Remote Sensing , year=

  35. [43]

    2021 , pages =

    Liu, Ze and Lin, Yutong and Cao, Yue and Hu, Han and Wei, Yixuan and Zhang, Zheng and Lin, Stephen and Guo, Baining , booktitle=. 2021 , pages =

    2021

  36. [44]

    Aligning geometric spatial layout in cross-view geo-localization via feature recombination , author=. Proc. AAAI Conf. Artif. Intell. (AAAI) , volume=

  37. [45]

    2022 , publisher=

    Tian, Xiaoyang and Shao, Jie and Ouyang, Deqiang and Zhu, Anjie and Chen, Feiyu , booktitle=. 2022 , publisher=

    2022

  38. [46]

    Accurate 3-

    Shi, Yujiao and Yu, Xin and Liu, Liu and Campbell, Dylan and Koniusz, Piotr and Li, Hongdong , journal=. Accurate 3-. 2022 , publisher=

    2022

  39. [47]

    arXiv preprint arXiv:1807.03748 , year=

    Representation learning with contrastive predictive coding , author=. arXiv preprint arXiv:1807.03748 , year=

    arXiv

  40. [48]

    Pattern Recognit

    Multiple-environment self-adaptive network for aerial-view geo-localization , author=. Pattern Recognit. , volume=. 2024 , publisher=

    2024

  41. [49]

    2023 , publisher=

    Zhu, Runzhe and Yin, Ling and Yang, Mingze and Wu, Fei and Yang, Yuncheng and Hu, Wenbo , journal=. 2023 , publisher=

    2023

  42. [50]

    2020 , booktitle=

    Zheng, Zhedong and Wei, Yunchao and Yang, Yi , title =. 2020 , booktitle=

    2020

  43. [51]

    Swin transformer v2:

    Liu, Ze and Hu, Han and Lin, Yutong and Yao, Zhuliang and Xie, Zhenda and Wei, Yixuan and Ning, Jia and Cao, Yue and Zhang, Zheng and Dong, Li and others , booktitle =. Swin transformer v2:. 2022 , pages =

    2022

  44. [52]

    Van der Maaten, Laurens and Hinton, Geoffrey , journal=

  45. [53]

    Lending orientation to neural networks for cross-view geo-localization , year=

    Liu, Liu and Li, Hongdong , booktitle=. Lending orientation to neural networks for cross-view geo-localization , year=

  46. [54]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    Geobridge: A semantic-anchored multi-view foundation model bridging images and text for geo-localization , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

  47. [55]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    Cross-modality image matching network With modality-invariant feature representation for airborne-ground thermal infrared and visible datasets , author=. IEEE Transactions on Geoscience and Remote Sensing , volume=. 2021 , publisher=

    2021

  48. [56]

    Multilevel Embedding and Alignment Network With Consistency and Invariance Learning for Cross-View Geo-Localization , year=

    Chen, Zhongwei and Yang, Zhao-Xu and Rong, Hai-Jun , journal=. Multilevel Embedding and Alignment Network With Consistency and Invariance Learning for Cross-View Geo-Localization , year=

  49. [57]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    SURFNet: A Surface-Aware UAV--Satellite Geolocation Framework via Feature Aggregation and Dual Positional Encoding , author=. IEEE Transactions on Geoscience and Remote Sensing , volume=. 2026 , publisher=

    2026

  50. [58]

    IEEE Transactions on Geoscience and Remote Sensing , year=

    Unsupervised multi-view uav image geo-localization via iterative rendering , author=. IEEE Transactions on Geoscience and Remote Sensing , year=

  51. [59]

    arXiv preprint arXiv:2604.01747 , year=

    Unifying UAV cross-view geo-localization via 3D geometric perception , author=. arXiv preprint arXiv:2604.01747 , year=

  52. [60]

    2026 , eprint=

    Satellite-Free Training for Drone-View Geo-Localization , author=. 2026 , eprint=

    2026

  53. [61]

    Vision-Based UAV Self-Positioning in Low-Altitude Urban Environments , year=

    Dai, Ming and Zheng, Enhui and Feng, Zhenhua and Qi, Lei and Zhuang, Jiedong and Yang, Wankou , journal=. Vision-Based UAV Self-Positioning in Low-Altitude Urban Environments , year=

  54. [62]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Video2bev: Transforming drone videos to bevs for video-based geo-localization , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  55. [63]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages=

    MMGeo: Multimodal Compositional Geo-Localization for UAVs , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages=

  56. [64]

    Without Paired Labeled Data: End-to-End Self-Supervised Learning for Drone-View Geo-Localization , year=

    Chen, Zhongwei and Yang, Zhao-Xu and Rong, Hai-Jun and Li, Guoqi , journal=. Without Paired Labeled Data: End-to-End Self-Supervised Learning for Drone-View Geo-Localization , year=

  57. [65]

    IEEE Transactions on Image Processing , volume=

    A real-time memory updating strategy for unsupervised person re-identification , author=. IEEE Transactions on Image Processing , volume=. 2023 , publisher=

    2023

  58. [66]

    Cluster contrast for unsupervised person re-identification , author=. Proc. Asian Conf. Comput. Vis. , pages=

  59. [67]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Unsupervised feature learning via non-parametric instance discrimination , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  60. [68]

    Neural Networks , pages=

    Source-free unsupervised domain adaptation: A survey , author=. Neural Networks , pages=. 2024 , publisher=

    2024

  61. [69]

    Unveiling the power of clip in unsupervised visible-infrared person re-identification , author=. Proc. ACM Int. Conf. Multimedia , pages=

  62. [70]

    Cross-Modality Hierarchical Clustering and Refinement for Unsupervised Visible-Infrared Person Re-Identification , year=

    Pang, Zhiqi and Wang, Chunyu and Zhao, Lingling and Liu, Yang and Sharma, Gaurav , booktitle=. Cross-Modality Hierarchical Clustering and Refinement for Unsupervised Visible-Infrared Person Re-Identification , year=

  63. [71]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

    Zheng, Yi and Tang, Shixiang and Teng, Guolong and Ge, Yixiao and Liu, Kaijian and Qin, Jing and Qi, Donglian and Chen, Dapeng , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

  64. [72]

    Cho, Yoonki and Kim, Woo Jae and Hong, Seunghoon and Yoon, Sung-Eui , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , year =

  65. [73]

    Rethinking Pseudo-Label-Based Unsupervised Person Re-ID with Hierarchical Prototype-based Graph , author=. Proc. ACM Int. Conf. Multimedia , pages=

  66. [74]

    Chen, Hao and Wang, Yaohui and Lagadec, Benoit and Dantcheva, Antitza and Bremond, Francois , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , year =

  67. [75]

    Learning Invariance From Generated Variance for Unsupervised Person Re-Identification , year=

    Chen, Hao and Wang, Yaohui and Lagadec, Benoit and Dantcheva, Antitza and Bremond, Francois , journal=. Learning Invariance From Generated Variance for Unsupervised Person Re-Identification , year=

  68. [76]

    2024 , issn =

    GAN-based data augmentation and pseudo-label refinement with holistic features for unsupervised domain adaptation person re-identification , journal =. 2024 , issn =

    2024

  69. [77]

    Generative adversarial nets , author=. Proc. Adv. Neural Inf. Process. Syst. , volume=

  70. [78]

    A density-based algorithm for discovering clusters in large spatial databases with noise , author=. Proc. ACM SIGKDD Int. Conf. Knowl. Discov. Data Min. , volume=

  71. [79]

    Some methods for classification and analysis of multivariate observations , author=. Proc. Fifth Berkeley Symp. Math. Statist. Prob. , volume=

  72. [80]

    Li, Guopeng and Qian, Ming and Xia, Gui-Song , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , year =

  73. [81]

    Augmented Dual-Contrastive Aggregation Learning for Unsupervised Visible-Infrared Person Re-Identification , author=. Proc. ACM Int. Conf. Multimedia , year=

  74. [82]

    Hu, Sixing and Feng, Mengdan and Nguyen, Rang M. H. and Lee, Gim Hee , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , pages =

  75. [83]

    Zhang, Xiaohan and Li, Xingyu and Sultani, Waqas and Zhou, Yi and Wshah, Safwan , booktitle=

  76. [84]

    , title =

    Zhu, Jun-Yan and Park, Taesung and Isola, Phillip and Efros, Alexei A. , title =. Proc. IEEE Int. Conf. Comput. Vis. , pages=

  77. [85]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Shallow-Deep Collaborative Learning for Unsupervised Visible-Infrared Person Re-Identification , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  78. [86]

    arXiv preprint arXiv:1911.05371 , year=

    Self-labelling via simultaneous clustering and representation learning , author=. arXiv preprint arXiv:1911.05371 , year=

    1911 arXiv

  79. [87]

    Proceedings of the Ninth International Conference on Learning Representations: ICLR , pages=

    Prototypical contrastive learning of unsupervised representations (2020) , author=. Proceedings of the Ninth International Conference on Learning Representations: ICLR , pages=

    2020

  80. [88]

    He, Kaiming and Fan, Haoqi and Wu, Yuxin and Xie, Saining and Girshick, Ross , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , year =

  81. [89]

    and Lin, Dahua , title =

    Wu, Zhirong and Xiong, Yuanjun and Yu, Stella X. and Lin, Dahua , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , pages=

  82. [90]

    Learning representations by maximizing mutual information across views , author=. Proc. Adv. Neural Inf. Process. Syst. , volume=

  83. [91]

    International conference on machine learning , pages=

    A simple framework for contrastive learning of visual representations , author=. International conference on machine learning , pages=. 2020 , organization=

    2020

  84. [92]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

    Yang, Bin and Chen, Jun and Ye, Mang , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

  85. [93]

    IEEE Trans

    Dual consistency-constrained learning for unsupervised visible-infrared person re-identification , author=. IEEE Trans. Inf. Forensics Secur. , volume=. 2023 , publisher=

    2023

  86. [94]

    European conference on computer vision , year=

    Feature Representation Learning for Unsupervised Cross-domain Image Retrieval , author=. European conference on computer vision , year=

  87. [95]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Imagenet: A large-scale hierarchical image database , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  88. [96]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Person transfer gan to bridge domain gap for person re-identification , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  89. [97]

    ACM Trans

    A discriminatively learned cnn embedding for person reidentification , author=. ACM Trans. Multimedia Comput. Commun. Appl. , volume=

  90. [98]

    Deep learning face representation by joint identification-verification , author=. Proc. Adv. Neural Inf. Process. Syst. , volume=

  91. [99]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Scalable person re-identification: A benchmark , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  92. [100]

    Wu, Zesen and Ye, Mang , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , year =

  93. [101]

    Cai, Sudong and Guo, Yulan and Khan, Salman and Hu, Jiwei and Wen, Gongjian , title =. Proc. IEEE Int. Conf. Comput. Vis. , year =

  94. [102]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

    Qian, Ming and Xiong, Jincheng and Xia, Gui-Song and Xue, Nan , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

  95. [103]

    Shi, Yujiao and Yu, Xin and Campbell, Dylan and Li, Hongdong , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , year =

  96. [104]

    2022 , booktitle =

    Hu, Wenmiao and Zhang, Yichen and Liang, Yuxuan and Yin, Yifang and Georgescu, Andrei and Tran, An and Kruppa, Hannes and Ng, See-Kiong and Zimmermann, Roger , title =. 2022 , booktitle =

    2022

  97. [105]

    Liu, Liu and Li, Hongdong , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , pages =

  98. [106]

    Optimal Feature Transport for Cross-View Image Geo-Localization , volume=. Proc. AAAI Conf. Artif. Intell. , author=. 2020 , pages=

    2020

  99. [107]

    arXiv preprint arXiv:2409.16925 , year =

    Game4Loc: A UAV Geo-Localization Benchmark from Game Data , author =. arXiv preprint arXiv:2409.16925 , year =

    arXiv

  100. [108]

    Baldrati, Alberto and Bertini, Marco and Uricchio, Tiberio and Del Bimbo, Alberto , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , year =

  101. [109]

    Learning Joint Visual Semantic Matching Embeddings for Language-Guided Retrieval

    Chen, Yanbei and Bazzani, Loris. Learning Joint Visual Semantic Matching Embeddings for Language-Guided Retrieval. Proc. Eur. Conf. Comput. Vis. (ECCV). 2020

    2020

  102. [110]

    Zhang, Yukang and Wang, Hanzi , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , year =

  103. [111]

    IEEE Transactions on Image Processing , volume=

    Translation, association and augmentation: Learning cross-modality re-identification from single-modality annotation , author=. IEEE Transactions on Image Processing , volume=. 2023 , publisher=

    2023

  104. [112]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

    Xu, Peng and Zhu, Xiatian , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

  105. [113]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

    Fang, Xingye and Yang, Yang and Fu, Ying , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

  106. [114]

    Unsupervised learning of visual features by contrasting cluster assignments , author=. Proc. Adv. Neural Inf. Process. Syst. , volume=

  107. [115]

    Caron, Mathilde and Bojanowski, Piotr and Joulin, Armand and Douze, Matthijs , title =. Proc. Eur. Conf. Comput. Vis. (ECCV) , pages =

  108. [116]

    Augmented dual-contrastive aggregation learning for unsupervised visible-infrared person re-identification , author=. Proc. ACM Int. Conf. Multimedia , pages=

  109. [117]

    Random erasing data augmentation , author=. Proc. AAAI Conf. Artif. Intell. (AAAI) , volume=

  110. [118]

    Liu, Shuai and Li, Xin and Lu, Huchuan and He, You , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , year =

  111. [119]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

    Li, Bowen and Huang, Ziyuan and Ye, Junjie and Li, Yiming and Scherer, Sebastian and Zhao, Hang and Fu, Changhong , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

  112. [120]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Bridging the domain gap for ground-to-aerial image matching , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  113. [121]

    Toker, Aysim and Zhou, Qunjie and Maximov, Maxim and Leal-Taixe, Laura , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , year =

  114. [122]

    Advances in Neural Information Processing Systems , pages=

    Spatial-aware feature aggregation for image based cross-view geo-localization , author=. Advances in Neural Information Processing Systems , pages=

  115. [123]

    Bridging Viewpoints in Cross-View Geo-Localization With Siamese Vision Transformer , year=

    Ahn, Woo-Jin and Park, So-Yeon and Pae, Dong-Sung and Choi, Hyun-Duck and Lim, Myo-Taeg , journal=. Bridging Viewpoints in Cross-View Geo-Localization With Siamese Vision Transformer , year=

  116. [124]

    TirSA: A Three Stage Approach for UAV-Satellite Cross-View Geo-Localization Based on Self-Supervised Feature Enhancement , year=

    Sun, Jian and Sun, Hao and Lei, Lin and Ji, Kefeng and Kuang, Gangyao , journal=. TirSA: A Three Stage Approach for UAV-Satellite Cross-View Geo-Localization Based on Self-Supervised Feature Enhancement , year=

  117. [125]

    ICE: Inter-Instance Contrastive Encoding for Unsupervised Person Re-Identification , booktitle=

    Chen, Hao and Lagadec, Benoit and Bremond, Fran. ICE: Inter-Instance Contrastive Encoding for Unsupervised Person Re-Identification , booktitle=. 2021 , pages =

    2021

  118. [126]

    Xuan, Shiyu and Zhang, Shiliang , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , year =

  119. [127]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    NetVLAD: CNN architecture for weakly supervised place recognition , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  120. [128]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Predicting ground-level scene layout from aerial imagery , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  121. [129]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Vigor: Cross-view image geo-localization beyond one-to-one retrieval , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  122. [130]

    Learning Cross-View Visual Geo-Localization Without Ground Truth , year=

    Li, Haoyuan and Xu, Chang and Yang, Wen and Yu, Huai and Xia, Gui-Song , journal=. Learning Cross-View Visual Geo-Localization Without Ground Truth , year=

  123. [131]

    arXiv preprint arXiv:2304.07193 , year=

    Dinov2: Learning robust visual features without supervision , author=. arXiv preprint arXiv:2304.07193 , year=

    arXiv

  124. [132]

    International Conference on Machine Learning , pages=

    Learning transferable visual models from natural language supervision , author=. International Conference on Machine Learning , pages=

  125. [133]

    arXiv preprint arXiv:2503.07520 , year=

    From Limited Labels to Open Domains: An Efficient Learning Paradigm for UAV-view Geo-Localization , author=. arXiv preprint arXiv:2503.07520 , year=

    arXiv

  126. [134]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Ice: Inter-instance contrastive encoding for unsupervised person re-identification , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  127. [135]

    European conference on computer vision , pages=

    Optimal transport for label-efficient visible-infrared person re-identification , author=. European conference on computer vision , pages=

  128. [136]

    Dual-Level Matching with Outlier Filtering for Unsupervised Visible-Infrared Person Re-Identification , year=

    Ye, Mang and Wu, Zesen and Du, Bo , journal=. Dual-Level Matching with Outlier Filtering for Unsupervised Visible-Infrared Person Re-Identification , year=

  129. [137]

    Proceedings of the 31st ACM International Conference on Multimedia , pages=

    Efficient bilateral cross-modality cluster matching for unsupervised visible-infrared person reid , author=. Proceedings of the 31st ACM International Conference on Multimedia , pages=

  130. [138]

    Global Relation-Aware Contrast Learning for Unsupervised Person Re-Identification , year=

    Zhang, Hongwei and Zhang, Guoqing and Chen, Yuhao and Zheng, Yuhui , journal=. Global Relation-Aware Contrast Learning for Unsupervised Person Re-Identification , year=

  131. [139]

    FLAG: A Framework With Explicit Learning Based on Appearance and Gait for Video-Based Clothes-Changing Person Re-Identification , year=

    Lu, Hengjie and Gao, Yilin and Xu, Shugong , journal=. FLAG: A Framework With Explicit Learning Based on Appearance and Gait for Video-Based Clothes-Changing Person Re-Identification , year=

  132. [140]

    Corruption-Invariant Person Re-Identification via Coarse-to-Fine Feature Alignment , year=

    Zhang, Xinyu and Zhang, Peng and Shan, Caifeng , journal=. Corruption-Invariant Person Re-Identification via Coarse-to-Fine Feature Alignment , year=

  133. [141]

    Disentangled Representations for Short-Term and Long-Term Person Re-Identification , year=

    Eom, Chanho and Lee, Wonkyung and Lee, Geon and Ham, Bumsub , journal=. Disentangled Representations for Short-Term and Long-Term Person Re-Identification , year=

  134. [142]

    Proceedings of the AAAI Conference on Artificial Intelligence (AAAI) , volume=

    From Coarse to Fine: A Matching and Alignment Framework for Unsupervised Cross-View Geo-Localization , author=. Proceedings of the AAAI Conference on Artificial Intelligence (AAAI) , volume=

  135. [143]

    Meta Pairwise Relationship Distillation for Unsupervised Person Re-Identification , year=

    Ji, Haoxuanye and Wang, Le and Zhou, Sanping and Tang, Wei and Zheng, Nanning and Hua, Gang , journal=. Meta Pairwise Relationship Distillation for Unsupervised Person Re-Identification , year=

  136. [144]

    Pixel and Feature Transfer Fusion for Unsupervised Cross-Dataset Person Reidentification , year=

    Yang, Yang and Wang, Guan’an and Tiwari, Prayag and Pandey, Hari Mohan and Lei, Zhen , journal=. Pixel and Feature Transfer Fusion for Unsupervised Cross-Dataset Person Reidentification , year=

  137. [145]

    Global–Local Multiple Granularity Learning for Cross-Modality Visible–Infrared Person Reidentification , year=

    Zhang, Liyan and Du, Guodong and Liu, Fan and Tu, Huawei and Shu, Xiangbo , journal=. Global–Local Multiple Granularity Learning for Cross-Modality Visible–Infrared Person Reidentification , year=

  138. [146]

    Unsupervised Domain Adaptation on Person Reidentification Via Dual-Level Asymmetric Mutual Learning , year=

    Wu, Qiong and Li, Jiahan and Dai, Pingyang and Ye, Qixiang and Cao, Liujuan and Wu, Yongjian and Ji, Rongrong , journal=. Unsupervised Domain Adaptation on Person Reidentification Via Dual-Level Asymmetric Mutual Learning , year=

  139. [147]

    MHSA-Net: Multihead Self-Attention Network for Occluded Person Re-Identification , year=

    Tan, Hongchen and Liu, Xiuping and Yin, Baocai and Li, Xin , journal=. MHSA-Net: Multihead Self-Attention Network for Occluded Person Re-Identification , year=

  140. [148]

    Structured Domain Adaptation With Online Relation Regularization for Unsupervised Person Re-ID , year=

    Ge, Yixiao and Zhu, Feng and Chen, Dapeng and Zhao, Rui and Wang, Xiaogang and Li, Hongsheng , journal=. Structured Domain Adaptation With Online Relation Regularization for Unsupervised Person Re-ID , year=

  141. [149]

    , author=

    hdbscan: Hierarchical density based clustering. , author=. J. Open Source Softw. , volume=

  142. [150]

    Hierarchical grouping to optimize an objective function , author=. J. Am. Stat. Assoc. , volume=. 1963 , publisher=

    1963

  143. [151]

    Proceedings of the National Academy of Sciences , volume=

    Overcoming catastrophic forgetting in neural networks , author=. Proceedings of the National Academy of Sciences , volume=. 2017 , publisher=

    2017

  144. [152]

    Rebuffi, Sylvestre-Alvise and Kolesnikov, Alexander and Sperl, Georg and Lampert, Christoph H , booktitle=

  145. [153]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    Learning without forgetting , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2017 , publisher=

    2017

  146. [154]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Expandable subspace ensemble for pre-trained model-based class-incremental learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  147. [155]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Inflora: Interference-free low-rank adaptation for continual learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  148. [156]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    CL-LoRA: Continual low-rank adaptation for rehearsal-free class-incremental learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  149. [157]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Learning to prompt for continual learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  150. [158]

    European Conference on Computer Vision , pages=

    Dualprompt: Complementary prompting for rehearsal-free continual learning , author=. European Conference on Computer Vision , pages=. 2022 , organization=

    2022

  151. [159]

    Advances in Neural Information Processing Systems , volume=

    Vector quantization prompting for continual learning , author=. Advances in Neural Information Processing Systems , volume=

  152. [160]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Knowledge memorization and rumination for pre-trained model-based class-incremental learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  153. [161]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Mind the gap: Preserving and compensating for the modality gap in clip-based continual learning , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  154. [162]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Critical Patch-Aware Sparse Prompting with Decoupled Training for Continual Learning on the Edge , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  155. [163]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Learning from Itself: Mining Internal Knowledge from Vision Language Models for Continual Learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  156. [164]

    arXiv preprint arXiv:2508.10104 , year=

    Dinov3 , author=. arXiv preprint arXiv:2508.10104 , year=

    arXiv

  157. [165]

    Proceedings of the 32nd ACM International Conference on Multimedia (ACMMM) , pages=

    Mfrgn: Multi-scale feature representation generalization network for ground-to-aerial geo-localization , author=. Proceedings of the 32nd ACM International Conference on Multimedia (ACMMM) , pages=

  158. [166]

    IEEE Transactions on Geoscience and Remote Sensing , year=

    Cdm-net: A framework for cross-view geo-localization with multimodal data , author=. IEEE Transactions on Geoscience and Remote Sensing , year=

  159. [167]

    arXiv preprint arXiv:2511.02489 , year=

    Object Detection as an Optional Basis: A Graph Matching Network for Cross-View UAV Localization , author=. arXiv preprint arXiv:2511.02489 , year=

  160. [168]

    2025 IEEE International Conference on Robotics and Automation (ICRA) , pages=

    APA-BI: Adaptive Partition Aggregation and Bidirectional Integration for UAV-View Geo-Localization , author=. 2025 IEEE International Conference on Robotics and Automation (ICRA) , pages=. 2025 , organization=

    2025

  161. [169]

    SHAA: Spatial Hybrid Attention Network With Adaptive Cross-Entropy Loss Function for UAV-View Geo-Localization , year=

    Chen, Nanhua and Zhang, Dongshuo and Jiang, Kai and Yu, Meng and Zhu, Yeqing and Lou, Tai-Shan and Zhao, Liangyu , journal=. SHAA: Spatial Hybrid Attention Network With Adaptive Cross-Entropy Loss Function for UAV-View Geo-Localization , year=

  162. [170]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    Lifelong unsupervised domain adaptive person re-identification with coordinated anti-forgetting and adaptation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

  163. [171]

    Long Short-Term Knowledge Decomposition and Consolidation for Lifelong Person Re-Identification , year=

    Xu, Kunlun and Liu, Zichen and Zou, Xu and Peng, Yuxin and Zhou, Jiahuan , journal=. Long Short-Term Knowledge Decomposition and Consolidation for Lifelong Person Re-Identification , year=

  164. [172]

    Proceedings of the AAAI Conference on Artificial Intelligence (AAAI) , volume=

    Lstkc: Long short-term knowledge consolidation for lifelong person re-identification , author=. Proceedings of the AAAI Conference on Artificial Intelligence (AAAI) , volume=

  165. [173]

    Proceedings of the AAAI Conference on Artificial Intelligence (AAAI) , volume=

    Lifelong person re-identification via knowledge refreshing and consolidation , author=. Proceedings of the AAAI Conference on Artificial Intelligence (AAAI) , volume=

  166. [174]

    Proceedings of the AAAI Conference on Artificial Intelligence (AAAI) , volume=

    Dask: Distribution rehearsing via adaptive style kernel learning for exemplar-free lifelong person re-identification , author=. Proceedings of the AAAI Conference on Artificial Intelligence (AAAI) , volume=

  167. [175]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    Distribution-aware knowledge prototyping for non-exemplar lifelong person re-identification , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

  168. [176]

    Advances in Neural Information Processing Systems , volume=

    Beyond not-forgetting: Continual learning with backward knowledge transfer , author=. Advances in Neural Information Processing Systems , volume=

  169. [177]

    Proceedings of the European conference on computer vision (ECCV) , pages=

    Riemannian walk for incremental learning: Understanding forgetting and intransigence , author=. Proceedings of the European conference on computer vision (ECCV) , pages=

  170. [178]

    Advances in neural information processing systems , volume=

    Gradient episodic memory for continual learning , author=. Advances in neural information processing systems , volume=

  171. [179]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    A comprehensive survey of continual learning: Theory, method and application , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2024 , publisher=

    2024

  172. [180]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    A continual learning survey: Defying forgetting in classification tasks , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2021 , publisher=

    2021

  173. [181]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Cui, Zhenyu and Zhou, Jiahuan and Wang, Xun and Zhu, Manyu and Peng, Yuxin , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2024 , pages =

    2024

  174. [182]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    Prompt-Anchored Vision-Text Distillation for Lifelong Person Re-identification , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

  175. [183]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    Representation-Steered Incremental Adapter-Tuning for Class-Incremental Learning with Pre-Trained Models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

  176. [184]

    Proceedings of the AAAI Conference on Artificial Intelligence (AAAI) , volume=

    First Learn, Then Review: Human-Like Continual Learning for Cross-View Geo-Localization with Limited Field of View , author=. Proceedings of the AAAI Conference on Artificial Intelligence (AAAI) , volume=

  177. [185]

    arXiv preprint arXiv:2604.09038 , year=

    Towards Lifelong Aerial Autonomy: Geometric Memory Management for Continual Visual Place Recognition in Dynamic Environments , author=. arXiv preprint arXiv:2604.09038 , year=

    arXiv

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.