Pith. sign in

REVIEW 3 major objections 5 minor 37 references

VICI: VLM-Instructed Cross-view Image-localisation

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read Using a vision-language model to re-rank its top ten matches lifts street-to-satellite retrieval from 27.49% to 30.21% top-1 accuracy.

desk verdict A plausible two-stage CVGL recipe whose headline gains are undercut by internally inconsistent tables; the re-ranking gain is more solid than the drone augmentation gain. read the letter →

arxiv 2507.04107 v2 pith:XLP34ZDE submitted 2025-07-05 cs.CV

classification cs.CV
keywords Cross-viewgeo-localisationVision-languagemodelImageretrievalLimitedfield-of-viewDroneaugmentationRe-rankingUniversity-1652Street-to-satellitematching
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims a two-stage pipeline markedly improves limited-field-of-view street-to-satellite image retrieval on the University-1652 benchmark. Stage one uses a DINOv2-L embedding network, trained with a 30% chance of substituting a same-location drone image for the satellite reference, to retrieve a coarse top-10 ranking. Stage two feeds the query and those ten candidates to a general-purpose vision-language model, Gemini 2.5 Flash, which re-ranks them and writes a textual justification. The authors report 30.21% top-1 accuracy (R@1) and 63.13% top-10 accuracy (R@10), with the re-ranking step alone adding 2.72 percentage points of R@1. If the results hold, they show that a general-purpose VLM can act as an interpretable re-ranker for practical geo-localisation.

What carries the argument

The machinery is the two-stage VICI pipeline. Stage I is a Siamese embedding network with separate parameters for street and satellite branches, built on DINOv2-L and trained with the InfoNCE loss; during training, the satellite branch receives a randomly sampled same-location drone image with probability P=0.3, using the intermediate oblique viewpoint to bridge the ground-to-nadir domain gap. Stage II passes the query and the top-10 retrieved satellite candidates, together with a structured prompt, to Gemini 2.5 Flash, which re-ranks the candidates and justifies the top choice. The components that carry the argument are the drone augmentation probability, the fixed top-10 candidate bottleneck, and the VLM prompt.

What would settle it

Run the Stage II re-ranking with a text-only large language model given the same candidate summaries but no images; if R@1 still improves over the 27.49% embedding-only baseline, the gain is not coming from visual reasoning, and the claim about VLM vision would collapse.

Watch

Extended reading notes

Core claim

The paper's central discovery, on its own terms, is that a two-stage design—self-supervised DINOv2-L coarse retrieval plus a Gemini 2.5 Flash re-ranking of the top ten candidates—achieves 30.21% R@1 and 63.13% R@10 on the University-1652 limited-FOV street-to-satellite task, using no extra training data. The VLM re-ranking raises R@1 from 27.49% to 30.21% while R@10 stays fixed at 63.13%, so the improvement comes purely from reordering candidates already present in the coarse top-10. The same model also produces natural-language justifications that name matching visual features and an estimated camera position, which the authors present as evidence that VLM reasoning improves accuracy while adding interpretability.

Load-bearing premise

The load-bearing premise is that substituting same-location drone images for satellite images during 30% of training steps teaches the embedding model features that transfer to true satellite imagery at inference; if the model instead learns cues absent from real nadir satellite views, the augmentation's reported gain would not survive deployment.

Editorial extensions

If this is right

  • The 63.13% coarse R@10 sets a hard ceiling on this re-ranking scheme, so improving Stage-I recall is the clearest path to higher top-1 accuracy.
  • A general-purpose VLM can add 2.72 R@1 points at inference without retraining or expanding the reference database.
  • The pipeline returns natural-language justifications, so it can supply human-auditable explanations alongside location estimates.
  • Drone-view augmentation only helps in moderation: raising the substitution probability to 0.5 erases the gain, matching no-augmentation performance.

Reading between the lines

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

  • A natural extension the paper does not test is whether oblique satellite imagery, which is more widely available than drone video, can substitute for the drone views and produce a similar augmentation benefit.
  • Because re-ranking is restricted to the top-10 candidates, a perfect re-ranker cannot exceed 63.13% R@1; the largest headroom lies in raising coarse recall, not in better re-ranking.
  • If the VLM re-ranking gain transfers to other datasets, specialised re-ranking heads in retrieval pipelines may become less necessary, since a foundation model can reorder candidates and explain itself.
  • The temperature-0 JSON output makes the re-ranking stage reproducible, but the cost scales with 10 VLM calls per query; distilling the VLM's ranking behaviour into a small model could preserve the gain at lower cost.
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 VICI, a two-stage cross-view geo-localisation pipeline for limited-FOV street-to-satellite retrieval on the University-1652 challenge benchmark. Stage I trains a Siamese DINOv2-L model with InfoNCE loss and a drone-image augmentation that randomly substitutes drone views for satellite images during training; Stage II re-ranks the top-10 retrieval candidates with Gemini 2.5 Flash and produces textual justifications. The headline results are R@1 30.21% and R@10 63.13%, with claimed gains of +2.72 pp from VLM re-ranking and +2.83 pp from drone augmentation, using only the official challenge training data.

Significance. If the results are reproducible, VICI provides a practical two-stage recipe and evidence that a general-purpose VLM can act as an effective re-ranker for cross-view geo-localisation while also yielding interpretable justifications. The paper releases code, uses only the official training split, fixes VLM temperature and thinking budget for reproducibility, and compares multiple backbones under a stated common training setup. However, the evaluation protocol has internal inconsistencies in the baseline numbers and no uncertainty quantification, so the quantitative claims are not yet securely established.

major comments (3)
  1. [Section 4, Tables 1-4] The Stage I reference point for the re-ranking gain is not uniquely defined. Table 1 reports DINOv2-L as 27.49/51.96/63.13; Table 2 reports P=0.3 as 27.49/51.96/63.13 and P=0 as 24.47/48.16/60.99; Table 4 reports DINOv2-L as 24.66/48.00/59.02 and '+ Drone Data' as 27.49/51.96/63.13. Since Table 3 uses 27.49 as 'Without Re-ranking', the claimed +2.72 pp gain from Gemini 2.5 Flash depends on whether the underlying coarse model used drone augmentation, and the reader cannot uniquely resolve from the paper which configuration produced the 27.49 row. Please clarify the training condition for each table (especially whether Table 1 used P=0 or P=0.3) and provide a single consistent baseline row, or explain why identical numbers appear for different configurations.
  2. [Section 4, Tables 2-4] All results appear to be single runs with no error bars. The no-drone baseline appears as 24.47/48.16/60.99 in Table 2 and 24.66/48.00/59.02 in Table 4; if these are meant to be the same configuration, the discrepancy is an internal inconsistency, and if they are different configurations, the difference is not stated. Because the reported incremental gains are small (2.72-2.83 pp), the paper should report means and standard deviations over multiple seeds or otherwise justify that the differences exceed run-to-run variability. The paper should also state explicitly whether any hyperparameters (P, VLM variant, thinking budget) were selected using the test split; if so, the reported gains are optimistic and should be re-estimated on a held-out validation split.
  3. [Section 3.1 and Section 4] The evaluation protocol for the drone augmentation is underspecified. The paper trains with drone images substituted into the satellite branch but does not explicitly state whether inference uses only satellite images as references or also includes drone images. If the same-location drone images are available at test time, the reported +2.83 pp gain from drone augmentation may not transfer to a deployment setting where only satellite references exist. Please state clearly that the reference database at inference contains only satellite images and that drone images are used only during training, or report results for both settings.
minor comments (5)
  1. [Abstract and metadata] The abstract in the provided text still contains placeholder tokens '\topone\%' and '\topten\%' instead of the actual numbers 30.21% and 63.13%; these should be replaced before submission.
  2. [Figure 1 caption] The caption contains typos such as 'T op left' and 'T op right'; these should read 'Top left' and 'Top right'.
  3. [Section 2 and References] References [1] and [38] are the same workshop paper and should be merged or cited consistently throughout the text.
  4. [Section 4, Table 2] The text states that P=0.5 causes performance to 'drop significantly', but Table 2 shows R@5 at 52.03 for P=0.5 versus 51.96 for P=0.3; without error bars or a statistical test, 'significantly' is unsupported and should be softened or replaced with quantitative evidence.
  5. [Section 3.3] The implementation details for Stage I omit weight decay and warm-up settings for the AdamW optimiser; adding these, or pointing to the exact configuration in the released code, would improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: VICI reports measured retrieval accuracy on held-out test splits; no fitted parameter is renamed as a prediction.

full rationale

The paper is an empirical retrieval pipeline, not a derivation. Stage I trains a Siamese DINOv2 model with InfoNCE loss on the official train split, and Stage II re-ranks the top-10 retrievals using a frozen Gemini 2.5 VLM; all reported R@1/R@5/R@10 numbers are measurements on the official test set of University-1652. There is no equation in which a fitted quantity is algebraically identical to the reported prediction, and no parameter is fitted to a subset and then renamed as a prediction of the same subset. The selection of P=0.3 and the choice of Gemini Flash are made by comparing test-set metrics, which is a model-selection / benchmark-overfitting concern, not circularity. Self-citations to GeoDTR and related prior work are contextual references and are not load-bearing for the central VICI result; no uniqueness theorem or ansatz is imported from the authors' own prior work to force the method. The internal inconsistency in the reported baseline numbers (e.g., DINOv2-L appears as 27.49 in Tables 1/2/3 but as 24.66 in Table 4) is a reproducibility and attribution issue that a reader cannot resolve from the paper alone, but it does not make any claim equivalent to its own input by construction. Under the provided circularity criteria, no specific reduction can be exhibited, so the honest finding is no significant circularity.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim depends mainly on the University-1652 evaluation setup, the transferability of DINOv2 features, the validity of drone substitution as an augmentation, and the reliability of Gemini's structured re-ranking output. No new physical entities are introduced. The main free parameter is P=0.3, selected on the test split, and other hyperparameters such as K and the VLM temperature are fixed without sensitivity analysis.

free parameters (5)
  • drone substitution probability P = 0.3
    Chosen as best R@1 on the test split in Table 2 over P=0, 0.1, 0.3, and 0.5. This is a hyperparameter fit to the evaluation set.
  • number of re-ranking candidates K = 10
    Stage II re-ranks the Top-10 retrieved satellite images; no sensitivity analysis is reported for other values of K.
  • VLM inference temperature = 0
    Set to 0 for reproducibility, but API-side nondeterminism is not controlled.
  • VLM thinking budget = 1024
    Fixed for both Gemini variants; the effect of this budget on accuracy is not ablated.
  • input image size = not reported (candidates: 224, 384, 448, 518)
    Section 3.1 lists four possible sizes but the final configuration is not stated; image resolution affects retrieval accuracy.
assumptions (4)
  • domain assumption University-1652 provides aligned street, drone, and satellite images of the same buildings, and matching a street query to its satellite image is a well-posed retrieval task.
    Section 4 uses the dataset's train/test split and assumes the ground-truth correspondences are correct.
  • ad hoc to paper Randomly replacing satellite images with same-location drone images during training preserves the retrieval task at inference, where only satellite references are used.
    Section 3.1 introduces this as the drone augmentation strategy; the training-time distribution differs from the inference-time distribution.
  • domain assumption DINOv2 embeddings pretrained on LVD-142M transfer well to limited-FOV street-satellite matching.
    Stage I backbone choice is motivated by Table 1; no analysis of dataset overlap or domain shift is provided.
  • domain assumption Gemini 2.5 Flash can reliably rank 10 satellite images against one street image and return a structured JSON ranking with temperature 0.
    Section 3.2 relies on the VLM's reasoning and structured output; no validation of output parsing failures is reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VICI: VLM-Instructed Cross-view Image-localisation." pith.science (2026). https://pith.science/paper/XLP34ZDE

@misc{pith2026250704107,
  author       = {Pith},
  title        = {Pith review of: VICI: VLM-Instructed Cross-view Image-localisation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XLP34ZDE}},
  note         = {Machine review of arXiv:2507.04107}
}
read the original abstract

In this paper, we present a high-performing solution to the UAVM 2025 Challenge, which focuses on matching narrow FOV street-level images to corresponding satellite imagery using the University-1652 dataset. As panoramic Cross-View Geo-Localisation nears peak performance, it becomes increasingly important to explore more practical problem formulations. Real-world scenarios rarely offer panoramic street-level queries; instead, queries typically consist of limited-FOV images captured with unknown camera parameters. Our work prioritises discovering the highest achievable performance under these constraints, pushing the limits of existing architectures. Our method begins by retrieving candidate satellite image embeddings for a given query, followed by a re-ranking stage that selectively enhances retrieval accuracy within the top candidates. This two-stage approach enables more precise matching, even under the significant viewpoint and scale variations inherent in the task. Through experimentation, we demonstrate that our approach achieves competitive results -specifically attaining R@1 and R@10 retrieval rates of \topone\% and \topten\% respectively. This underscores the potential of optimised retrieval and re-ranking strategies in advancing practical geo-localisation performance. Code is available at https://github.com/tavisshore/VICI.

Figures

Figures reproduced from arXiv: 2507.04107 by the authors.

Figure 1
Figure 1. localisation example from VICI. Top left: query image. Top right: top retrieved satellite image. Bottom: justification for this satellite image being re-ranked to Top-1. query images are not fully explored due to the extreme lack of con￾textual surrounding information. Instead, existing methods primarily focused on panoramic imagery, leveraging its wide FOV to extract descriptive features and optimise matching accur… view at source ↗
Figure 2
Figure 2. Street-level, drone, and satellite images from various [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. 3.1 Stage I: Coarse Retrieval In Stage I, we employ a Siamese network 𝑓 , without weight sharing, parametrised by 𝜃 to minimise the domain gap between street-level and satellite image features, generating embeddings 𝜂𝑡 where 𝑡 ∈ {𝑠𝑡𝑟𝑒𝑒𝑡, 𝑠𝑎𝑡 }. All input images 𝐼 are in RGB space: 𝐼 ∈ R 3×𝑊 ×𝐻 , and are resized to 𝑊 × 𝐻, where 𝐻 = 𝑊 ,𝑊 ∈ {224, 384, 448, 518} for different backbone extractor configurations. Mathemati… view at source ↗
Figures from the paper (1 more)
Figure 3
Figure 3. Figure 3: Overview of the architecture: Features extracted by separate DINOv2 branches, references retrieved by descending similarity, [PITH_FULL_IMAGE:figures/full_fig_p003_3.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

37 extracted references · 28 canonical work pages

  1. [2]

    Cross- view geo-localization via learning disentangled geometric layout correspondence

    Xiaohan Zhang, Xingyu Li, Waqas Sultani, Yi Zhou, and Safwan Wshah. Cross- view geo-localization via learning disentangled geometric layout correspondence. InProceedings of the AAAI conference on artificial intelligence, volume 37, pages 3480–3488, 2023

  2. [3]

    Geodtr+: Toward generic cross-view geolocalization via geometric disentan- glement.IEEE Transactions on Pattern Analysis and Machine Intelligence, 46(12):10419–10433, 2024

    Xiaohan Zhang, Xingyu Li, Waqas Sultani, Chen Chen, and Safwan Wshah. Geodtr+: Toward generic cross-view geolocalization via geometric disentan- glement.IEEE Transactions on Pattern Analysis and Machine Intelligence, 46(12):10419–10433, 2024

  3. [4]

    Sample4geo: Hard negative sampling for cross-view geo-localisation, 2023

    Fabian Deuser, Konrad Habel, and Norbert Oswald. Sample4geo: Hard negative sampling for cross-view geo-localisation, 2023

  4. [5]

    Yujiao Shi, Xin Yu, Dylan Campbell, and Hongdong Li. Where am i looking at? joint location and orientation estimation by cross-view matching.2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 4063– 4071, 2020

  5. [6]

    Simple, effective and general: A new backbone for cross-view image geo-localization, 2023

    Yingying Zhu, Hongji Yang, Yuxin Lu, and Qiang Huang. Simple, effective and general: A new backbone for cross-view image geo-localization, 2023

  6. [7]

    Cross-view image sequence geo-localization

    Xiaohan Zhang, Waqas Sultani, and Safwan Wshah. Cross-view image sequence geo-localization. InProceedings of the IEEE/CVF Winter Conference on Applica- tions of Computer Vision, pages 2914–2923, 2023

  7. [8]

    Image and object geo-localization.International Journal of Computer Vision, 132(4):1350– 1392, 2024

    Daniel Wilson, Xiaohan Zhang, Waqas Sultani, and Safwan Wshah. Image and object geo-localization.International Journal of Computer Vision, 132(4):1350– 1392, 2024

  8. [9]

    Congeo: Robust cross-view geo-localization across ground view variations

    Li Mi, Chang Xu, Javiera Castillo-Navarro, Syrielle Montariol, Wen Yang, Antoine Bosselut, and Devis Tuia. Congeo: Robust cross-view geo-localization across ground view variations. InEuropean Conference on Computer Vision, pages 214–230. Springer, 2024

Show all 37 references
  1. [10]

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

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

  2. [11]

    ChatGPT (july 5 version)

    OpenAI. ChatGPT (july 5 version). https://chat.openai.com/, 2024. Accessed: 2025-07-05

  3. [12]

    Gemini: a family of highly capable multimodal models.arXiv preprint arXiv:2312.11805, 2023

    Gemini Team, Rohan Anil, Sebastian Borgeaud, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, Katie Millican, et al. Gemini: a family of highly capable multimodal models.arXiv preprint arXiv:2312.11805, 2023

  4. [13]

    Llama: Open and efficient foundation language models.arXiv preprint arXiv:2302.13971, 2023

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models.arXiv preprint arXiv:2302.13971, 2023

  5. [14]

    Visual instruction tuning.Advances in neural information processing systems, 36:34892–34916, 2023

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning.Advances in neural information processing systems, 36:34892–34916, 2023

  6. [15]

    Bliva: A simple multimodal llm for better handling of text-rich visual questions

    Wenbo Hu, Yifan Xu, Yi Li, Weiyue Li, Zeyuan Chen, and Zhuowen Tu. Bliva: A simple multimodal llm for better handling of text-rich visual questions. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 2256–2264, 2024

  7. [16]

    Visual autoregressive modeling: Scalable image generation via next-scale prediction

    Keyu Tian, Yi Jiang, Zehuan Yuan, Bingyue Peng, and Liwei Wang. Visual autoregressive modeling: Scalable image generation via next-scale prediction. Advances in neural information processing systems, 37:84839–84865, 2024

  8. [17]

    On the location dependence of convolutional neural network features

    Scott Workman and Nathan Jacobs. On the location dependence of convolutional neural network features. In2015 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pages 70–78, 2015

  9. [18]

    Wide-area image geolo- calization with aerial reference imagery

    Scott Workman, Richard Souvenir, and Nathan Jacobs. Wide-area image geolo- calization with aerial reference imagery. InIEEE International Conference on Computer Vision (ICCV), pages 1–9, 2015

  10. [19]

    Lending orientation to neural networks for cross-view geo-localization

    Liu Liu and Hongdong Li. Lending orientation to neural networks for cross-view geo-localization. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5624–5633, 2019

  11. [20]

    Vigor: Cross-view image geo- localization beyond one-to-one retrieval

    Sijie Zhu, Taojiannan Yang, and Chen Chen. Vigor: Cross-view image geo- localization beyond one-to-one retrieval. InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, pages 3640–3649, 2021

  12. [21]

    Spagbol: Spatial-graph-based orientated localisation

    Tavis Shore, Oscar Mendez, and Simon Hadfield. Spagbol: Spatial-graph-based orientated localisation. In2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 6858–6867, 2025

  13. [22]

    Peng: Pose-enhanced geo- localisation.IEEE Robotics and Automation Letters, 10(4):3835–3842, 2025

    Tavis Shore, Oscar Mendez, and Simon Hadfield. Peng: Pose-enhanced geo- localisation.IEEE Robotics and Automation Letters, 10(4):3835–3842, 2025

  14. [23]

    Cv-cities: Ad- vancing cross-view geo-localization in global cities.IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 18:1592–1606, 2025

    Gaoshuang Huang, Yang Zhou, Luying Zhao, and Wenjian Gan. Cv-cities: Ad- vancing cross-view geo-localization in global cities.IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 18:1592–1606, 2025

  15. [24]

    Attention is all you need

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

  16. [25]

    Hongji Yang, Xiufan Lu, and Ying J. Zhu. Cross-view geo-localization with layer-to-layer transformer. InNeural Information Processing Systems, 2021

  17. [26]

    Transgeo: Transformer is all you need for cross-view image geo-localization.2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 1152–1161, 2022

    Sijie Zhu, Mubarak Shah, and Chen Chen. Transgeo: Transformer is all you need for cross-view image geo-localization.2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 1152–1161, 2022

  18. [27]

    Georeasoner: Geo-localization with reasoning in street views using a large vision-language model

    Ling Li, Yu Ye, Bingchuan Jiang, and Wei Zeng. Georeasoner: Geo-localization with reasoning in street views using a large vision-language model. InForty-first International Conference on Machine Learning, 2024

  19. [28]

    Where am i? cross-view geo-localization with natural language descriptions.arXiv preprint arXiv:2412.17007, 2024

    Junyan Ye, Honglin Lin, Leyan Ou, Dairong Chen, Zihao Wang, Conghui He, and Weijia Li. Where am i? cross-view geo-localization with natural language descriptions.arXiv preprint arXiv:2412.17007, 2024

  20. [29]

    Geovlm: Improving auto- mated vehicle geolocalisation using vision-language matching, 2025

    Barkin Dagda, Muhammad Awais, and Saber Fallah. Geovlm: Improving auto- mated vehicle geolocalisation using vision-language matching, 2025

  21. [30]

    University-1652: A multi-view multi-source benchmark for drone-based geo-localization.ACM Multimedia, 2020

    Zhedong Zheng, Yunchao Wei, and Yi Yang. University-1652: A multi-view multi-source benchmark for drone-based geo-localization.ACM Multimedia, 2020

  22. [31]

    Pytorch: An imperative style, high-performance deep learning library.Advances in neural information processing systems, 32, 2019

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gre- gory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An imperative style, high-performance deep learning library.Advances in neural information processing systems, 32, 2019

  23. [32]

    Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

  24. [33]

    A convnet for the 2020s

    Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, and Saining Xie. A convnet for the 2020s. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 11976– 11986, June 2022

  25. [34]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xi- aohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition ...

  26. [35]

    Dinov2: Learning robust visual features without supervision

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

  27. [36]

    Gemini 2.5 flash model

    Google Cloud. Gemini 2.5 flash model. https://cloud.google.com/vertex-ai/ generative-ai/docs/models/gemini/2-5-flash, 2025. Accessed: 2025-06-15

  28. [37]

    Each part matters: Local patterns facilitate cross-view geo- localization.IEEE Transactions on Circuits and Systems for Video Technology, 32(2):867–879, 2021

    Tingyu Wang, Zhedong Zheng, Chenggang Yan, Jiyong Zhang, Yaoqi Sun, Bolun Zheng, and Yi Yang. Each part matters: Local patterns facilitate cross-view geo- localization.IEEE Transactions on Circuits and Systems for Video Technology, 32(2):867–879, 2021

  29. [38]

    The 3rd workshop on uavs in multime- dia: Capturing the world from a new perspective

    Tingyu Wang, Yujiao Shi, Fabian Deuser, Shaofei Huang, Guosheng Hu, Si Liu, Zhedong Zheng, and Roger Zimmermann. The 3rd workshop on uavs in multime- dia: Capturing the world from a new perspective. InProceedings of the 33rd ACM International Conference on Multimedia Workshop, 2025

Pith tools

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