Pith. sign in

REVIEW 3 major objections 3 minor 54 references

GeoExplorer: Active Geo-localization with Curiosity-Driven Exploration

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

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

Pith's one-line read GeoExplorer claims that replacing distance-based rewards with goal-agnostic curiosity rewards makes active geo-localization agents generalize to unfamiliar targets and environments.

desk verdict A plausible application of curiosity-driven RL to geo-localization, but the generalization claim rests on an unverified environment-model transfer and the supplied full text is unreadable. read the letter →

arxiv 2508.00152 v1 pith:QTHPLLCJ submitted 2025-07-31 cs.CV

Li Mi , Manon Bechaz , Zeming Chen , Antoine Bosselut , Devis Tuia This is my paper
classification cs.CV
keywords activegeo-localizationcuriosity-drivenexplorationintrinsicrewardreinforcementlearningenvironmentmodelinggeneralizationgoal-agnostic
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

Active geo-localization asks an agent to find a goal described by an image or text inside a search area, and current agents learn this by minimizing a distance-based reward. GeoExplorer proposes to replace that signal with a curiosity-driven intrinsic reward: the agent is rewarded for reducing the prediction error of a learned environment model, not for getting closer to the goal. Because the reward does not refer to the goal, the exploration policy it induces can stay useful when distance estimates are unreliable and can transfer to unseen targets and environments. The paper reports experiments across four AGL benchmarks supporting this generalization claim. A sympathetic reader would take the central claim to be that goal-agnostic surprise is a better training signal for geo-localization than goal-distance minimization.

What carries the argument

The load-bearing object is the intrinsic curiosity reward, $r^{\mathrm{cur}}_t$, generated by an environment model: at each step the model tries to predict the next observation from the current observation and action, and the reward is the model's prediction error. The agent's policy is trained by reinforcement learning to maximize the accumulated intrinsic reward, so it learns to seek out states where its model is surprised. Because the reward contains no information about the goal's location, the resulting exploration is goal-agnostic by construction, which is what the paper claims yields generalizable behavior. The abstract's phrase 'effective environment modeling' identifies this prediction-error mechanism as the source of the exploration signal.

What would settle it

Train GeoExplorer on one region and evaluate it in a visually repetitive or featureless environment where the environment model's prediction error is uniformly low; if localization success collapses to random-search level while a distance-reward baseline still succeeds, the curiosity signal is not carrying the generalization the paper claims.

Watch

Extended reading notes

Core claim

The central claim is that the exploration policy in active geo-localization should be learned from a goal-agnostic curiosity reward rather than a distance-based reward. GeoExplorer defines this reward through an environment model: the agent gains reward when its model fails to predict the next observation, so it is driven to visit states that are surprising and therefore informative. The paper argues that this produces exploration that is reliable, diverse, and contextually relevant, and that it directly addresses the failure mode of distance-based RL agents when distance estimates are noisy or the target or environment is unfamiliar. The paper's experiments on four AGL benchmarks are offered as evidence that curiosity-driven exploration matches or exceeds distance-reward baselines, particularly in localizing unfamiliar targets and environments.

Load-bearing premise

The load-bearing premise is that the environment model's prediction error remains a useful and well-calibrated signal in unfamiliar environments, so the curiosity reward drives informative exploration instead of noise-chasing or aimless wandering.

Editorial extensions

If this is right

  • Distance-based reward can be dropped or de-emphasized in AGL, so agents no longer need reliable distance estimation to explore well.
  • An AGL agent trained with curiosity should localize goals it has never seen, and in environments outside its training distribution, more reliably than distance-reward agents.
  • The same goal-agnostic exploration module can be reused across different goals within a search area, since the reward does not depend on a particular target.
  • Benchmark comparisons should include generalization splits, not only average success on familiar targets, because that is where curiosity rewards are claimed to help.
  • The environment model, not the goal-reaching policy, becomes the component that determines exploration quality and therefore localization success.

Reading between the lines

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

  • If curiosity-driven exploration transfers as claimed, the same environment-model surprise reward could be applied to other embodied search tasks such as object finding, disaster response, or visual navigation, where distance to an unseen target is hard to measure.
  • The paper leaves implicit that the environment model's prediction error must stay informative in deployment environments; a direct test would be to measure the reward signal's magnitude on unseen environments and check whether it tracks useful exploration or saturates.
  • One testable extension is to balance the curiosity reward against a sparse goal-reached reward to prevent the agent from exploring indefinitely without committing to a localization; the paper does not report such an ablation.
  • A stronger claim implicit in the method is that surprise correlates with geospatial information gain, which could be tested by comparing curiosity-selected views against a mutual-information baseline on satellite imagery.
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 / 3 minor

Summary. The paper proposes GeoExplorer, an active geo-localization (AGL) agent that replaces distance-based rewards with a goal-agnostic curiosity reward derived from environment-model prediction error. The authors argue that distance-based rewards are unreliable when targets and environments are unfamiliar, and that a curiosity-driven intrinsic reward enables more robust and generalizable exploration. The abstract claims extensive experiments across four AGL benchmarks, but the supplied manuscript is heavily corrupted: the mathematical formulation is garbled and all experimental tables are unreadable, so the central empirical claims cannot be checked.

Significance. The conceptual idea is relevant to the AGL community, since distance-based rewards are known to be brittle when distance estimates are noisy, and intrinsic motivation is a plausible remedy. If validated, the method could provide a useful baseline for curiosity-driven exploration in geo-localization. However, the manuscript provides no machine-checked proofs, no code, and no readable experimental results, so the significance of the contribution cannot be assessed from this version.

major comments (3)
  1. [Experiments (tables)] The central empirical claim is not verifiable because all numeric entries in the experimental tables are unreadable in the supplied manuscript; the claimed improvements over baselines across the four AGL benchmarks are therefore unsupported. Since the abstract's conclusion rests entirely on these numbers, this is a load-bearing gap.
  2. [Section 3 (Method)] The intrinsic reward is defined through environment-model prediction error, but the paper provides no argument or experiment showing that this error remains informative under distribution shift. On unseen environments, prediction error can become uniformly high, yielding an uninformative reward and undirected exploration; an analysis or ablation on test-environment error maps and reward-scale sensitivity is needed to support the generalization claim.
  3. [Section 3 (reward equations)] The reward equations are not readable in the supplied manuscript, so I cannot confirm that the curiosity reward is goal-agnostic or how it is combined with the goal-reaching objective. A clean version of the mathematical formulation is required before the method can be evaluated.
minor comments (3)
  1. [Abstract] The phrase 'proven through extensive experiments' overclaims for a benchmark study; wording such as 'suggest' or 'indicate' would be more appropriate.
  2. [Experiments] The captions for figures and tables are missing from the supplied text, making it difficult to interpret the structure of the experimental section.
  3. [References] The bibliography is truncated in the supplied manuscript, with several references appearing incomplete; the final version should contain the full reference list.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: GeoExplorer's central claim is an empirical benchmark comparison, not a derivation from its own inputs.

full rationale

The paper's central claim is empirical: an active geo-localization RL agent that augments goal-reaching with a goal-agnostic intrinsic curiosity reward—defined from environment-model prediction error—should localize unfamiliar targets more robustly than distance-reward baselines. This is a design proposal validated by benchmark results, not a derivation whose conclusion is identical to its premise. Nothing in the supplied abstract or readable manuscript text fits a parameter to the reported localization outcomes and then renames that fit as a prediction; the curiosity reward is defined from the agent's environment-model prediction error, and the claimed generalization is evaluated by an independent localization task. The concern that the environment model may produce uninformative prediction error under distribution shift is a missing-evidence or correctness concern about empirical generalization, not a circularity: the paper does not define 'generalization' as high prediction error, and it does not appear to invoke a load-bearing self-citation or imported uniqueness theorem to force its design choice. Because the full text is largely unreadable in the supplied rendering, no equation was available to exhibit a reduction of a predicted result to its inputs; absent such a specific reduction, no circular step should be asserted. The honest finding is therefore no significant circularity.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

Only the abstract is readable in the supplied extraction, so this ledger is necessarily incomplete. The method will surely introduce free parameters such as the intrinsic-reward weight, the environment-model architecture, and the novelty-scale constants, but they cannot be enumerated here. The two axioms listed are premises the abstract explicitly depends on. No invented entities are proposed at the abstract level.

assumptions (2)
  • domain assumption The fragility of distance-based AGL agents is caused by unreliable learned exploration, not by representation or goal-encoding deficiencies.
    The abstract motivates the entire design with this diagnosis of prior methods ('reduced robustness and generalization ability due to the less reliable exploration strategy learned during training'). If the real bottleneck is elsewhere, the intended improvement mechanism is misdirected.
  • domain assumption The environment model behind the intrinsic reward stays predictive enough to produce useful novelty signals in unseen environments.
    The abstract claims 'robust, diverse, and contextually relevant exploration based on effective environment modeling'. The method works only if prediction-error-based curiosity remains informative out of distribution and does not displace goal-reaching, which is exactly the weakest assumption identified in the report.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GeoExplorer: Active Geo-localization with Curiosity-Driven Exploration." pith.science (2026). https://pith.science/paper/QTHPLLCJ

@misc{pith2026250800152,
  author       = {Pith},
  title        = {Pith review of: GeoExplorer: Active Geo-localization with Curiosity-Driven Exploration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QTHPLLCJ}},
  note         = {Machine review of arXiv:2508.00152}
}
read the original abstract

Active Geo-localization (AGL) is the task of localizing a goal, represented in various modalities (e.g., aerial images, ground-level images, or text), within a predefined search area. Current methods approach AGL as a goal-reaching reinforcement learning (RL) problem with a distance-based reward. They localize the goal by implicitly learning to minimize the relative distance from it. However, when distance estimation becomes challenging or when encountering unseen targets and environments, the agent exhibits reduced robustness and generalization ability due to the less reliable exploration strategy learned during training. In this paper, we propose GeoExplorer, an AGL agent that incorporates curiosity-driven exploration through intrinsic rewards. Unlike distance-based rewards, our curiosity-driven reward is goal-agnostic, enabling robust, diverse, and contextually relevant exploration based on effective environment modeling. These capabilities have been proven through extensive experiments across four AGL benchmarks, demonstrating the effectiveness and generalization ability of GeoExplorer in diverse settings, particularly in localizing unfamiliar targets and environments.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

54 extracted references · 45 canonical work pages

  1. [1]

    The Falcon series of open language models

    Ebtesam Almazrouei, Hamza Alobeidli, Abdulaziz Alshamsi, Alessandro Cappelli, Ruxandra Cojocaru, M \'e rouane Debbah, \'E tienne Goffinet, Daniel Hesslow, Julien Launay, Quentin Malartic, et al. The Falcon series of open language models. arXiv preprint arXiv:2311.16867, 2023

    arXiv 2023

  2. [2]

    OpenStreetView-5M : The many roads to global visual geolocation

    Guillaume Astruc, Nicolas Dufour, Ioannis Siglidis, Constantin Aronssohn, Nacim Bouia, Stephanie Fu, Romain Loiseau, Van Nguyen Nguyen, Charles Raude, Elliot Vincent, et al. OpenStreetView-5M : The many roads to global visual geolocation. In CVPR, pages 21967--21977, 2024

    work page 2024

  3. [3]

    Navigation world models

    Amir Bar, Gaoyue Zhou, Danny Tran, Trevor Darrell, and Yann LeCun. Navigation world models. In CVPR, pages 15791--15801, 2025

    work page 2025

  4. [4]

    Rethinking visual geo-localization for large-scale applications

    Gabriele Berton, Carlo Masone, and Barbara Caputo. Rethinking visual geo-localization for large-scale applications. In CVPR, pages 4878--4888, 2022 a

    work page 2022

  5. [5]

    Deep visual geo-localization benchmark

    Gabriele Berton, Riccardo Mereu, Gabriele Trivigno, Carlo Masone, Gabriela Csurka, Torsten Sattler, and Barbara Caputo. Deep visual geo-localization benchmark. In CVPR, pages 5396--5407, 2022 b

    work page 2022

  6. [6]

    Large-scale study of curiosity-driven learning

    Yuri Burda, Harri Edwards, Deepak Pathak, Amos Storkey, Trevor Darrell, and Alexei A Efros. Large-scale study of curiosity-driven learning. In ICLR, 2018

    work page 2018

  7. [7]

    Chen, Georges Baatz, Kevin Köser, Sam S

    David M. Chen, Georges Baatz, Kevin Köser, Sam S. Tsai, Ramakrishna Vedantham, Timo Pylvänäinen, Kimmo Roimela, Xin Chen, Jeff Bach, Marc Pollefeys, Bernd Girod, and Radek Grzeszczuk. City-scale landmark identification on mobile devices. In CVPR, pages 737--744, 2011

    work page 2011

  8. [8]

    Decision transformer: Reinforcement learning via sequence modeling

    Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Misha Laskin, Pieter Abbeel, Aravind Srinivas, and Igor Mordatch. Decision transformer: Reinforcement learning via sequence modeling. In NeurIPS, pages 15084--15097, 2021

    work page 2021

Show all 54 references
  1. [9]

    Sat2Cap : Mapping fine-grained textual descriptions from satellite images

    Aayush Dhakal, Adeel Ahmad, Subash Khanal, Srikumar Sastry, Hannah Kerner, and Nathan Jacobs. Sat2Cap : Mapping fine-grained textual descriptions from satellite images. In CVPRW, pages 533--542, 2024

    2024

  2. [10]

    Scalable place recognition under appearance change for autonomous driving

    Dzung Doan, Yasir Latif, Tat-Jun Chin, Yu Liu, Thanh-Toan Do, and Ian Reid. Scalable place recognition under appearance change for autonomous driving. In ICCV, 2019

    2019

  3. [11]

    RvS : What is essential for offline RL via supervised learning? In ICLR, 2022

    Scott Emmons, Benjamin Eysenbach, Ilya Kostrikov, and Sergey Levine. RvS : What is essential for offline RL via supervised learning? In ICLR, 2022

    2022

  4. [12]

    Aerial vision-and-dialog navigation

    Yue Fan, Winson Chen, Tongzhou Jiang, Chun Zhou, Yi Zhang, and Xin Eric Wang. Aerial vision-and-dialog navigation. In ACL Findings 2023, pages 3043--3061, 2022

    2023

  5. [13]

    Uncertainty-aware vision-based metric cross-view geolocalization

    Florian Fervers, Sebastian Bullinger, Christoph Bodensteiner, Michael Arens, and Rainer Stiefelhagen. Uncertainty-aware vision-based metric cross-view geolocalization. In CVPR, pages 21621--21631, 2023

    2023

  6. [14]

    Vision-and-language navigation : A survey of tasks, methods, and future directions

    Jing Gu, Eliana Stefani, Qi Wu, Jesse Thomason, and Xin Eric Wang. Vision-and-language navigation : A survey of tasks, methods, and future directions. In ACL, pages 7606--7623, 2022

    2022

  7. [15]

    Creating xBD : A dataset for assessing building damage from satellite imagery

    Ritwik Gupta, Bryce Goodman, Nirav Patel, Ricky Hosfelt, Sandra Sajeev, Eric Heim, Jigar Doshi, Keane Lucas, Howie Choset, and Matthew Gaston. Creating xBD : A dataset for assessing building damage from satellite imagery. In CVPRW, pages 10--17, 2019

    2019

  8. [16]

    Mastering diverse domains through world models

    Danijar Hafner, Jurgis Pasukonis, Jimmy Ba, and Timothy Lillicrap. Mastering diverse domains through world models. arXiv preprint arXiv:2301.04104, 2023

    2023 arXiv

  9. [17]

    Offline reinforcement learning as one big sequence modeling problem

    Michael Janner, Qiyang Li, and Sergey Levine. Offline reinforcement learning as one big sequence modeling problem. In NeurIPS, pages 1273--1286, 2021

    2021

  10. [18]

    Goat-bench: A benchmark for multi-modal lifelong navigation

    Mukul Khanna, Ram Ramrakhya, Gunjan Chhablani, Sriram Yenamandra, Theophile Gervet, Matthew Chang, Zsolt Kira, Devendra Singh Chaplot, Dhruv Batra, and Roozbeh Mottaghi. Goat-bench: A benchmark for multi-modal lifelong navigation. In CVPR, pages 16373--16383, 2024

    2024

  11. [19]

    Active world model learning with progress curiosity

    Kuno Kim, Megumi Sano, Julian De Freitas, Nick Haber, and Daniel Yamins. Active world model learning with progress curiosity. In ICML, pages 5306--5315, 2020

    2020

  12. [20]

    Visual instruction tuning

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. NeurIPS, 36: 0 34892--34916, 2023 a

    2023

  13. [21]

    AerialVLN : Vision-and-language navigation for uavs

    Shubo Liu, Hongsheng Zhang, Yuankai Qi, Peng Wang, Yanning Zhang, and Qi Wu. AerialVLN : Vision-and-language navigation for uavs. In ICCV, pages 15384--15394, 2023 b

    2023

  14. [22]

    ELIGN : Expectation alignment as a multi-agent intrinsic reward

    Zixian Ma, Rose Wang, Fei-Fei Li, Michael Bernstein, and Ranjay Krishna. ELIGN : Expectation alignment as a multi-agent intrinsic reward. In NeurIPS, pages 8304--8317, 2022

    2022

  15. [23]

    Obstacle-aware adaptive informative path planning for uav-based target search

    Ajith Anil Meera, Marija Popović, Alexander Millane, and Roland Siegwart. Obstacle-aware adaptive informative path planning for uav-based target search. In ICRA, pages 718--724, 2019

    2019

  16. [24]

    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. In ECCV, pages 214--230, 2024

    2024

  17. [25]

    Machine learning for aerial image labeling

    Volodymyr Mnih. Machine learning for aerial image labeling. PhD thesis, University of Toronto (Canada), 2013

    2013

  18. [26]

    Vision-based navigation planning for autonomous post-earthquake inspection of reinforced concrete railway viaducts using unmanned aerial vehicles

    Yasutaka Narazaki, Vedhus Hoskere, Girish Chowdhary, and Billie F Spencer Jr. Vision-based navigation planning for autonomous post-earthquake inspection of reinforced concrete railway viaducts using unmanned aerial vehicles. Automation in Construction, 137: 0 104214, 2022

    2022

  19. [27]

    Representation learning with contrastive predictive coding

    Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748, 2018

    2018 arXiv

  20. [28]

    Curiosity-driven exploration by self-supervised prediction

    Deepak Pathak, Pulkit Agrawal, Alexei A Efros, and Trevor Darrell. Curiosity-driven exploration by self-supervised prediction. In ICML, pages 2778--2787, 2017

    2017

  21. [29]

    Aerial view localization with reinforcement learning: Towards emulating search-and-rescue

    Aleksis Pirinen, Anton Samuelsson, John Backsund, and Kalle str \"o m. Aerial view localization with reinforcement learning: Towards emulating search-and-rescue. Swedish Artificial Intelligence Society, pages 28--37, 2023

    2023

  22. [30]

    Improving language understanding by generative pre-training

    Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. Improving language understanding by generative pre-training. OpenAI blog, 2018

    2018

  23. [31]

    Language models are unsupervised multitask learners

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. Language models are unsupervised multitask learners. OpenAI blog, 2019

    2019

  24. [32]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In ICML, pages 8748--8763, 2021

    2021

  25. [33]

    GOMAA-Geo : Goal modality agnostic active geo-localization

    Anindya Sarkar, Srikumar Sastry, Aleksis Pirinen, Chongjie Zhang, Nathan Jacobs, and Yevgeniy Vorobeychik. GOMAA-Geo : Goal modality agnostic active geo-localization. In NeurIPS, pages 104934--104964, 2024

    2024

  26. [34]

    A possibility for implementing curiosity and boredom in model-building neural controllers

    J \"u rgen Schmidhuber. A possibility for implementing curiosity and boredom in model-building neural controllers. In Proc. of the International Conference on Simulation of Adaptive Behavior: From Animals to Animats, pages 222--227, 1991

    1991

  27. [35]

    Reinforcement learning upside down: Don't predict rewards--just map them to actions

    J \"u rgen Schmidhuber. Reinforcement learning upside down: Don't predict rewards--just map them to actions. arXiv preprint arXiv:1912.02875, 2019

    1912 arXiv

  28. [36]

    Proximal policy optimization algorithms

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347, 2017

    2017 arXiv

  29. [37]

    Vi NT : A foundation model for visual navigation

    Dhruv Shah, Ajay Sridhar, Nitish Dashora, Kyle Stachowicz, Kevin Black, Noriaki Hirose, and Sergey Levine. Vi NT : A foundation model for visual navigation. In CoRL, 2023

    2023

  30. [38]

    Boosting 3-dof ground-to-satellite camera localization accuracy via geometry-guided cross-view transformer

    Yujiao Shi, Fei Wu, Akhil Perincherry, Ankit Vora, and Hongdong Li. Boosting 3-dof ground-to-satellite camera localization accuracy via geometry-guided cross-view transformer. In ICCV, pages 21516--21526, 2023

    2023

  31. [39]

    PEnG : Pose-enhanced geo-localisation

    Tavis Shore, Oscar Mendez, and Simon Hadfield. PEnG : Pose-enhanced geo-localisation. IEEE Robotics and Automation Letters, 10 0 (4): 0 3835--3842, 2025

    2025

  32. [40]

    Training agents using upside-down reinforcement learning

    Rupesh Kumar Srivastava, Pranav Shyam, Filipe Mutz, Wojciech Ja \'s kowski, and J \"u rgen Schmidhuber. Training agents using upside-down reinforcement learning. arXiv preprint arXiv:1912.02877, 2019

    1912 arXiv

  33. [41]

    Incentivizing exploration in reinforcement learning with deep predictive models

    Bradly C Stadie, Sergey Levine, and Pieter Abbeel. Incentivizing exploration in reinforcement learning with deep predictive models. arXiv preprint arXiv:1507.00814, 2015

    2015 arXiv

  34. [42]

    GeoCLIP : Clip-inspired alignment between locations and images for effective worldwide geo-localization

    Vicente Vivanco Cepeda, Gaurav Kumar Nayak, and Mubarak Shah. GeoCLIP : Clip-inspired alignment between locations and images for effective worldwide geo-localization. In NeurIPS, pages 8690--8701, 2023

    2023

  35. [43]

    MimicPlay : Long-horizon imitation learning by watching human play

    Chen Wang, Linxi Fan, Jiankai Sun, Ruohan Zhang, Li Fei-Fei, Danfei Xu, Yuke Zhu, and Anima Anandkumar. MimicPlay : Long-horizon imitation learning by watching human play. In CoRL, 2023 a

    2023

  36. [44]

    Fine-grained cross-view geo-localization using a correlation-aware homography estimator

    Xiaolong Wang, Runsen Xu, Zhuofan Cui, Zeyu Wan, and Yu Zhang. Fine-grained cross-view geo-localization using a correlation-aware homography estimator. In NeurIPS, pages 5301--5319, 2023 b

    2023

  37. [45]

    Towards realistic uav vision-language navigation: Platform, benchmark, and methodology

    Xiangyu Wang, Donglin Yang, Ziqin Wang, Hohin Kwan, Jinyu Chen, Wenjun Wu, Hongsheng Li, Yue Liao, and Si Liu. Towards realistic uav vision-language navigation: Platform, benchmark, and methodology. In ICLR, 2025

    2025

  38. [46]

    Attention is all you need

    A Waswani, N Shazeer, N Parmar, J Uszkoreit, L Jones, A Gomez, L Kaiser, and I Polosukhin. Attention is all you need. In NeurIPS, 2017

    2017

  39. [47]

    Wide-area image geolocalization with aerial reference imagery

    Scott Workman, Richard Souvenir, and Nathan Jacobs. Wide-area image geolocalization with aerial reference imagery. In ICCV, pages 3961--3969, 2015

    2015

  40. [48]

    Unleashing large-scale video generative pre-training for visual robot manipulation

    Hongtao Wu, Ya Jing, Chilam Cheang, Guangzeng Chen, Jiafeng Xu, Xinghang Li, Minghuan Liu, Hang Li, and Tao Kong. Unleashing large-scale video generative pre-training for visual robot manipulation. In ICLR, 2024

    2024

  41. [49]

    FG 2 : Fine-grained cross-view localization by fine-grained feature matching

    Zimin Xia and Alexandre Alahi. FG 2 : Fine-grained cross-view localization by fine-grained feature matching. In CVPR, pages 6362--6372, 2025

    2025

  42. [50]

    Learning interactive real-world simulators

    Sherry Yang, Yilun Du, Seyed Kamyar Seyed Ghasemipour, Jonathan Tompson, Leslie Pack Kaelbling, Dale Schuurmans, and Pieter Abbeel. Learning interactive real-world simulators. In ICLR, 2024

    2024

  43. [51]

    Introduction to large-scale visual geo-localization

    Amir R Zamir, Asaad Hakeem, Luc Van Gool, Mubarak Shah, and Richard Szeliski. Introduction to large-scale visual geo-localization. In Large-Scale Visual Geo-Localization, pages 1--18. Springer, 2016

    2016

  44. [52]

    GeoDTR+ : Toward generic cross-view geolocalization via geometric disentanglement

    Xiaohan Zhang, Xingyu Li, Waqas Sultani, Chen Chen, and Safwan Wshah. GeoDTR+ : Toward generic cross-view geolocalization via geometric disentanglement. IEEE Transactions on Pattern Analysis and Machine Intelligence, 46 0 (12): 0 10419--10433, 2024

    2024

  45. [53]

    DINO-WM : World models on pre-trained visual features enable zero-shot planning

    Gaoyue Zhou, Hengkai Pan, Yann LeCun, and Lerrel Pinto. DINO-WM : World models on pre-trained visual features enable zero-shot planning. In ICML, 2025

    2025

  46. [54]

    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. In CVPR, pages 3640--3649, 2021

    2021

Pith tools

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