Pith. sign in

REVIEW 4 major objections 6 minor 4 cited by

GeoLocSFT: Efficient Visual Geolocation via Supervised Fine-Tuning of Multimodal Foundation Models

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that fine-tuning a large multimodal foundation model on just ~2,700 curated image–GPS pairs, each annotated with an LLM-generated "geo-caption", delivers competitive planet-scale visual geolocation in a single forward pass.

desk verdict Useful efficiency study with a novel SFT data pipeline, but the abstract oversells the results and the new benchmark needs a train/test overlap check. read the letter →

arxiv 2506.01277 v1 pith:EVBCKCFK submitted 2025-06-02 cs.AI

Qiang Yi , Lianlei Shan This is my paper
classification cs.AI
keywords visualgeolocationsupervisedfine-tuningmultimodalfoundationmodelsgeo-captionsMR40kbenchmarkimageGPSestimationdataqualityversusquantityplanet-scalelocalization
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

Visual geolocation — inferring where a photograph was taken from pixels alone — has usually required millions of geotagged images or multi-stage pipelines with large training budgets. GeoLocSFT claims a much cheaper route: take an instruction-tuned multimodal foundation model, fine-tune it for a single epoch on about 2,700 carefully chosen image–GPS pairs, and let each pair's training label be a detailed "geo-caption" written by a large language model. These captions analyze the scene at regional and local scales, name micro-features such as road markings and tree species, and argue why similar-looking regions are wrong, ending in coordinates. On standard benchmarks (Im2GPS-3k, YFCC-4k, OSV5M, GWS15K) and the paper's new 40,000-image MR40k set of sparsely populated areas, the fine-tuned 27B model beats its untuned baseline at every distance threshold, and a 3B model with the same data shows strong regional gains. If this holds, high-quality supervision can substitute for massive databases in planet-scale geolocation.

What carries the argument

The central object is the geo-caption: a multi-part textual annotation that converts a raw image–GPS pair into a supervised reasoning target. A large language model is instructed to analyze the image at two scales — regional context out to roughly 25 km and local micro-features within 1 km — to cite specific evidence such as line-marking measurements, plant species, and infrastructure standards, and to state why plausible alternative regions are excluded, before emitting coordinates inside <answer> tags. The caption does double duty: it is the supervision signal during fine-tuning, because the model learns to generate it, and it is the inference scaffold, because coordinates are read out of the generated text. The paper's ablation with 2,700 curated captions versus 100,000 unlabeled pairs is what ties the measured gains to the caption's content rather than to data volume.

What would settle it

Run an overlap audit between MR40k and the 2,700 training images: compare GPS coordinates at street level (say, within 50 meters) and detect near-duplicate images. If a substantial fraction of MR40k images lie at or near training locations, and accuracy on the non-overlapping remainder drops toward the untuned baseline, the generalization claim fails; a cleaner test is to fine-tune with all near-duplicates of MR40k removed and check whether the benchmark scores survive.

Watch

Extended reading notes

Core claim

GeoLocSFT's central claim is that a foundation model adapted on a few thousand reasoning-rich examples can geolocate images competitively without a reference database or a complicated inference pipeline. The training target is the innovation: each of roughly 2,700 images is paired with a structured geo-caption — produced by prompting Claude 3.7 Sonnet as an "expert geographer" — that combines a 25-km regional analysis, a 1-km micro-feature inventory (road engineering, vegetation species, building styles, signage), and an explicit disambiguation of visually similar regions, ending with latitude and longitude in a fixed tag format. Fine-tuning with the standard next-token prediction loss teaches the model to reproduce that reasoning along with the coordinates; prediction is then one forward pass. On Im2GPS-3k the 27B model goes from 42.08% to 47.20% accuracy at 200 km, and on MR40k from 76.58% to 88.95% at 2,500 km, with similar gains on YFCC-4k, OSV5M, and GWS15K. One epoch of training suffices, multi-candidate re-ranking adds little, and the curated captions beat 100,000 unlabeled image–GPS pairs in the paper's ablation.

Load-bearing premise

The load-bearing assumption is that MR40k is a genuinely held-out test set, but both MR40k and the 2,700 training images are drawn from the same Mapillary pool, and the paper never states or verifies that test images were excluded from training; if locations overlap, the reported MR40k gains could be memorization rather than generalization.

Editorial extensions

If this is right

  • Fine-tuning a 27B multimodal model for one epoch on ~2,700 curated geo-caption pairs improves geolocation accuracy over the untuned baseline at every distance threshold on all five benchmarks tested.
  • The curated captions, not just more images, drive the gain: 2,700 high-quality pairs outperform 100,000 unlabeled image–GPS pairs on the 3B model in the paper's controlled comparison.
  • The fine-tuning stage carries nearly all the benefit: sampling ten candidates and re-ranking them with an LLM consensus prompt improves accuracy only marginally over the single-pass prediction.
  • The paper's new MR40k set, proposed for public release, offers 40,000 street-level images from areas with under 5,000 inhabitants, where every tested model scores markedly lower than on urban-centric benchmarks.
  • The recipe extends to smaller models: a fine-tuned 3B model reaches 67.67% at 200 km on the Austria subset, above GPT-4.1's 58.96%, suggesting curated data can produce expert-level regional specialization.

Reading between the lines

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

  • The caption-generation recipe is not tied to geolocation: any dense-label perceptual task where a language model can verbalize discriminative cues — building-age estimation, plant-species identification, dialect or signage mapping — could reuse the same curate-few, caption-deeply, fine-tune-briefly template.
  • If the single-epoch result generalizes, the effective bottleneck is caption quality rather than compute or data volume; a natural next experiment is scaling curated captions from 2,700 to tens of thousands and watching whether accuracy keeps climbing or saturates.
  • The MR40k benchmark would be strengthened by publishing explicit train/test overlap statistics; until then, comparisons on it should be read with the shared Mapillary source pool in mind.
  • The observed behavior of low error variance across repeated samples without mode collapse suggests the fine-tuned model encodes a calibrated distribution over plausible locations, which could support uncertainty-aware downstream uses such as filtering untrustworthy predictions.
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

4 major / 6 minor

Summary. The paper proposes GeoLocSFT, a framework that fine-tunes multimodal foundation models (Gemma 3 27B and Qwen2.5-VL-3B) on roughly 2,700 image-GPS pairs with LLM-generated geo-captions, and evaluates the result on OSV5M, GWS15K, YFCC4k, IM2GPS3K, and a new MR40k benchmark for sparsely populated regions. The authors report single-pass inference results, explore multi-candidate re-ranking, and claim that high-quality supervised data can substitute for massive databases in planet-scale visual geolocation.

Significance. The training-efficiency claim is concrete and attractive: one epoch on about 2,700 examples on 8 A100 GPUs in roughly 50 minutes, followed by single-pass inference, is a practically useful recipe if the accuracy claims hold. The paper also provides useful ablations separating data quality from data quantity, and the proposed MR40k benchmark could fill a real gap if it is a clean held-out set. However, the central 'highly competitive' claim is not supported by the paper's own comparison against specialized geolocation systems (Table 5), and the validity of the new benchmark is undermined by the absence of any stated train/test disjointness from the SFT pool. The significance of the contribution therefore depends on substantial revision of both the claims and the benchmark validation.

major comments (4)
  1. [Abstract and Table 5] The abstract's claim of 'highly competitive geolocation performance' on standard benchmarks is contradicted by Table 5. On YFCC4k, GeoLocSFT (Gemma 3 27B-SFT) achieves 5.21% at 1 km versus G3's 23.99%; on IM2GPS3K it achieves 8.80% at 1 km versus G3's 16.65% and 32.70% at 25 km versus G3's 40.94%. The paper needs either to add experiments that close this gap or to reframe its central claim as showing that SFT improves general-purpose LMMs with very little data, rather than claiming competitiveness with state-of-the-art geolocation pipelines.
  2. [Sections 3.2, 3.3, Tables 1 and 6] MR40k is not established as a held-out benchmark. Both the 2,700 SFT images (Section 3.2) and the 40,000 MR40k images (Section 3.3) are sampled from the same MR600k Mapillary pool, with no stated exclusion of SFT images from MR40k and no reported minimum pairwise distance. Because Mapillary images come in bursts and sequences, even non-identical images can be near-duplicate views of the same road. The MR40k rows in Tables 1 and 6 are therefore potentially inflated by memorization. The authors must report exact image-ID overlap and the minimum pairwise GPS distance between the SFT set and MR40k, and should re-evaluate on a guaranteed-disjoint split.
  3. [NeurIPS checklist item 7 and Tables 1, 5, 6] No error bars, confidence intervals, or significance tests are reported, and the checklist explicitly answers 'No' to the statistical-significance question. This matters because several headline improvements are small in absolute terms (e.g., OSV5M 1 km: 2.35 vs 1.74 in Table 1, and GWS15K 750 km: 69.65 vs 65.12 in Table 6), while Table 4 shows that a 100k-sample weak SFT dataset gives no improvement over baseline. Bootstrap intervals or multiple-seed runs are needed before 'substantial improvement' can be assessed.
  4. [Tables 1 and 6, MR40k row for Claude 3.7 Sonnet] The MR40k row for Claude 3.7 Sonnet lists 2500 km accuracy as 4.70%, which is lower than the 750 km value of 70.44% and violates the required monotonicity of Acc@R. This appears to be a transcription error, and it casts doubt on the reliability of the other numeric entries in the main results tables. The entries should be rechecked against raw evaluation logs and corrected.
minor comments (6)
  1. [Footnotes, Section 3.3] The paper states that MR40k will be publicly released, but the footnote says the link will be 'added upon publication or hosting.' A verifiable URL, data card, and license should be included with the submission.
  2. [Section 5.2, Table 2] The Austria subset is introduced without a definition of how it was constructed, how large it is, or why it is representative; also, Table 2 contains a footnote marker '†' on OSV5M that is never defined in the table caption or text.
  3. [Appendix D] The MCR consensus aggregation is said to use 'the fine-tuned GeoLocSFT (Gemma 3 27B) model'; if this is the same model that generated the candidates, the aggregation analysis may be biased, and the paper should clarify the exact prompt and model used for judging.
  4. [Tables 1, 5, and 6] Several numeric cells are malformed due to missing delimiters, for example '50.6664.84' in the IM2GPS3K Claude row and '46.7166.83' in the OSV5M row of Table 5; all table formatting should be normalized.
  5. [Figure 2 and Section 3.2] It is unclear whether the GPS coordinates (the 'Geometry' field listed in Step 1) were provided to Claude 3.7 Sonnet when generating the geo-captions; the paper should state explicitly whether the caption generator saw the true coordinates, since this affects what the SFT supervision actually teaches the model.
  6. [Section 5.1 and Tables 1, 6] The phrase 'substantially improves' is used for both models, but Qwen2.5-VL-3B-SFT degrades relative to its baseline at the 1 km threshold on IM2GPS3K (2.15 vs 2.75) and YFCC4k (0.75 vs 0.85) in Table 6; the language should be made threshold-specific.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: training signal is external, gains are measured on independent standard benchmarks, and the MR40k overlap concern is a data-hygiene risk, not a circular derivation.

full rationale

The paper's central claim is empirical: supervised fine-tuning on about 2,700 curated image-GPS pairs with LLM-generated geo-captions improves geolocation accuracy over zero-shot baselines. The training signal is generated by Claude 3.7 Sonnet from image content and metadata, not by the fine-tuned model or by the evaluation metrics, so the supervised signal is external to the claim being tested. The main results are reported on established benchmarks (Im2GPS-3k, YFCC-4k, OSV5M, GWS15K) as well as the authors' new MR40k set; the standard benchmarks provide independent support that does not depend on the paper's own data construction. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the authors' prior work, and no equation-level reduction makes a reported result equal to an input by construction. The one concern raised by the reviewer is that MR40k and the SFT training set are both sampled from the same MR600k pool without a stated exclusion or overlap check; if MR40k images are near-duplicates of training images, reported MR40k accuracy could be inflated by memorization. That is a legitimate benchmark-validity and data-hygiene risk, but it is not a circularity in the derivation: MR40k is not defined in terms of the model's outputs, and the paper does not fit a parameter to MR40k and then claim that fit as a prediction. Because the central SFT result is corroborated by external benchmarks, no circular step is established, and the circularity score is 0.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim rests on a small set of hand-picked data choices and unverified assumptions about caption accuracy and benchmark separation. There are no new physical entities. The main free parameters are curation thresholds and training hyperparameters.

free parameters (4)
  • MR40k population threshold = <5,000 inhabitants
    Manual cut used to define 'sparsely populated' for benchmark sampling; not justified empirically. Affects which images enter the new benchmark.
  • LoRA rank and alpha = r=32, alpha=32
    Chosen after tuning over r=16 and r=256 and several learning rates; affects final accuracy and is part of the method recipe.
  • Number of SFT examples = ~2,700
    Selection criteria for these images are not specified; the count is presented as a key efficiency claim.
  • Learning rate = 8e-6 (Gemma), tuned
    Selected from {1e-5, 2e-5, 3e-5, 8e-6} to maximize GWS15K and MR40k accuracy; a hyperparameter, not a physical constant.
assumptions (3)
  • domain assumption Mapillary GPS metadata is accurate and the images are representative of global street scenes.
    Relied on throughout for both SFT data and MR40k construction (Sections 3.1 and 3.3).
  • ad hoc to paper Claude 3.7 Sonnet's geo-captions are geographically accurate and free of hallucination.
    The entire SFT supervision depends on this; no validation of caption accuracy is reported.
  • ad hoc to paper MR40k is a fair held-out benchmark, disjoint from the SFT training images.
    Assumed disjointness from the 2,700 SFT images is never stated (Sections 3.2 vs 3.3).

how reviews work

0 comments
Cite this review

Pith. "Pith review of GeoLocSFT: Efficient Visual Geolocation via Supervised Fine-Tuning of Multimodal Foundation Models." pith.science (2026). https://pith.science/paper/EVBCKCFK

@misc{pith2026250601277,
  author       = {Pith},
  title        = {Pith review of: GeoLocSFT: Efficient Visual Geolocation via Supervised Fine-Tuning of Multimodal Foundation Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EVBCKCFK}},
  note         = {Machine review of arXiv:2506.01277}
}
read the original abstract

Accurately determining the geographic location where a single image was taken, visual geolocation, remains a formidable challenge due to the planet's vastness and the deceptive similarity among distant locations. We introduce GeoLocSFT, a framework that demonstrates how targeted supervised fine-tuning (SFT) of a large multimodal foundation model (Gemma 3) using a small, high-quality dataset can yield highly competitive geolocation performance. GeoLocSFT is trained with only 2700 carefully selected image-GPS pairs from our geographically diverse MR600k dataset. Despite this limited data, our SFT-centric approach substantially improves over baseline models and achieves robust results on standard benchmarks such as Im2GPS-3k and YFCC-4k, as well as on our newly proposed and challenging MR40k benchmark, aimed specifically at sparsely populated regions. Further, we explore multi-candidate inference and aggregation strategies but find that the core gains are already realized at the SFT stage. Our findings highlight the power of high-quality supervision and efficient SFT for planet-scale image geolocation, especially when compared to prior methods that require massive databases or complex pipelines. To foster further research, we publicly release the MR40k benchmark dataset.

Figures

Figures reproduced from arXiv: 2506.01277 by the authors.

Figure 1
Figure 1. Geolocation approach comparison. Traditional methods (Top) struggle with semantic [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Generation pipeline for high-quality SFT "geo-captions". Step 1: Image/metadata extraction. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Geographic distribution of the 2700 SFT training images, detailing country/region specifics and continental proportions. Explicit Reasoning for Disambiguation: A crucial component of our prompting strategy requires the LLM to perform comparative anal￾ysis by identifying potentially visually similar regions and then articulating specific, verifiable reasons (based on micro-features) why the in￾put image cannot belong… view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Conceptual illustration of the Supervised Fine-tuning data preparation process. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Left: Examples from prior benchmarks often focusing on landmarks or urban areas. Right: MR40k examples showcasing street-level views from sparsely populated regions. Curated from our MR600k dataset, MR40k (40,000 images) was constructed using a tar￾geted sampling strat…
Figure 6
Figure 6. Figure 6: Generic prompt and baseline model evaluation setup. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Prediction error across K = 10 at￾tempts. GeoLocSFT models (e.g., Gemma 27B-SFT as ’our_model’) show lower vari￾ance. Our GeoLocSFT approach consistently enhances foundation model performance. Beyond aggregate accuracy, the nature of predictions is also refined by SFT …
Figure 8
Figure 8. Figure 8: Target Image: Former Executive Yuan at the Nanjing Presidential Palace. True Latitude [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Idealized analytical reasoning for the target in Figure 8, simulating LLM-like analysis to [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]
Figure 10
Figure 10. Figure 10: Actual Top 5 predictions from the GeoCLIP model [ [PITH_FULL_IMAGE:figures/full_fig_p014_10.png]
Figure 11
Figure 11. Figure 11: MCR success case: Windmill image with prediction list. The red starred entry indicates the [PITH_FULL_IMAGE:figures/full_fig_p017_11.png]
Figure 12
Figure 12. Figure 12: MCR challenge: Landmark confusion (Geese image with prediction list). True location: [PITH_FULL_IMAGE:figures/full_fig_p017_12.png]
Figure 13
Figure 13. Figure 13: MCR challenge: Input ambiguity (Food image with prediction list). True location: Athens, [PITH_FULL_IMAGE:figures/full_fig_p018_13.png]
Figure 14
Figure 14. Figure 14: GeoLocSFT training loss comparison (1 vs. 3 epochs) showing diminishing returns after the first epoch [PITH_FULL_IMAGE:figures/full_fig_p021_14.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. DisasterTD: Disaster Toponym Disambiguation Using Multimodal LLMs and Cross-View Geolocalization

    cs.CV 2026-07 conditional novelty 5.0 of 10

    Combining MLLM candidate place extraction with DINOv2 cross-view matching of social-media, street-view, and satellite images raises disaster photo geolocation accuracy to 71.6% within 1 km on Hurricane Harvey data.

  2. GDGS: 3D Gaussian Splatting Via Geometry-Guided Initialization And Dynamic Density Control

    cs.CV 2025-07 conditional novelty 4.0 of 10

    A 3DGS variant that adds MLP initialization, normal alignment, and region-aware density control reports consistent but modest quality gains over vanilla 3DGS on three standard benchmarks.

  3. Building Lightweight Semantic Segmentation Models for Aerial Images Using Dual Relation Distillation

    eess.IV 2025-06 conditional novelty 4.0 of 10

    A student segmentation network trained with spatial and channel relation distillation from a PSPNet ResNet101 teacher gains about 3 to 5 mIoU points on Vaihingen, Potsdam, and Cityscapes.

  4. A Global-Local Cross-Attention Network for Ultra-high Resolution Remote Sensing Image Semantic Segmentation

    cs.CV 2025-06 reject novelty 3.0 of 10

    GLCANet is a dual-branch global-local attention network that reports top mIoU on DeepGlobe, Vaihingen, and Potsdam, but the method and experiments are internally inconsistent and lack code.

Reference graph

Works this paper leans on

39 extracted references · 25 canonical work pages · cited by 4 Pith papers

  1. [1]

    Learning transferable visual models from natural language supervision.arXiv preprint arXiv:2103.00020, 2021

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agar- wal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision.arXiv preprint arXiv:2103.00020, 2021. URLhttps://arxiv.org/abs/2103.00020

    arXiv 2021

  2. [2]

    James Hays and Alexei A. Efros. Im2GPS: Estimating geographic information from a single image. In2008 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 1–8. IEEE, 2008

    work page 2008

  3. [3]

    PlaNet - photo geolocation with con- volutional neural networks

    Tobias Weyand, Ilya Kostrikov, and James Philbin. PlaNet - photo geolocation with con- volutional neural networks. In Bastian Leibe, Jiri Matas, Nicu Sebe, and Max Welling, editors,Computer Vision – ECCV 2016, volume 9905 ofLecture Notes in Computer Sci- ence, pages 37–55. Springer International Publishing, 2016. ISBN 978-3-319-46484-8. doi: 10.1007/978-3...

    doi:10.1007/978-3-319-46484-8_3 2016

  4. [4]

    PIGEON: Predicting Image Geolocations

    Lukas Haas, Michal Skreta, Silas Alberti, and Chelsea Finn. PIGEON: Predicting image geolocations.arXiv preprint arXiv:2307.05845, 2024. URL https://arxiv.org/abs/2307. 05845

    work page Pith review arXiv 2024

  5. [6]

    Image-based geolocation using large vision-language models

    Yi Liu, Junchen Ding, Gelei Deng, Yuekang Li, Tianwei Zhang, Weisong Sun, Yaowen Zheng, Jingquan Ge, and Yang Liu. Image-based geolocation using large vision-language models. arXiv preprint arXiv:2408.09474, 2024. URLhttps://arxiv.org/abs/2408.09474

    arXiv 2024

  6. [7]

    Chain-of-thought prompting elicits reasoning in large language models.arXiv preprint arXiv:2201.11903, 2023

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. Chain-of-thought prompting elicits reasoning in large language models.arXiv preprint arXiv:2201.11903, 2023. URL https://arxiv.org/abs/2201. 11903

    arXiv 2023

  7. [8]

    Self-consistency improves chain of thought reasoning in language models.arXiv preprint arXiv:2203.11171, 2023

    Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. Self-consistency improves chain of thought reasoning in language models.arXiv preprint arXiv:2203.11171, 2023. URL https://arxiv.org/abs/2203. 11171

    arXiv 2023

  8. [9]

    Neural network ensembles.IEEE Transactions on Pattern Analysis and Machine Intelligence, 12(10):993–1001, 1990

    Lars Kai Hansen and Peter Salamon. Neural network ensembles.IEEE Transactions on Pattern Analysis and Machine Intelligence, 12(10):993–1001, 1990

    work page 1990

Show all 39 references
  1. [10]

    Shamma, Gerald Friedland, Benjamin Elizalde, Karl Ni, Douglas Poland, Damian Borth, and Li-Jia Li

    Bart Thomee, David A. Shamma, Gerald Friedland, Benjamin Elizalde, Karl Ni, Douglas Poland, Damian Borth, and Li-Jia Li. YFCC100M: The new data in multimedia research. Communications of the ACM, 59(2):64–73, 2016. doi: 10.1145/2812802

    2016 doi

  2. [11]

    Where we are and what we’re looking at: Query based worldwide image geo-localization using hierarchies and scenes.arXiv preprint arXiv:2303.04249, 2023

    Brandon Clark, Alec Kerrigan, Parth Parag Kulkarni, Vicente Vivanco Cepeda, and Mubarak Shah. Where we are and what we’re looking at: Query based worldwide image geo-localization using hierarchies and scenes.arXiv preprint arXiv:2303.04249, 2023. URLhttps://arxiv. org/abs/2303.04249

    2023 arXiv

  3. [12]

    The Mapil- lary Vistas dataset for semantic understanding of street scenes

    Gerhard Neuhold, Tobias Ollmann, Samuel Rota Bulò, and Peter Kontschieder. The Mapil- lary Vistas dataset for semantic understanding of street scenes. InProceedings of the IEEE International Conference on Computer Vision (ICCV), pages 4990–4999. IEEE, 2017

    2017

  4. [13]

    GeoNames geographical database

    GeoNames. GeoNames geographical database. https://www.geonames.org, 2025. Ac- cessed: 2025-01-01

    2025

  5. [15]

    Qwen2.5-VL technical report

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, Jiabo Ye, Xi Zhang, Tianbao Xie, Zesen Cheng, Han...

    2025 arXiv

  6. [16]

    Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2019

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2019. URLhttps://arxiv.org/abs/1711.05101

    2019 arXiv

  7. [17]

    OpenStreetView-5M: The many roads to global visual geolocation.arXiv preprint arXiv:2404.18873, 2024

    Guillaume Astruc, Nicolas Dufour, Ioannis Siglidis, Constantin Aronssohn, Nacim Bouia, Stephanie Fu, Romain Loiseau, Van Nguyen Nguyen, Charles Raude, Elliot Vincent, Lintao XU, Hongyu Zhou, and Loic Landrieu. OpenStreetView-5M: The many roads to global visual geolocation.arXi...

    2024 arXiv

  8. [18]

    The Claude 3 model family: Opus, Sonnet, Haiku

    Anthropic. The Claude 3 model family: Opus, Sonnet, Haiku. https://www.anthropic. com/news/claude-3-7-sonnet, 2025

    2025

  9. [19]

    Visual instruction tuning.arXiv preprint arXiv:2304.08485, 2023

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning.arXiv preprint arXiv:2304.08485, 2023. URLhttps://arxiv.org/abs/2304.08485

    2023 arXiv

  10. [20]

    Mistral-Small-3.1-24B-Instruct-2503

    Mistral AI. Mistral-Small-3.1-24B-Instruct-2503. Hugging Face model mistralai/Mistral- Small-3.1-24B-Instruct-2503, March 2025. URL https://huggingface.co/mistralai/ Mistral-Small-3.1-24B-Instruct-2503

    2025

  11. [21]

    Revisiting IM2GPS in the deep learning era.arXiv preprint arXiv:1705.04838, 2017

    Nam V o, Nathan Jacobs, and James Hays. Revisiting IM2GPS in the deep learning era.arXiv preprint arXiv:1705.04838, 2017. URLhttps://arxiv.org/abs/1705.04838

    2017 arXiv

  12. [23]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-rank adaptation of large language models.arXiv preprint arXiv:2106.09685, 2021. URLhttps://arxiv.org/abs/2106.09685

    2021 arXiv

  13. [24]

    Consensus- Geo

    Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, Zheyan Luo, Zhangchi Feng, and Yongqiang Ma. LlamaFactory: Unified efficient fine-tuning of 100+ language models.arXiv preprint arXiv:2403.13372, 2024. URLhttps://arxiv.org/abs/2403.13372. 12 A Qualitative Example: Nanjing ...

    2024 arXiv

  14. [25]

    URLhttps://arxiv.org/abs/2412.06781

    arXiv

  15. [28]

    Guidelines: • The answer NA means that the abstract and introduction do not include the claims made in the paper

    Claims Question: Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? Answer: [Yes] Justification: Our abstract and introduction clearly outline the three key contributions of our work: the GeoLocSFT framework, explor...

  16. [29]

    Guidelines: • The answer NA means that the paper has no limitation while the answer No means that the paper has limitations, but those are not discussed in the paper

    Limitations Question: Does the paper discuss the limitations of the work performed by the authors? Answer: [Yes] Justification: In our ablation studies, we acknowledge limitations in our approach, particularly in sampling and aggregation strategies, and identify challenges in ...

  17. [30]

    Guidelines: • The answer NA means that the paper does not include theoretical results

    Theory Assumptions and Proofs Question: For each theoretical result, does the paper provide the full set of assumptions and a complete (and correct) proof? Answer: [NA] Justification: Our paper focuses on empirical methods for visual geolocation and does not include theoretica...

  18. [31]

    The appendices include hyperparameters for SFT and information about computational resources used

    Experimental Result Reproducibility Question: Does the paper fully disclose all the information needed to reproduce the main experimental results of the paper to the extent that it affects the main claims and/or conclusions of the paper (regardless of whether the code and data...

  19. [32]

    Our appendices provide detailed instructions regarding implementation, hyperparameters, and experimental setup to facilitate reproduction

    Open access to data and code Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: We will publicly release our ...

  20. [33]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experimental Setting/Details Question: Does the paper specify all the training and test details (e.g., data splits, hyperparameters, how they were chosen, type of optimizer, etc.) necessary to understand the results? Answer: [Yes] Justification: Our experimental section and ap...

  21. [34]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experiment Statistical Significance Question: Does the paper report error bars suitably and correctly defined or other appropriate information about the statistical significance of the experiments? Answer: [No] Justification: While we do not report error bars, we perform compr...

  22. [35]

    We also report model sizes and memory requirements

    Experiments Compute Resources Question: For each experiment, does the paper provide sufficient information on the computer re- sources (type of compute workers, memory, time of execution) needed to reproduce the experiments? Answer: [Yes] Justification: We specify that our SFT...

  23. [36]

    We use publicly available datasets, acknowledge relevant prior work, and are transparent about our methodologies

    Code Of Ethics Question: Does the research conducted in the paper conform, in every respect, with the NeurIPS Code of Ethics https://neurips.cc/public/EthicsGuidelines? 26 Answer: [Yes] Justification: Our research adheres to the NeurIPS Code of Ethics. We use publicly availabl...

  24. [37]

    Guidelines: • The answer NA means that there is no societal impact of the work performed

    Broader Impacts Question: Does the paper discuss both potential positive societal impacts and negative societal impacts of the work performed? Answer: [Yes] Justification: We acknowledge that visual geolocation has positive applications such as photo organization and journalis...

  25. [38]

    27 Guidelines: • The answer NA means that the paper poses no such risks

    Safeguards Question: Does the paper describe safeguards that have been put in place for responsible release of data or models that have a high risk for misuse (e.g., pretrained language models, image generators, or scraped datasets)? Answer: [NA] Justification: Our research fo...

  26. [39]

    Guidelines: • The answer NA means that the paper does not use existing assets

    Licenses for existing assets Question: Are the creators or original owners of assets (e.g., code, data, models), used in the paper, properly credited and are the license and terms of use explicitly mentioned and properly respected? Answer: [Yes] Justification: We properly cred...

  27. [40]

    This documentation will be released alongside the dataset

    New Assets Question: Are new assets introduced in the paper well documented and is the documentation provided alongside the assets? Answer: [Yes] Justification: Our MR40k benchmark dataset is thoroughly documented, including details on data collection, filtering criteria, geog...

  28. [41]

    Guidelines: • The answer NA means that the paper does not involve crowdsourcing nor research with human subjects

    Crowdsourcing and Research with Human Subjects Question: For crowdsourcing experiments and research with human subjects, does the paper include the full text of instructions given to participants and screenshots, if applicable, as well as details about compensation (if any)? A...

  29. [42]

    Guidelines: • The answer NA means that the paper does not involve crowdsourcing nor research with human subjects

    Institutional Review Board (IRB) Approvals Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals (or an equivalent approval/review based on the...

  30. [2024]

    URLhttps://arxiv.org/abs/2405.14702

    arXiv

  31. [2025]

    arXiv preprint arXiv:2503.19786

    URLhttps://arxiv.org/abs/2503.19786. arXiv preprint arXiv:2503.19786. 11

    arXiv

Pith tools

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