Pith. sign in

REVIEW 3 major objections 6 minor 43 references

AstroLoc: Robust Space to Ground Image Localizer

T0 review · 3 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read Training on 221k automatically footprinted astronaut photos gives a retrieval model a 35% average recall@1 gain over prior state of the art in astronaut photography localization, with recall@100 above 99%.

desk verdict Solid, valuable APL contribution; the -L and historical claims are weakened by shared automatic labels, and the 35% headline is not backed by the numbers. read the letter →

arxiv 2502.07003 v2 pith:ZIH3BQDM submitted 2025-02-10 cs.CV

Gabriele Berton , Alex Stoken , Carlo Masone This is my paper · ORCID
classification cs.CV
keywords AstronautPhotographyLocalizationImageRetrievalCross-domainSatelliteimageryUnsupervisedminingVisualplacerecognitionFootprintestimationSpace-to-ground
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

Astronaut photographs from the ISS are a rich, manually unlocalized record of Earth, but previous retrieval-based localizers were trained only on satellite imagery, ignoring the millions of open-source astronaut photos. AstroLoc is the first pipeline to train on astronaut photos themselves: it automatically estimates the ground footprint of 300k weakly labeled photos using a feature-matching pipeline, pairs them with overlapping satellite tiles, and trains a retrieval model with two complementary losses—a pairwise cross-domain loss and a cluster-mining loss that samples satellite imagery according to the geographic distribution of astronaut photos. The paper reports a 35% average improvement in recall@1 over prior state of the art, recall@100 above 99% on existing test sets, and strong results on new, harder test sets that include small-area photos. It also reports transfer without fine-tuning to lost-in-space satellite orbit determination and to 40-year-old Space Shuttle film imagery.

What carries the argument

The central mechanism is the training data and objective combination: (1) an automated footprint estimation pipeline that converts weak center-point labels into full four-corner footprints for 221k astronaut photos, yielding 865k query–satellite training pairs with IoU > 0.2; (2) a pairwise contrastive loss over these cross-domain pairs; and (3) the MUM loss, which k-means clusters the satellite database into K=50 clusters in feature space, weights clusters by how many astronaut query features fall into them, and applies a Multi-Similarity loss on quadruplets sampled from a cluster. The weighting is what makes the satellite-only loss focus on the visual environments astronauts actually photograph (glaciers, volcanoes, coasts) rather than uniformly sampling featureless oceans and deserts.

What would settle it

Compare the automated footprints against human-verified corner coordinates on a held-out set of astronaut photos; if the recovered footprints show systematic spatial error (e.g., a consistent shift toward the center of the weak label or a consistent rotation), then the training pairs and the -L test labels are constructed from the same biased source, and retrieval accuracy measured on human-verified queries would drop noticeably below the reported recall@1 and recall@100.

Watch

Extended reading notes

Core claim

The central claim is that astronaut photos can and should be used as training data for the space-to-ground image retrieval task, rather than only satellite imagery. The paper argues that the previously untapped 300k manually weakly labeled astronaut photos, once their full footprints are recovered by an automated matching pipeline (SuperPoint + LightGlue + EarthMatch), provide the missing supervision for cross-domain retrieval. With a pairwise loss that pulls matching astronaut–satellite pairs together while pushing apart geographically disjoint pairs, plus a 'Multi-similarity with Unsupervised Mining' loss that clusters the entire satellite database and samples clusters according to the distribution of astronaut queries, AstroLoc learns an Earth-surface representation that outperforms all prior methods on the standard APL benchmarks, saturates them at recall@100 > 99%, and extends to new, more realistic test sets covering the full range of query areas. The same model, without fine-tuning, achieves strong results on lost-in-space satellite localization and historical Space Shuttle imagery.

Load-bearing premise

The whole pipeline rests on the assumption that the automated footprint estimation (SuperPoint + LightGlue + EarthMatch) is accurate enough that the 865k training pairs and the new -L test-set labels are both correct; if those footprints are systematically biased, the training and test labels share the same error and the reported recalls could be inflated.

Editorial extensions

If this is right

  • Existing APL test sets are effectively saturated at recall@100 > 99%; the new -L test sets covering the full range of query areas should serve as the standard for future evaluation.
  • A single retrieval model can handle astronaut photography localization, lost-in-space orbit determination, and historical Space Shuttle imagery without task-specific fine-tuning.
  • The unsupervised-mining objective offers a way to use an unlabeled query distribution to mine a larger unlabeled database for contrastive training, a technique that could be reused in other cross-domain retrieval problems.
  • The model is already deployed at scale: it has localized hundreds of thousands of astronaut photos, and the paper expects the backlog of unlocalized ISS imagery to be nearly cleared within months.

Reading between the lines

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

  • Because the new -L test sets are labeled by the same automated footprint pipeline that generates the training pairs, an independent human-verified footprint benchmark is needed to rule out the possibility that training and test labels share a systematic bias that inflates the reported recalls.
  • The unsupervised-mining principle—weighting clusters of a large unlabeled database by the distribution of a query stream—could transfer to other cross-domain retrieval settings, such as UAV-view queries over satellite maps, where the drone's flight path defines the weighting.
  • The footprint pipeline succeeds on 221k of 300k photos; the roughly 79k failures (cloud occlusion, horizon shots, label errors) form a hard tail that a pure retrieval model may never localize, so a verification stage like EarthMatch appears necessary for full-coverage deployment.
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 / 6 minor

Summary. The paper proposes AstroLoc, an image retrieval model for Astronaut Photography Localization (APL). The authors first estimate footprints for 221k astronaut photos via an automated matching pipeline (SuperPoint + LightGlue + EarthMatch), pair these photos with overlapping satellite tiles to form 865k training pairs, and train a DINOv2-SALAD-based model with a pairwise cross-domain loss and a new 'unsupervised mining' Multi-Similarity loss. Experiments report large gains over prior methods on the six original EarthLoc test sets, on newly introduced extended '-L' test sets, on a lost-in-space satellite localization task, on historical Space Shuttle imagery, and on a worldwide-search variant. The paper claims a 'staggering 35% average improvement in recall@1 over previous SOTA' and recall@100 consistently above 99% on existing datasets.

Significance. If the results hold, the paper makes a strong practical contribution: it is the first APL method to exploit astronaut photos at training time, and the reported gains on the original EarthLoc test sets (e.g., R@1 rising from about 80 to 96 on Texas, and similar gains elsewhere) are compelling evidence that the approach is effective. The lost-in-space results on the independent VINSat dataset are particularly encouraging, showing large improvements over strong baselines with an external label source. The unsupervised-mining formulation is a reasonable and reusable idea. However, the significance is currently tempered by two issues: the headline '35%' claim is not reproducible from the tables, and the newly introduced -L test sets and the historical imagery evaluation rely on the same automated footprint pipeline used to generate training labels, which risks inflating measured recall through shared label bias. These issues are fixable and do not undermine the core training idea, but they must be resolved before the broader claims are accepted.

major comments (3)
  1. [§3.1, §3.2, §7.2] The automated footprint pipeline (SuperPoint + LightGlue + EarthMatch) is used to generate the 865k training pairs (Sec. 3.1) and also to define correctness on the newly proposed -L test sets (Sec. 3.2) and on the historical Space Shuttle evaluation (Sec. 7.2: 'We first precisely localize 704 images with the pipeline described in Sec. 3.1'). No accuracy of these footprints against human-verified labels is reported. If EarthMatch has systematic bias, the training supervision and the evaluation labels are aligned, so the recall numbers on -L and historical sets can be inflated. The original EarthLoc test sets (Tab. 2) use external human-derived labels and therefore are not affected by this circularity; the large gains there are credible. But the -L results in Tab. 3 and the historical results in Tab. 5 are not yet evidence of real localization capability until the footprint accuracy is independently quantified. Please report a human-verified evaluation of a random sample of the 221k footprints (e.g., IoU against manually drawn footprints) and, if the bias is non-negligible, re-evaluate the -L and historical sets with independent labels.
  2. [Abstract, Tab. 2-3] The abstract's claim of a 'staggering 35% average improvement in recall@1 over previous SOTA' is not reproducible from the reported tables. Computing from Tab. 2, the average relative R@1 gain of AstroLoc over the strongest baseline EarthLoc++ is approximately 22%, and the average absolute gain is about 19 percentage points. No aggregation in Tabs. 2 or 3 yields 35%. Please specify exactly which baseline and which aggregation (relative vs. absolute, which test sets) the 35% figure refers to, or correct the claim.
  3. [Abstract, §3.1] There is an internal inconsistency about the scale of the annotated data. The abstract states that the authors 'produce full localization information for 300,000 manually weakly labeled astronaut photos', but Sec. 3.1 reports that the automated method was successful for only 221k queries. The introduction also says 'produce a precise annotation of these 300,000 photos'. Please reconcile these numbers and state clearly how many astronaut photos have estimated footprints and how many training pairs were actually used.
minor comments (6)
  1. [§3.2] The construction of the new -L test sets is under-specified: the paper does not state explicitly whether the ground-truth footprints for queries in Texas-L, Alps-L, etc. come from the automated pipeline of Sec. 3.1 or from independent manual labeling. Please state the label source explicitly in the main text.
  2. [§5.1, Tabs. 2-7] No error bars or multiple-seed results are reported for the main experiments. Since the gains are large, this is not critical, but reporting the variance across at least three seeds would strengthen the claims, especially for the ablation in Tab. 6.
  3. [§4.3, Eq. (7)] The notation 'k ∼ B(Q, 1, k)' in Eq. (7) is unconventional and appears to be a typo for 'k ∼ B(b_1,...,b_K)' or similar. Please define the weighted distribution clearly.
  4. [§7 (supplementary), Fig. 6-8] In the qualitative results, correct predictions are defined as those that have 'any overlap with the query' footprint. This is a very loose criterion; a prediction that overlaps by a tiny sliver would be counted as correct. Please report the IoU threshold used to define correctness in the qualitative evaluation or in the quantitative protocol.
  5. [§3.1] The sentence about rotating potential positives by 90°, 180° and 270° is ambiguous: it should clarify whether the satellite tiles are rotated in the matching step and how the final footprint is derived from the rotation that yields the best match.
  6. [§4.3] The claim that Unsupervised Mining is 'the first mining technique' of its kind is strong; please soften it to 'to the best of our knowledge' in the main text (the phrase appears in the intro but not in Sec. 4.3).

Circularity Check

2 steps flagged · score 6.0 of 10

Partial circularity: the new -L and historical test sets score retrieval against the same automated EarthMatch footprints that generated the training pairs, so those headline gains partly measure agreement with the label generator.

  1. self definitional [Sec. 3.1-3.2 (Tables 2-3)]
    "Sec. 3.1: 'We then perform image matching with SuperPoint [10] + LightGlue [22] and the EarthMatch pipeline [5] to get the footprint coordinates of each query.' ... 'we pair each query with the all database images with an IoU over tiou = 0.2 producing 865k query-database training pairs.' Sec. 3.2: 'We therefore propose new evaluation sets which include all available geolocated queries within an evaluation area.'"

    The IoU>0.2 threshold that defines a positive training pair is computed from footprints produced by the Sec. 3.1 EarthMatch pipeline, and those footprints are the only full-localization labels described in the paper (weak labels are single points). The -L test sets are then built from 'all available geolocated queries' and scored by whether a retrieved tile overlaps the query footprint, so the test correctness criterion is the same automatically estimated footprint that generated the training supervision. If EarthMatch has systematic bias, AstroLoc can learn to retrieve tiles overlapping the biased footprints and will be scored as correct by those same biased footprints.

  2. self definitional [Supplementary Sec. 7.2-7.3, Tab. 5]
    "'We first precisely localize 704 images with the pipeline described in Sec. 3.1, and then compute localization results, reported in Tab. 5.' Sec. 7.3: 'Correct predictions, defined as those that have any overlap with the query, are outlined in green.'"

    The historical Space Shuttle queries are localized with the same SuperPoint+LightGlue+EarthMatch pipeline used to create the 865k training pairs, and the Tab. 5 recall is computed against those pipeline footprints using the same overlap criterion. The model was trained to bring database tiles with IoU>0.2 on those footprints closer to astronaut photos, so the historical experiment predominantly measures consistency with the label generator rather than independently verified localization of 1981-1984 film photos. Without human-verified footprints for the 704 queries, the reported 82.0 R@1 is not independent transfer evidence.

full rationale

The original EarthLoc evaluation sets (Tab. 2) are a genuinely independent checkpoint: they come from prior work, use externally defined labels, and the paper removes training queries that appear in those sets, so the large gains there are not circular. The circularity is confined to the newly introduced -L sets and the historical Shuttle experiment. For -L, the paper never states that its footprint labels are human-verified; its only full-localization source is the Sec. 3.1 automated pipeline, and overlap-based retrieval correctness on those queries requires those footprints. For historical imagery, the supplementary explicitly says the queries were localized with the same Sec. 3.1 pipeline. Thus those R@1 numbers partly reduce to agreement between the retrieval model and the same EarthMatch label generator used for training supervision. The paper should report footprint accuracy against human-verified labels before the -L and historical claims are treated as independent. Separately, the abstract's '35% average improvement' is not reproduced from the tables (Tab. 2 shows roughly 22% average relative R@1 gain over EarthLoc++); that is a reporting/aggregation problem, not a circularity. Overall score 6: partial circularity in the new-set and transfer claims, while the central EarthLoc-benchmark result retains independent content.

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

The paper is an empirical systems contribution. The central claims rest on the accuracy of the automated footprint labels, the suitability of Sentinel-2 composited tiles as a retrieval database, the chosen evaluation protocol, and the hyperparameters of the two losses. No new physical entities are introduced.

free parameters (5)
  • number of clusters K = 50
    K-means cluster count for the MUM loss; chosen by the authors without a reported sensitivity analysis (Sec. 5.1).
  • IoU threshold t_iou = 0.2
    Threshold for building query-satellite training pairs; affects the number and quality of pairs (865k pairs, Sec. 3.1).
  • loss gains alpha1, beta1, alpha2, beta2 = 1, 50, 1, 50
    Contrastive loss hyperparameters in Eqs. 1, 3, and 9; set without reported sensitivity analysis (Sec. 5.1).
  • lambda1, lambda2 = 1, 1
    Weights for the two losses in Eq. 10; no ablation on the weighting is shown (Sec. 5.1).
  • zoom levels = 8-12
    Database tile zoom levels, extended from EarthLoc to cover a wider range of query extents (Sec. 3).
assumptions (4)
  • domain assumption The EarthMatch pipeline (SuperPoint + LightGlue) produces sufficiently accurate footprints to define training pairs and test-set labels.
    All training pairs and the -L and historical test labels depend on this; no quantitative accuracy check is reported (Sec. 3.1 and supplementary Sec. 7.2).
  • domain assumption Sentinel-2 yearly composites (2018-2021) are an adequate database for localizing astronaut photos taken between 2000 and 2024.
    The database is fixed to S2 tiles at zoom 8-12; temporal and seasonal domain gaps are not studied (Tab. 1, Sec. 3).
  • domain assumption Recall@N with IoU-overlapping tiles as ground truth is a valid proxy for real-world APL success.
    Evaluations use this protocol from EarthLoc; the paper also notes post-processing with EarthMatch is common in practice (Sec. 5.2).
  • domain assumption The manually provided weak labels (single lat/lon) are accurate enough to seed footprint estimation.
    Errors in manual labels are cited as one cause of footprint failure (Sec. 3.1), so their accuracy directly impacts the training set.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AstroLoc: Robust Space to Ground Image Localizer." pith.science (2026). https://pith.science/paper/ZIH3BQDM

@misc{pith2026250207003,
  author       = {Pith},
  title        = {Pith review of: AstroLoc: Robust Space to Ground Image Localizer},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZIH3BQDM}},
  note         = {Machine review of arXiv:2502.07003}
}
read the original abstract

Astronauts take thousands of photos of Earth per day from the International Space Station, which, once localized on Earth's surface, are used for a multitude of tasks, ranging from climate change research to disaster management. The localization process, which has been performed manually for decades, has recently been approached through image retrieval solutions: given an astronaut photo, find its most similar match among a large database of geo-tagged satellite images, in a task called Astronaut Photography Localization (APL). Yet, existing APL approaches are trained only using satellite images, without taking advantage of the millions open-source astronaut photos. In this work we present the first APL pipeline capable of leveraging astronaut photos for training. We first produce full localization information for 300,000 manually weakly labeled astronaut photos through an automated pipeline, and then use these images to train a model, called AstroLoc. AstroLoc learns a robust representation of Earth's surface features through two losses: astronaut photos paired with their matching satellite counterparts in a pairwise loss, and a second loss on clusters of satellite imagery weighted by their relevance to astronaut photography via unsupervised mining. We find that AstroLoc achieves a staggering 35% average improvement in recall@1 over previous SOTA, pushing the limits of existing datasets with a recall@100 consistently over 99%. Finally, we note that AstroLoc, without any fine-tuning, provides excellent results for related tasks like the lost-in-space satellite problem and historical space imagery localization.

Figures

Figures reproduced from arXiv: 2502.07003 by the authors.

Figure 1
Figure 1. One model, many space to ground applications. We train a single model, AstroLoc, that succeeds in multiple space￾based image retrieval settings: astronaut photography localization, “lost in space” orbit determination, and historical (Space Shuttle) photography localization. In this figure, each group of 3 images represent a query and its top-2 predictions from searching over a worldwide database of millions of satel… view at source ↗
Figure 2
Figure 2. Visual example of weak (manual) annotation and full annotation of an astronaut photo. Weak annotation is the geo￾graphic coordinates of a single point, which does not provide in￾formation about the image’s size (i.e. it could cover a town or an entire continent), whereas full annotation provides coordinates for all 4 corners (called footprint), from which the coordinate of any pixel within the image can be easily ca… view at source ↗
Figure 3
Figure 3. Distribution of queries by covered area. The red mark at 5,000 sqkm shows that 78% of astronaut photographs cover an area lower than 5,000 sqkm. Thus, the test sets in Tab. 2 do not contain this vast majority of queries, leading us to propose test sets containing all queries used for experiments in Tab. 3. These datasets were named after the geographic location of their center, like Texas, Gobi, and Amazon, so we ca… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: AstroLoc’s training pipeline. The upper branch feeds pairs of matching query-database images to the pairwise loss (Sec. 4.2). The lower branch (Sec. 4.3) first creates clusters of satellite images, then queries are assigned to these clusters, which are sampled accordin…
Figure 5
Figure 5. Figure 5: Examples of training batches, using the three different sampling solutions presented. For each solution, we show two examples of batches with batch size 12 (i.e. 3 quadruplets), so that each image has 3 positives and 8 negatives. Solution 1 leads to training a non-robu…
Figure 6
Figure 6. Figure 6: Qualitative examples from the Amazon-L test set. Each triplet shows one query and its top-2 predictions, red if wrong and green if correct. mass, can help alleviate the mostly water retrieval results. Further hard negative mining may help disambiguate similar forested …
Figure 7
Figure 7. Figure 7: Qualitative examples from the historical Space Shuttle imagery. Each triplet shows one query and its top-2 predictions, red if wrong and green if correct. The queries were taken with analog cameras between 1981 and 1984 and then later digitized [PITH_FULL_IMAGE:figure…
Figure 8
Figure 8. Figure 8: Qualitative examples from the VINSat dataset [23]. Each triplet shows one query and its top-2 predictions, red if wrong and green if correct. Queries are mosaics of Sentinel 2 imagery. Worth 16x16 Words: Transformers for Image Recognition at Scale. ArXiv, abs/2010.1192…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 39 canonical work pages

  1. [1]

    Global proxy-based hard mining for visual place recognition

    Amar Ali-Bey, Brahim Chaib-draa, and Philippe Giguere. Global proxy-based hard mining for visual place recognition. In 33rd British Machine Vision Conference 2022, BMVC 2022, London, UK, November 21-24, 2022 . BMV A Press,

    work page 2022

  2. [2]

    Gsv-cities: Toward appropriate supervised visual place recognition

    Amar Ali-bey, Brahim Chaib-draa, and Philippe Gigu `ere. Gsv-cities: Toward appropriate supervised visual place recognition. Neurocomputing, 513:194–203, 2022. 6

    work page 2022

  3. [3]

    NetVLAD: CNN architecture for weakly supervised place recognition

    Relja Arandjelovi ´c, Petr Gronat, Akihiko Torii, Tomas Pa- jdla, and Josef Sivic. NetVLAD: CNN architecture for weakly supervised place recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence , 40(6):1437– 1451, 2018. 7

    work page 2018

  4. [4]

    Geography-aware self-supervised learning

    Kumar Ayush, Burak Uzkent, Chenlin Meng, Kumar Tan- may, Marshall Burke, David Lobell, and Stefano Ermon. Geography-aware self-supervised learning. IEEE Interna- tional Conference on Computer Vision, 2021. 8

    work page 2021

  5. [5]

    Earthmatch: Iterative coregistration for fine-grained localization of astro- naut photography

    Gabriele Berton, Gabriele Goletto, Gabriele Trivigno, Alex Stoken, Barbara Caputo, and Carlo Masone. Earthmatch: Iterative coregistration for fine-grained localization of astro- naut photography. In Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR) Workshops, 2024. 2, 4, 7

    work page 2024

  6. [6]

    Earthloc: Astronaut photography localization by indexing earth from space

    Gabriele Berton, Alex Stoken, Barbara Caputo, and Carlo Masone. Earthloc: Astronaut photography localization by indexing earth from space. InIEEE Conference on Computer Vision and Pattern Recognition, 2024. 2, 3, 4, 6, 7, 8

    work page 2024

  7. [7]

    Mollie Bianchi and Timothy D. Barfoot. Uav localiza- tion using autoencoded satellite images. arXiv preprint arXiv:2102.05692, 2021. 3

    work page Pith review arXiv 2021

  8. [8]

    Drone Referring Localization: An Efficient Heterogeneous Spatial Feature Interaction Method For UAV Self-Localization

    Ming Dai, Enhui Zheng, Zhenhua Feng, Jiahao Chen, and Wankou Yang. Finding point with image: A simple and efficient method for uav self-localization. arXiv preprint arXiv:2208.06561, 2022. 3

    work page Pith review arXiv 2022

Show all 43 references
  1. [9]

    Alejandro S ´anchez de Miguel, Jonathan Bennie, Emma Rosenfeld, Simon Dzurjak, and Kevin J. Gaston. Environ- mental risks from artificial nighttime lighting widespread and increasing across europe. Science Advances , 8(37): eabl6891, 2022. 1, 3

    2022

  2. [10]

    Superpoint: Self-supervised interest point detection and description

    Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabi- novich. Superpoint: Self-supervised interest point detection and description. In Proceedings of the IEEE conference on computer vision and pattern recognition workshops , pages 224–236, 2018. 4

    2018

  3. [11]

    Orientation-guided contrastive learning for uav- view geo-localisation

    Fabian Deuser, Konrad Habel, Martin Werner, and Norbert Oswald. Orientation-guided contrastive learning for uav- view geo-localisation. In Proceedings of the 2023 Workshop on UAVs in Multimedia: Capturing the World from a New Perspective, page 7–11, New York, NY , USA, 2023. A...

    2023

  4. [12]

    An Image is 2 Figure 7

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An Image is 2 Figure 7. Qualitative examples from the historical S...

    1981 arXiv

  5. [13]

    The faiss library

    Matthijs Douze, Alexandr Guzhva, Chengqi Deng, Jeff Johnson, Gergely Szilvasy, Pierre-Emmanuel Mazar´e, Maria Lomeli, Lucas Hosseini, and Herv´e J´egou. The faiss library. arXiv, 2024. 8

    2024

  6. [14]

    Gaston and Alejandro S ´anchez de Miguel

    Kevin J. Gaston and Alejandro S ´anchez de Miguel. Environ- mental impacts of artificial light at night. Annual Review of Environment and Resources, 47(1):373–398, 2022. 1, 3

    2022

  7. [15]

    Self-supervising fine-grained region similarities for large-scale image localization

    Yixiao Ge, Haibo Wang, Feng Zhu, Rui Zhao, and Hong- sheng Li. Self-supervising fine-grained region similarities for large-scale image localization. In Computer Vision – ECCV 2020 , pages 369–386, Cham, 2020. Springer Inter- national Publishing. 7 3

    2020

  8. [16]

    Foundloc: Vision-based onboard aerial localization in the wild

    Yao He, Ivan Cisneros, Nikhil Keetha, Jay Patrikar, Zelin Ye, Ian Higgins, Yaoyu Hu, Parv Kapoor, and Sebastian Scherer. Foundloc: Vision-based onboard aerial localization in the wild. arXiv preprint arXiv:2310.16299, 2023. 3

    2023 arXiv

  9. [17]

    Optimal transport aggre- gation for visual place recognition

    Sergio Izquierdo and Javier Civera. Optimal transport aggre- gation for visual place recognition. In IEEE Conference on Computer Vision and Pattern Recognition, 2024. 7

    2024

  10. [18]

    Color pictures of sprites from non-dedicated observation on board the international space station

    Augustin Jehl, Thomas Farges, and Elisabeth Blanc. Color pictures of sprites from non-dedicated observation on board the international space station. Journal of Geophysical Re- search: Space Physics, 118(1):454–461, 2013. 1, 3

    2013

  11. [19]

    Prod- uct quantization for nearest neighbor search

    Herv ´e J´egou, Matthijs Douze, and Cordelia Schmid. Prod- uct quantization for nearest neighbor search. IEEE Trans. Pattern Anal. Mach. Intell., 33(1):117–128, 2011. 1

    2011

  12. [20]

    Anyloc: Towards universal vi- sual place recognition

    Nikhil Keetha, Avneesh Mishra, Jay Karhade, Kr- ishna Murthy Jatavallabhula, Sebastian Scherer, Madhava Krishna, and Sourav Garg. Anyloc: Towards universal vi- sual place recognition. arXiv, 2023. 3, 7, 8

    2023

  13. [21]

    A transformer-based adaptive semantic aggregation method for uav visual geo-localization

    Shishen Li, Cuiwei Liu, Huaijun Qiu, and Zhaokui Li. A transformer-based adaptive semantic aggregation method for uav visual geo-localization. arXiv preprint arXiv:2401.01574, 2024. 3

    2024 arXiv

  14. [22]

    LightGlue: Local Feature Matching at Light Speed

    Philipp Lindenberger, Paul-Edouard Sarlin, and Marc Polle- feys. LightGlue: Local Feature Matching at Light Speed. In ICCV, 2023. 4

    2023

  15. [23]

    Vin- sat: Solving the lost-in-space problem with visual-inertial navigation

    Kyle McCleary, Swaminathan Gurumurthy, Paulo RM Fisch, Saral Tayal, Zachary Manchester, and Brandon Lucia. Vin- sat: Solving the lost-in-space problem with visual-inertial navigation. In IEEE International Conference on Robotics and Automation (ICRA), 2024. 3, 7, 8, 1

    2024

  16. [24]

    A met- ric learning reality check

    Kevin Musgrave, Serge Belongie, and Ser-Nam Lim. A met- ric learning reality check. In Computer Vision – ECCV 2020, pages 681–699, Cham, 2020. Springer International Publish- ing. 6

    2020

  17. [25]

    Belongie, and Ser-Nam Lim

    Kevin Musgrave, Serge J. Belongie, and Ser-Nam Lim. Py- torch metric learning. ArXiv, abs/2008.09164, 2020. 7

    2008 arXiv

  18. [26]

    Observing Earth From Space: Using Astronaut Photography to Analyze Geograph- ical Climate Patterns

    Aadil Nathani, Rishi Iyer, Annabelle Wang, Aarnav Chitari, Adele Wilson, and Hannah Norris. Observing Earth From Space: Using Astronaut Photography to Analyze Geograph- ical Climate Patterns. In AGU Fall Meeting Abstracts, pages ED44C–06, 2022. 1, 3

    2022

  19. [27]

    Zoom levels

    OpenStreetMap Contributors. Zoom levels. https:// wiki.openstreetmap.org/wiki/Zoom_levels ,

  20. [28]

    Maxime Oquab, Timoth ´ee Darcet, Theo Moutakanni, Huy V . V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Rus- sell Howes, Po-Yao Huang, Hu Xu, Vasu Sharma, Shang- Wen Li, Wojciech Galuba, Mike Rabbat, Mido Assran, Ni...

    2023

  21. [29]

    Mapping behaviorally relevant light pollution levels to im- prove urban habitat planning

    Aaron Schirmer, Caleb Gallemore, Ting Liu, Seth Magle, Elisabeth DiNello, Humerah Ahmed, and Thomas Gilday. Mapping behaviorally relevant light pollution levels to im- prove urban habitat planning. Scientific Reports, 9, 2019. 1, 3

    2019

  22. [30]

    Shockley and Robert A

    Liberty M. Shockley and Robert A. Bettinger. Real-time aerospace vehicle position estimation using terrestrial illu- mination matching. In IEEE 8th International Workshop on Metrology for AeroSpace (MetroAeroSpace),, 2021. 3

    2021

  23. [31]

    Skinner, Michael Coletti, Matthew C

    Mark A. Skinner, Michael Coletti, Matthew C. V oss, Tomas Svitek, John C. Lee, Kerstyn Auman, Hemanshu Patel, and Eamonn J. Moyer. Mitigating cubesat confusion: Results of in-flight technical demonstrations of candidate tracking and identification technologies. Journal of Spac...

    2022

  24. [32]

    Spectrometry of the urban lightscape

    Christopher Small. Spectrometry of the urban lightscape. Technologies, 10(4), 2022. 1, 3

    2022

  25. [33]

    W. L. Stefanov and C. A. Evans. Data collection for disaster response from the international space station. The Interna- tional Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences , XL-7/W3:851–855, 2015. 1, 3

    2015

  26. [34]

    Stewart, Caleb Robinson, Isaac A

    Adam J. Stewart, Caleb Robinson, Isaac A. Corley, Anthony Ortiz, Juan M. Lavista Ferres, and Arindam Banerjee. Torch- Geo: Deep learning with geospatial data. In Proceedings of the 30th International Conference on Advances in Geo- graphic Information Systems, pages 1–12, Seatt...

    2022

  27. [35]

    Find my astronaut photo: Automated localization and georectification of astronaut photography

    Alex Stoken and Kenton Fisher. Find my astronaut photo: Automated localization and georectification of astronaut photography. In CVPRW, pages 6196–6205, 2023. 2, 3, 7

    2023

  28. [36]

    (street) lights will guide you: Georeferencing night- time astronaut photography of earth

    Alex Stoken, Peter Ilhardt, Mark Lambert, and Kenton Fisher. (street) lights will guide you: Georeferencing night- time astronaut photography of earth. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 492–501, 2024. 2, 3

    2024

  29. [37]

    Christian

    Miranda Straub and John A. Christian. Autonomous opti- cal navigation for earth-observing satellites using coastline matching. In AIAA Guidance, Navigation, and Control Con- ference, 2015. 3

    2015

  30. [38]

    The unreasonable effectiveness of pre- trained features for camera pose refinement

    Gabriele Trivigno, Carlo Masone, Barbara Caputo, and Torsten Sattler. The unreasonable effectiveness of pre- trained features for camera pose refinement. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 12786–12798, 2024. 7

    2024

  31. [39]

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

    Xun Wang, Xintong Han, Weilin Huang, Dengke Dong, and Matthew R Scott. Multi-similarity loss with general pair weighting for deep metric learning. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion, pages 5022–5030, 2019. 6

    2019

  32. [40]

    Observing lightning and transient luminous events from the international space station during ilan-es: An astronaut’s per- spective

    Yoav Yair, Melody Korman, Colin Price, and Eytan Stibbe. Observing lightning and transient luminous events from the international space station during ilan-es: An astronaut’s per- spective. Acta Astronautica, 211:592–599, 2023. 1, 3

    2023

  33. [41]

    Seco: Exploring sequence supervision for unsu- pervised representation learning

    Ting Yao, Yiheng Zhang, Zhaofan Qiu, Yingwei Pan, and Tao Mei. Seco: Exploring sequence supervision for unsu- pervised representation learning. In AAAI, 2021. 8

    2021

  34. [42]

    Uavm ’23: 2023 workshop on uavs in multimedia: Capturing the world from a new perspective

    Zhedong Zheng, Yujiao Shi, Tingyu Wang, Jun Liu, Jianwu Fang, Yunchao Wei, and Tat-seng Chua. Uavm ’23: 2023 workshop on uavs in multimedia: Capturing the world from a new perspective. In ACM MM, page 9715–9717, New York, NY , USA, 2023. Association for Computing Machinery. 3 4

    2023

  35. [43]

    Modern backbone for efficient geo- localization

    Runzhe Zhu, Mingze Yang, Kaiyu Zhang, Fei Wu, Ling Yin, and Yujin Zhang. Modern backbone for efficient geo- localization. In Proceedings of the 2023 Workshop on UAVs in Multimedia: Capturing the World from a New Perspec- tive, page 31–37, New York, NY , USA, 2023. Association ...

    2023

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.