Pith. sign in

REVIEW 4 major objections 6 minor 62 references

Artifacts of Idiosyncracy in Global Street View Data

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Street view data skews inside cities even when coverage looks complete

desk verdict A useful 28-city distributional audit with real interview value, but the GSV retrieval method probably thins the very density signal the rankings claim to measure. read the letter →

arxiv 2505.11046 v1 pith:S77ZYYKA submitted 2025-05-16 cs.CV

Tim Alpherts , Sennay Ghebreab , Nanne van Noord This is my paper · ORCID
classification cs.CV
keywords StreetViewDataCoverageDistributionSpatialBiasUrbanAIDatasetsGoogleMapillaryCollectionArtifacts
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

Street view imagery is usually treated as a systematic sample of a city, especially when coverage looks complete. This paper argues that even densely covered cities carry a hidden distributional bias: some neighbourhoods are imaged far more often than others, so any dataset sampled from the imagery inherits that skew. For 28 cities it measures the gap between real panorama locations and a uniform road-coverage prior, using KL-divergence and Earth Mover's distance, and shows that binary coverage percentages barely predict these gaps. A case study of Amsterdam with six interviews traces the skew to human choices in the collection chain, from routes and driver discretion to camera hardware and seasons. The practical point is that auditing a street view dataset by 'was every street covered?' is not enough to know what biases are baked into it.

What carries the argument

The evaluation is carried by a comparison between two densities over the city: $C_{\text{Uniform}}$, a Gaussian kernel density estimate of evenly spaced points at 20 metre intervals along OpenStreetMap roads, and $C_{\text{Real}}$, the same estimate over retrieved panorama locations. Their difference $C_\Delta = C_{\text{Real}} - C_{\text{Uniform}}$ produces maps of over- and under-sampled $1\,\text{km}^2$ cells. Two distances quantify the gap: a $k$-nearest-neighbour estimator of Kullback-Leibler divergence, which is sensitive to roads being skipped, and Earth Mover's distance approximated with the debiased Sinkhorn divergence, which captures how far the centres of mass of over- and under-sampling are separated. The case study adds a qualitative mechanism: semi-structured interviews with six people across the Amsterdam collection chain, from data managers to contracted collectors.

What would settle it

Re-run the KL-divergence, EMD, and $C_\Delta$ analysis against a prior that weights roads by lane count or OpenStreetMap road class instead of weighting every 20-metre point equally; if the measured deviations mostly vanish or shrink dramatically, the claimed idiosyncratic bias is largely an artifact of the uniform prior rather than a property of collection.

Watch

Extended reading notes

Core claim

The paper's central claim is that the spatial distribution of street view coverage deviates from uniform road coverage in most of the 28 cities examined, and that these deviations are artifacts of the collection process rather than of the city itself. It finds cities like Kiev, Almaty, and Reykjavik with near-uniform distributions, while Los Angeles, Auckland, and Sydney show high Earth Mover's distance with relatively low KL-divergence, meaning coverage exists nearly everywhere but its mass is concentrated in certain neighbourhoods. Coverage percentage and distribution metrics do not track each other, so a city can look fully covered by binary measures yet be strongly skewed in where the images actually are. The Amsterdam interviews show that the same city centre gets revisited after traffic obstructions, that different contracted providers drive different distances per day, use different cameras, and collect in different months, and that deduplication only filters images taken on the same day, so the oversampling survives in the database. The paper concludes that domain knowledge of both city and collection process is necessary before treating street view imagery as representative.

Load-bearing premise

The benchmark for 'no bias' is that every road segment should receive the same image density per unit length, so the measured deviations partly reflect departures from that specific uniform ideal rather than from the true collection process.

Editorial extensions

If this is right

  • Machine learning datasets sampled at fixed intervals from street view services carry a neighbourhood-level skew that persists even in cities with near-total coverage.
  • Binary coverage audits are insufficient validation for street view based models; distributional metrics such as EMD and KL should accompany them.
  • In a single city, Google Street View and Mapillary can have opposite bias patterns, so the choice of provider changes what a model learns.
  • The Amsterdam case implies that collection contracts that require equal coverage and cross-day deduplication could remove some artifacts at their source.
  • Researchers building urban perception datasets should treat each city as having its own coverage idiosyncrasy rather than assuming one sampling recipe generalises.

Reading between the lines

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

  • The same vehicle-based collection mechanism applies to other geospatial data streams, so $C_\Delta$-style distribution checks could audit mobile LiDAR, dashcam, and similar fleets for the same obstruction-driven oversampling.
  • The uniform prior is a normative benchmark; re-running the metrics with a road-class-weighted prior would separate 'human idiosyncrasy' from mechanical differences between highway and residential driving, which would sharpen the paper's causal interpretation.
  • A concrete next test is to use the $C_\Delta$ maps as a stratified sampling guide and measure whether model performance becomes more equal across neighbourhoods, which would directly connect distribution bias to downstream AI behaviour.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 a method to evaluate the spatial distribution of street view imagery coverage relative to a uniform road-coverage prior, using the KL divergence and Earth Mover's Distance (approximated by Sinkhorn divergence) to quantify deviations. The authors apply this method to 28 cities for Google Street View and Mapillary, compare the distributional metrics with binary coverage percentages, and present a qualitative case study of Amsterdam based on six semi-structured interviews with people involved in the municipal panorama collection process. The central claim is that street view data shows distributional biases even in densely covered cities, that these biases are not captured by binary coverage measures, and that they trace to human idiosyncrasies in the collection process. The paper concludes that datasets sampled from street view services inherit these distributional biases and that domain knowledge is needed when constructing AI datasets.

Significance. If the quantitative results were fully valid, this paper would make an important contribution to the growing literature on street view data bias, moving the discussion beyond binary coverage to distributional coverage. The proposed pipeline (metadata retrieval, density estimation, distribution comparison) is simple and potentially reusable, and the interview-based case study is a valuable and relatively rare source of evidence linking collection practices to measurable spatial artifacts. The paper is clearly organized and the core intuition is compelling. However, the quantitative claims are currently compromised by a likely retrieval-induced thinning of the Google Street View data, by the absence of reported hyperparameters and uncertainty estimates for the metrics, and by an untested sensitivity to the uniform road prior. These issues bear directly on the rankings and C_delta maps that form the quantitative backbone of the paper.

major comments (4)
  1. [Section 3.1] The Google Street View metadata retrieval procedure is likely to return only the closest panorama per 20m grid cell, because the standard GSV metadata endpoint returns a single result per request. The paper states that 'API requests are then made at every point for the closest images within 100m' and that duplicate panorama IDs are filtered, but it does not account for panoramas that are never returned because they are not the closest panorama to any grid point. This means C_Real is a density-dependent thinning of the full panorama set, not an estimate of the true density of all available imagery. Repeated captures, which are exactly the oversampling the paper aims to measure (e.g., the 12-14 vs 4-5 images per street in Figure 6), are collapsed unless they are the closest to separate grid points. The thinning is more severe in high-density areas, so the KL/EMD values in Tables 1 and 3 and the C_delta maps systematically compress the very biases they claim to quantify. The authors must either show that their retrieval retrieves all panoramas (e.g., by validating against a known city with dense repeated captures or by using a pagination/offset endpoint), or explicitly model and bound the effect of this censoring on the reported metrics.
  2. [Sections 3.2 and 3.3] The paper does not report the Gaussian KDE bandwidth used for density estimation, the value of k used in the k-nearest-neighbor KL divergence estimator, or the Sinkhorn regularization parameter and iteration count used to approximate the EMD. These are free parameters that materially affect the values in Tables 1 and 3, the ranking of cities, and the C_delta maps. Without them, the central quantitative results cannot be reproduced, and no statement is made about how sensitive the rankings are to these choices. The authors should report all hyperparameter values and include a sensitivity analysis over a reasonable range, together with bootstrap or other uncertainty estimates for the KL/EMD scores and their ranking.
  3. [Sections 3.2 and 6.1] The uniform road-coverage prior assumes that every road segment should receive equal image density per unit length. The paper acknowledges in Section 6.1 that OpenStreetMap road classifications have limitations, but it never tests how sensitive the KL/EMD deviations and city rankings are to this central modeling choice. If multi-lane divided roads, slow-moving segments, or certain OSM road classes are systematically imaged more often for mechanical reasons, then part of the measured 'bias' reflects the mismatch between the prior and real collection mechanics rather than human idiosyncrasy. As a concrete test, the authors should reweight OSM road classes by functional class or road type, or use an alternative prior (e.g., one based on road length weighted by expected capture frequency), and show whether the main conclusions and the ranking of cities are preserved.
  4. [Section 4.1] The MANOVA significance scores reported in Tables 5-8 are effectively all zero because of the very large sample sizes and thus provide no discriminating information about whether the observed distributional differences are meaningful beyond statistical noise. The authors should instead report effect sizes (e.g., standardized mean differences or overlap coefficients) or bootstrap confidence intervals for the KL and EMD differences between C_Real and C_Uniform, which would give a more honest sense of the strength of the deviations.
minor comments (6)
  1. [Section 3.3] The formula for W_p is written as an infimum over permutations of n elements, which is only valid for equal-size empirical measures; the notation is also ambiguous about the value of p and the placement of the exponent. Please clarify whether p=1 or p=2 is used and give the exact Sinkhorn divergence formula used in the experiments.
  2. [Section 3.1] The phrase 'squares of400m^2 meters' is ambiguous; presumably the authors mean 400m x 400m squares, but the notation suggests 400 square meters. Please correct the unit description.
  3. [Section 2.1] There is a typo: 'which is is labour-intensive' should be 'which is labour-intensive'.
  4. [Section 4.1] The interpretations of the KL and EMD (e.g., 'A low KL indicates all roads have been covered' and 'A low EMD indicates all parts of the city are being covered equally') are not formally derived from the definitions. Please provide a brief justification or separate these descriptive statements from formal interpretation.
  5. [Section 4.2] The paper reports R^2 = 0.07 for the Mapillary EMD-coverage relationship but does not give Pearson or Spearman correlation coefficients, nor the information needed to assess whether the linear fit is even appropriate. Please report the underlying correlation coefficients and, if relevant, note whether the R^2 comes from a linear or nonlinear fit.
  6. [Abstract and throughout] The word 'idiosyncracy' is consistently misspelled; the correct spelling is 'idiosyncrasy'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the uniform road prior and interview evidence are external to the street-view measurements, and no fitted parameter or self-citation carries the derivation.

full rationale

The paper's central comparison is between two independently constructed distributions: C_Real, a kernel density estimate over panorama locations returned by provider APIs, and C_Uniform, a kernel density estimate over 20m-spaced points on OpenStreetMap road networks. The uniform prior is an external modeling choice, not derived from or fitted to the street view data, so the KL-divergence and Earth Mover's Distance values and the C_delta maps are measurements relative to a stated benchmark rather than quantities that reproduce their inputs. The Amsterdam interview findings come from six human participants describing collection workflows and are independent of the density maps; they are used to narrate, not to compute, the quantitative results. The only self-citations ([3] and [28]) appear in related-work enumerations for visual urban analytics and Flickr-based place recognition, and neither supplies a theorem, ansatz, or fitted value on which the derivation depends. The paper's own limitations (Section 6.1 on OSM boundary and road artifacts, Section 6.2 on low-coverage cities) are validity caveats, not circular reductions. The concern that the Google Street View metadata endpoint may return only the closest panorama per grid query is a measurement-validity issue, but it does not make any claimed result equal to an input by construction. No load-bearing step reduces to a self-citation, a fitted parameter, or a definitional equivalence.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The paper's quantitative claims rest on an external uniform prior and a set of standard statistical tools. The most consequential unverified inputs are the uniform-prior assumption and the unspecified density-estimation parameters. Compared with typical machine learning papers, the axiomatic load is moderate and mostly transparently stated.

free parameters (6)
  • Gaussian KDE bandwidth = not reported
    Used in Section 3.2 to estimate densities for both real and uniform coverage; the value affects the KL and EMD numbers and the city rankings.
  • k for k-NN KL divergence estimator = not reported
    Appears in the D_k estimator in Section 3.3; different k changes KL values, and the paper does not state the chosen k.
  • Sinkhorn regularization and iterations for EMD approximation = not reported
    Used to approximate EMD in Section 3.3; epsilon and iteration count are not given, so the EMD column is not exactly reproducible.
  • GSV retrieval grid spacing and search radius = 20m grid, 100m radius
    Chosen in Section 3.1; these values determine which panoramas are retrieved and therefore the measured distribution.
  • Mapillary retrieval square size = unspecified (described as 400m^2)
    Chosen in Section 3.1; the exact square geometry affects retrieval completeness and the resulting density estimate.
  • C_delta evaluation grid = 1km^2
    Used in all C_delta visualizations; the grid resolution changes the apparent spatial pattern of over- and undersampling.
assumptions (4)
  • domain assumption A mechanistic, unbiased collection process would yield uniform image density per unit length of drivable road.
    This is the uniform prior benchmark in Section 3.2. If real processes target roads differentially by importance, or if road width, lanes, or speed affect capture density, deviations from this prior are not necessarily 'bias'.
  • domain assumption OpenStreetMap provides a complete and accurate road network for all 28 cities for both driveable and publicly accessible streets.
    The reference distribution in Section 3.2 is built from OSM. The paper lists limitations in Section 6.1 (gated roads, missing minor roads, boundary artifacts) but does not quantify how they affect the metrics.
  • domain assumption The API retrieval methods (GSV grid at 20m with 100m radius, Mapillary squares) obtain the complete set of available panoramas in each city.
    Used in Section 3.1; if APIs cap results or miss dense areas, the real coverage distribution is misestimated.
  • domain assumption The Gaussian kernel density estimate is an appropriate density model at the chosen (unstated) bandwidth for both real and uniform point sets.
    The KL and EMD values are computed on these densities; no sensitivity analysis is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Artifacts of Idiosyncracy in Global Street View Data." pith.science (2026). https://pith.science/paper/S77ZYYKA

@misc{pith2026250511046,
  author       = {Pith},
  title        = {Pith review of: Artifacts of Idiosyncracy in Global Street View Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/S77ZYYKA}},
  note         = {Machine review of arXiv:2505.11046}
}
read the original abstract

Street view data is increasingly being used in computer vision applications in recent years. Machine learning datasets are collected for these applications using simple sampling techniques. These datasets are assumed to be a systematic representation of cities, especially when densely sampled. Prior works however, show that there are clear gaps in coverage, with certain cities or regions being covered poorly or not at all. Here we demonstrate that a cities' idiosyncracies, such as city layout, may lead to biases in street view data for 28 cities across the globe, even when they are densely covered. We quantitatively uncover biases in the distribution of coverage of street view data and propose a method for evaluation of such distributions to get better insight in idiosyncracies in a cities' coverage. In addition, we perform a case study of Amsterdam with semi-structured interviews, showing how idiosyncracies of the collection process impact representation of cities and regions and allowing us to address biases at their source.

Figures

Figures reproduced from arXiv: 2505.11046 by the authors.

Figure 1
Figure 1. Analysis of the available Google Street View in Johannesburg (middle) compared against a uniform prior over the [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of cities where the coverage distribution [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. From left to right for Nairobi: Distribution of retrieved metadata in Google Street View, uniform coverage based on [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: Visualisation of how KL-Divergence and Earth Mover’s Distance capture the differences between street view coverage [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Coverage percentages plotted against the Earth Mover’s Distance for Google Street View. Note that the Y and X axis [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Density plot (𝐶Δ) of Google Street View Coverage in Los Angeles. Oversampled neighbourhoods such as Beverly Hills can have 12-14 images in a suburban street whereas undersampled neighbourhoods such as Compton may only have 4-5 images on similar streets. (a) Density plo…
Figure 7
Figure 7. Figure 7: Differences and similarities in coverage for Lima and Johannesburg. While Google Street View and Mapillary have [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Density plots of 𝐶Δ for coverage through Google Street View, Mapillary, and the Amsterdam Municipality. While the degree varies, similar patterns are observable for all three street view databases. The city center is oversampled, while the northern and eastern parts ar…
Figure 10
Figure 10. Figure 10: Disjointed areas for in which external providers collect street view images for the munici￾pality of Amsterdam. what routes they drove, but the log was updated throughout the year so they could more systematically tackle different parts of the city in order. P1 mentio…
Figure 11
Figure 11. Figure 11: Coverage percentages plotted against the KL Divergence for Google Street View for all driveable streets. [PITH_FULL_IMAGE:figures/full_fig_p015_11.png]
Figure 12
Figure 12. Figure 12: Coverage percentages plotted against the EMD for Mapillary for all driveable streets. [PITH_FULL_IMAGE:figures/full_fig_p015_12.png]
Figure 13
Figure 13. Figure 13: Coverage percentages plotted against the KL Divergence for Mapillary for all driveable streets. [PITH_FULL_IMAGE:figures/full_fig_p016_13.png]
Figure 14
Figure 14. Figure 14: Coverage percentages plotted against the EMD for Google Street View for all publicly accessible streets. [PITH_FULL_IMAGE:figures/full_fig_p016_14.png]
Figure 15
Figure 15. Figure 15: Coverage percentages plotted against the KL Divergence for Google Street View for publicly accessible streets. [PITH_FULL_IMAGE:figures/full_fig_p017_15.png]
Figure 16
Figure 16. Figure 16: Coverage percentages plotted against the EMD for Mapillary for all publicly accessible streets. [PITH_FULL_IMAGE:figures/full_fig_p017_16.png]
Figure 17
Figure 17. Figure 17: Coverage percentages plotted against the KL Divergence for Mapillary for publicly accessible streets. [PITH_FULL_IMAGE:figures/full_fig_p018_17.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

62 extracted references · 34 canonical work pages

  1. [1]

    Alcantarilla, Simon Stent, German Ros, Roberto Arroyo, and Riccardo Gherardi

    Pablo F. Alcantarilla, Simon Stent, German Ros, Roberto Arroyo, and Riccardo Gherardi. 2016. Street-View Change Detection with Deconvolutional Networks. Robotics: Science and Systems 12 (2016). doi:10.15607/rss.2016.xii.044

    work page doi:10.15607/rss.2016.xii.044 2016

  2. [2]

    Amar Ali-bey, Brahim Chaib-draa, and Philippe Giguère. 2022. GSV-Cities: Toward appropriate supervised visual place recognition. Neurocomputing 513 (Nov. 2022), 194–203. doi:10.1016/j.neucom.2022.09.127

    doi:10.1016/j.neucom.2022.09.127 2022

  3. [3]

    Tim Alpherts, Sennay Ghebreab, Yen-Chia Hsu, and Nanne Van Noord. 2024. Perceptive Visual Urban Analytics Is Not (Yet) Suitable for Municipalities. In The 2024 ACM Conference on Fairness, Accountability, and Transparency . ACM, Rio de Janeiro Brazil, 1341–1354. doi:10.1145/3630106.3658976

    arXiv 2024

  4. [4]

    Relja Arandjelovic, Petr Gronat, Akihiko Torii, Tomas Pajdla, and Josef Sivic. 2018. NetVLAD: CNN Architecture for Weakly Supervised Place Recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence 40, 6 (2018), 1437–1451. doi:10.1109/TPAMI.2017.2711011 arXiv:1511.07247 NetVLAD, what else?. Artifacts of Idiosyncracy in Global Street View ...

    arXiv 2018

  5. [5]

    Arietta, Alexei A

    Sean M. Arietta, Alexei A. Efros, Ravi Ramamoorthi, and Maneesh Agrawala. 2014. City Forensics: Using Visual Elements to Predict Non-Visual City Attributes.IEEE Transactions on Visualization and Computer Graphics 20, 12 (2014), 2624–2633. doi:10.1109/TVCG.2014.2346446

    arXiv 2014

  6. [6]

    Gabriele Berton, Carlo Masone, and Barbara Caputo. 2022. Rethinking Visual Geo-localization for Large-Scale Applications. arXiv:2204.02287 [cs.CV] https: //arxiv.org/abs/2204.02287

    work page Pith review arXiv 2022

  7. [7]

    Luis G Camara, Carl Gäbert, and Libor P ˇ. 2019. Highly Robust Visual Place Recognition Through Spatial Matching of CNN Highly Robust Visual Place Recognition Through Spatial Matching of CNN Features. September (2019). Refers to Pittsburgh 30k

    work page 2019

  8. [8]

    Song Cao and Noah Snavely. 2015. Graph-Based Discriminative Learning for Location Recognition. International Journal of Computer Vision 112, 2 (2015), 239–254. doi:10.1007/s11263-014-0774-9 Represented visual place recognition database structure in a graph. Still only using bow-based location recognition

    work page doi:10.1007/s11263-014-0774-9 2015

Show all 62 references
  1. [9]

    Zhengping Che, Guangyu Li, Tracy Li, Bo Jiang, Xuefeng Shi, Xinsheng Zhang, Ying Lu, Guobin Wu, Yan Liu, and Jieping Ye. 2019. D 2-City: A Large-Scale Dashcam Video Dataset of Diverse Traffic Scenarios. arXiv:1904.01975 [cs.LG] https://arxiv.org/abs/1904.01975

    2019 arXiv

  2. [10]

    Curtis, Andrew Curtis, Jennifer Mapes, Andrea B

    Jacqueline W. Curtis, Andrew Curtis, Jennifer Mapes, Andrea B. Szell, and Adam Cinderich. 2013. Using Google Street View for Systematic Observation of the Built Environment: Analysis of Spatio-Temporal Instability of Imagery Dates. 12, 1 (2013), 53. doi:10.1186/1476-072X-12-53

    2013 doi

  3. [11]

    Mirthe Dankloff, Vanja Skoric, Giovanni Sileno, Sennay Ghebreab, Jacco van Ossenbruggen, and Emma Beauxis-Aussalet. 2024. Analysing and organising human communications for AI fairness assessment: Use cases from the Dutch Public Sector. AI & SOCIETY (June 2024). doi:10.1007/s00...

    2024 doi

  4. [12]

    Marieke De Goede, Bosma Esmé, and Pallister-Wilkins Polly. 2019. Se- crecy and Methods in Security Research A Guide to Qualitative Fieldwork. Routledge, New York, NY ; Abingdon, Oxon. https://www.routledge. com/Secrecy-and-Methods-in-Security-Research-A-Guide-to-Qualitative- F...

    2019

  5. [13]

    A Dubey, N Nikhil, D Parikh, R Raskar, and César A. Hidalgo. 2016. Deep Learning the City: Quantifying Urban Perception at a Global Scale. Eccv 3 (2016), 398–413. doi:10.1007/978-3-319-46448-0

    2016 doi

  6. [14]

    Christian Ertler, Jerneja Mislej, Tobias Ollmann, Lorenzo Porzi, Gerhard Neuhold, and Yubin Kuang. 2020. The Mapillary Traffic Sign Dataset for Detection and Classification on a Global Scale. arXiv:1909.04422 [cs.CV] https://arxiv.org/abs/ 1909.04422

    2020 arXiv

  7. [15]

    Jean Feydy, Thibault Séjourné, François-Xavier Vialard, Shun ichi Amari, Alain Trouvé, and Gabriel Peyré. 2018. Interpolating between Optimal Transport and MMD using Sinkhorn Divergences. arXiv:1810.08278 [math.ST] https: //arxiv.org/abs/1810.08278

    2018 arXiv

  8. [16]

    Zamfirescu-Pereira, Wendy Ju, and Emma Pierson

    Matt Franchi, J.D. Zamfirescu-Pereira, Wendy Ju, and Emma Pierson. 2023. De- tecting disparities in police deployments using dashcam data. In 2023 ACM Con- ference on Fairness, Accountability, and Transparency (FAccT ’23) . ACM, 534–544. doi:10.1145/3593013.3594020

    2023

  9. [17]

    Mooney, Daniel A

    Dustin Fry, Stephen J. Mooney, Daniel A. Rodríguez, Waleska T. Caiaffa, and Gina S. Lovasi. 2020. Assessing Google Street View Image Availability in Latin American Cities. Journal of Urban Health 97, 4 (Aug. 2020), 552–560. doi:10.1007/ s11524-019-00408-7

    2020

  10. [18]

    Kaiqun Fu, Zhiqian Chen, and Chang Tien Lu. 2018. StreetNet: Preference Learning with Convolutional Neural Network on Urban Crime Perception. GIS: Proceedings of the ACM International Symposium on Advances in Geographic Information Systems August 2019 (2018), 269–278. doi:10.1...

    2018

  11. [19]

    Lee Ann Fujii. 2018. Interviewing in Social Science Research: a Re- lational Approach . Routledge, New York, NY ; Abingdon, Oxon. https://www.routledge.com/Interviewing-in-Social-Science-Research-A- Relational-Approach/Fujii/p/book/9780415843744

    2018

  12. [20]

    Gauteng City-Region Observatory (GCRO). 2016. Quality of Life Survey 2015-2016 [dataset]. Round 4. Version 1 . Johannesburg; Cape Town. doi:10.25828/w490-a496

    2016 doi

  13. [21]

    Gauteng City-Region Observatory (GCRO). 2021. Quality of Life Survey 2020-2021 [dataset]. Round 6. Version 1. Johannesburg; Cape Town. doi:10.25828/wemz-vf31

    2021 doi

  14. [22]

    Gauteng City-Region Observatory (GCRO). 2022. Quality of Life Survey V 2017- 2018 [dataset]. Round 5. Version 2 . Johannesburg; Cape Town. doi:10.25828/8yf7- 9261

    2022 doi

  15. [23]

    Timnit Gebru, Jonathan Krause, Yilun Wang, Duyun Chen, Jia Deng, Erez Lieber- man Aiden, and Li Fei-Fei. 2017. Using Deep Learning and Google Street View to Estimate the Demographic Makeup of Neighborhoods across the United States. Proceedings of the National Academy of Scienc...

    2017 doi

  16. [24]

    Ruben Gomez-Ojeda, Manuel Lopez-Antequera, Nicolai Petkov, and Javier Gonzalez-Jimenez. 2015. Training a Convolutional Neural Network for Appearance-Invariant Place Recognition. (2015), 1–9. arXiv:1505.07428

    2015 arXiv

  17. [25]

    Inske Groenen, Stevan Rudinac, and Marcel Worring. 2022. PanorAMS: Automatic Annotation for Detecting Objects in Urban Context. (2022). arXiv:2208.14295

    2022 arXiv

  18. [26]

    Christiaan Hamann. 2024. Segregation and socio-economic sorting in Gauteng. Map of the Month. Gauteng City-Region Observatory (August 2024). doi:10.36634/ MIJV2656

    2024

  19. [27]

    Tianyuan Huang, Zejia Wu, Jiajun Wu, Jackelyn Hwang, and Ram Rajagopal

  20. [28]

    Sarah Ibrahimi, Nanne van Noord, Tim Alpherts, and Marcel Worring. 2021. Inside Out Visual Place Recognition. CoRR abs/2111.13546 (2021). arXiv:2111.13546 https://arxiv.org/abs/2111.13546

    2021 arXiv

  21. [29]

    Braswell, Andrew D

    Nathan Jacobs, Walker Burgin, Nick Fridrich, Austin Abrams, Kylia Miskell, Bobby H. Braswell, Andrew D. Richardson, and Robert Pless. 2009. The Global Network of Outdoor Webcams: Properties and Applications. InACM SIGSPATIAL International Conference on Advances in Geographic I...

    2009

  22. [30]

    Junghwan Kim and Kee Moon Jang. 2023. An Examination of the Spatial Coverage and Temporal Variability of Google Street View (GSV) Images in Small- and Medium-Sized Cities: A People-Based Approach. Computers, Environment and Urban Systems 102 (June 2023), 101956. doi:10.1016/j....

    2023

  23. [31]

    Stephen Law, Brooks Paige, and Chris Russell. 2019. Take a Look around: Using Street View and Satellite Images to Estimate House Prices. ACM Transactions on Intelligent Systems and Technology 10, 5 (2019), 1–19. doi:10.1145/3342240 arXiv:1807.07155

    2019 arXiv

  24. [32]

    Nachuan Ma, Jiahe Fan, Wenshuo Wang, Jin Wu, Yu Jiang, Lihua Xie, and Rui Fan. 2022. Computer Vision for Road Imaging and Pothole Detection: A State-of- the-Art Review of Systems and Algorithms. (2022), 1–16. doi:10.1093/tse/tdac026 arXiv:2204.13590

    2022 arXiv

  25. [33]

    Jeaneth Machicao, Alison Specht, Danton Vellenich, Leandro Meneguzzi, Romain David, Shelley Stall, Katia Ferraz, Laurence Mabile, Margaret O’brien, and Pedro Corrêa. 2022. A Deep-Learning Method for the Prediction of Socio-Economic Indicators from Street-View Imagery Using a C...

    2022 doi

  26. [34]

    Fabio Miranda, Maryam Hosseini, Marcos Lage, Harish Doraiswamy, Graham Dove, and Cláudio T. Silva. 2020. Urban Mosaic: Visual Exploration of Streetscapes Using Large-Scale Image Data. Conference on Human Factors in Computing Systems - Proceedings (2020). doi:10.1145/3313831.33...

    2020

  27. [35]

    Gabriele Moreno Berton, Valerio Paolicelli, Carlo Masone, and Barbara Caputo

  28. [36]

    Moura, Shizhan Zhu, and Orly Zvitia

    Daniel C. Moura, Shizhan Zhu, and Orly Zvitia. 2025. Nexar Dashcam Collision Prediction Dataset and Challenge. arXiv:2503.03848 [cs.CV] https://arxiv.org/ abs/2503.03848

    2025 arXiv

  29. [37]

    Emily Muller, Emily Gemmell, Ishmam Choudhury, Ricky Nathvani, Antje Bar- bara Metzler, James Bennett, Emily Denton, Seth Flaxman, and Majid Ezzati. 2022. City-Wide Perceptions of Neighbourhood Quality Using Street View Images . Vol. 1. Association for Computing Machinery. arX...

    2022 arXiv

  30. [38]

    Nikhil Naik, Jade Philipoom, Ramesh Raskar, and Cesar Hidalgo. 2014. Streetscore- Predicting the Perceived Safety of One Million Streetscapes. IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops January (2014), 793–799. doi:10.1109/CVPRW.2014.121

    2014 doi

  31. [39]

    Gerhard Neuhold, Tobias Ollmann, Samuel Rota Bulò, and Peter Kontschieder

  32. [40]

    Gerhard Neuhold, Tobias Ollmann, Samuel Rota Bulo, and Peter Kontschieder

  33. [41]

    OpenStreetMap contributors. 2017. Planet dump retrieved from https://planet.osm.org . https://www.openstreetmap.org

    2017

  34. [42]

    Vicente Ordonez and Tamara L. Berg. 2014. Learning High-Level Judgments of Urban Perception. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 8694 LNCS, PART 6 (2014), 494–510. doi:10.1007/978-...

    2014 doi

  35. [43]

    Fernando Perez-Cruz. 2008. Kullback-Leibler Divergence Estimation of Continu- ous Distributions. In 2008 IEEE International Symposium on Information Theory . 1666–1670. doi:10.1109/ISIT.2008.4595271

    2008

  36. [44]

    In Proceedings of the IEEE International Conference on Computer Vision (ICCV)

    The Mapillary Vistas Dataset for Semantic Understanding of Street Scenes. In Proceedings of the IEEE International Conference on Computer Vision (ICCV)

  37. [45]

    Sukanya Randhawa, Eren Aygun, Guntaj Randhawa, Benjamin Herfort, Sven Lautenbach, and Alexander Zipf. 2024. Paved or unpaved? A Deep Learn- ing derived Road Surface Global Dataset from Mapillary Street-View Imagery. arXiv:2410.19874 [cs.CV] https://arxiv.org/abs/2410.19874

    2024 arXiv

  38. [46]

    Pearson, Benjamin V

    Amanda Rzotkiewicz, Amber L. Pearson, Benjamin V. Dougherty, Ashton Short- ridge, and Nick Wilson. 2018. Systematic Review of the Use of Google Street View in Health Research: Major Themes, Strengths, Weaknesses and Possibilities for Future Research. Health & Place 52 (July 20...

    2018 doi

  39. [47]

    Ken Sakurada. 2018. Weakly Supervised Silhouette-based Semantic Change Detection. (2018). arXiv:1811.11985

    2018 arXiv

  40. [48]

    Sterling Quinn and Luis Alvarez León. 2019. Every Single Street? Rethinking Full Coverage across Street-Level Imagery Platforms. Transactions in GIS 23, 6 (2019), 1251–1272. doi:10.1111/tgis.12571

    2019 doi

  41. [49]

    Chanuki Illushka Seresinhe, Tobias Preis, and Helen Susannah Moat. 2017. Using Deep Learning to Quantify the Beauty of Outdoor Places. Royal Society Open Science 4, 7 (2017). doi:10.1098/rsos.170170

    2017 doi

  42. [50]

    Smith, Joel D

    Cara M. Smith, Joel D. Kaufman, and Stephen J. Mooney. 2021. Google street view image availability in the Bronx and San Diego, 2007–2020: Understanding potential biases in virtual audits of urban built environments. Health & Place 72 (2021), 102701. doi:10.1016/j.healthplace.2...

    2021

  43. [51]

    Polak, James E

    Esra Suel, John W. Polak, James E. Bennett, and Majid Ezzati. 2019. Measuring Social, Environmental and Health Inequalities Using Deep Learning and Street Imagery. Scientific Reports 9, 1 (2019), 1–10. doi:10.1038/s41598-019-42036-w

    2019 doi

  44. [52]

    Ian Seiferling, Nikhil Naik, Carlo Ratti, and Raphäel Proulx. 2017. Green Streets - Quantifying and Mapping Urban Trees with Street-Level Imagery and Computer Vision. Landscape and Urban Planning 165, July 2016 (2017), 93–101. doi:10.1016/ j.landurbplan.2017.05.010

    2017

  45. [53]

    Akihiko Torii, Relja Arandjelovi, Sivic Masatoshi, and Okutomi Tomas. [n. d.]. Torii-CVPR-2015-final. ([n. d.])

    2015

  46. [54]

    Akihiko Torii, Josef Sivic, Toma Pajdla, and Masatoshi Okutomi. 2013. Visual Place Recognition with Repetitive Structures. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition (2013), 883–890. doi:10.1109/CVPR.2013.119

    2013 doi

  47. [55]

    Farouk Umar, Josephine Amoah, Moses Asamoah, Mawuli Dzodzomenyo, Chid- inma Igwenagu, Lorna-Grace Okotto, Joseph Okotto-Okotto, Pete Shaw, and Jim Wright. 2023. On the Potential of Google Street View for Environmental Waste Quantification in Urban Africa: An Assessment of Bias...

    2023

  48. [56]

    Maarten Sukel, Stevan Rudinac, and Marcel Worring. 2020. Urban Object Detec- tion Kit: A System for Collection and Analysis of Street-Level Imagery. ICMR 2020 - Proceedings of the 2020 International Conference on Multimedia Retrieval (2020), 509–516. doi:10.1145/3372278.3390708

    2020

  49. [57]

    Frederik Warburg, Søren Hauberg, Manuel López-Antequera, Pau Gargallo, Yubin Kuang, and Javier Civera. 2020. Mapillary Street-Level Sequences: A Dataset for Lifelong Place Recognition. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2623–2632. do...

    2020

  50. [58]

    Tobias Weyand, André Araújo, Bingyi Cao, and Jack Sim. 2020. Google Land- marks Dataset v2 - A Large-Scale Benchmark for Instance-Level Recognition and Retrieval. CoRR abs/2004.01804 (2020). arXiv:2004.01804 https://arxiv.org/abs/ 2004.01804 Artifacts of Idiosyncracy in Global...

    2020 arXiv

  51. [60]

    Vaserstein, L

    N. Vaserstein, L. 1969. Markov Processes over Denumerable Products of Spaces, Describing Large Systems of Automata. Probl. Peredachi Inf. (1969), 64–72. http: //mi.mathnet.ru/ppi1811

    1969

  52. [2017]

    In 2017 IEEE International Conference on Computer Vision (ICCV)

    The Mapillary Vistas Dataset for Semantic Understanding of Street Scenes. In 2017 IEEE International Conference on Computer Vision (ICCV) . 5000–5009. doi:10.1109/ICCV.2017.534

    2017 doi

  53. [2021]

    In 2021 IEEE Winter Conference on Applications of Computer Vision (W ACV)

    Adaptive-Attentive Geolocalization from few queries: a hybrid approach. In 2021 IEEE Winter Conference on Applications of Computer Vision (W ACV). IEEE, 2917–2926. doi:10.1109/wacv48630.2021.00296

    2021

  54. [2024]

    arXiv:2401.01107 [cs] Comment: Accepted by AAAI 2024

    CityPulse: Fine-Grained Assessment of Urban Change with Street View Time Series. arXiv:2401.01107 [cs] Comment: Accepted by AAAI 2024

    2024 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.