Pith. sign in

REVIEW 4 major objections 4 minor 23 references

GlobalBuildingAtlas: An Open Global and Complete Dataset of Building Polygons, Heights and LoD1 3D Models

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

Pith's one-line read GlobalBuildingAtlas claims to be the first open global dataset with individual building polygons, heights, and LoD1 3D models, totalling 2.75 billion footprints and 2.68 billion height-assigned buildings with per-continent RMSEs from 1.5…

desk verdict A genuinely massive and useful dataset whose 'global completeness' claim currently outruns the evidence, mainly because Africa is unvalidated and the 97% completeness number means something weaker than it appears. read the letter →

arxiv 2506.04106 v1 pith:RSIEPIFD submitted 2025-06-04 cs.CV

classification cs.CV
keywords globalbuildingdatasetfootprintsheightestimationLoD13DmodelsPlanetScopesatelliteimagerymonocularquality-guidedpolygonfusionSustainableDevelopmentGoal11
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

GlobalBuildingAtlas aims to close a major gap in geospatial knowledge: there is no open, globally complete record of individual buildings in three dimensions. The paper claims to deliver one, built from 2019 PlanetScope optical imagery alone, with 2.75 billion building polygons, 3-meter height maps, and a LoD1 model in which 2.68 billion buildings (97% of the total) have predicted heights. Height accuracy is reported as RMSEs between 1.5 m (Oceania) and 8.9 m (South America), with a global average of 5.5 m. If these claims hold, researchers and policymakers get the first building-level 3D inventory of the entire planet, useful for urban planning, population and infrastructure analysis, and monitoring the UN's sustainable cities goal. The dataset also implies the global building stock is roughly 2.7 billion buildings, not the UN's often-cited 4 billion.

What carries the argument

The load-bearing mechanism is a two-stage satellite-only pipeline. First, a quality-guided polygon fusion selects a base footprint layer per administrative region (OpenStreetMap everywhere except South America and Africa, where Google Open Buildings is preferred), picks a complementary secondary source by recall and area gain, and merges instances so that the final polygons keep the best source's geometry while adding missing buildings. Second, a monocular height estimation network (HTC-DC Net) trained on LiDAR-derived normalized digital surface models predicts pixel-wise heights from 3 m PlanetScope imagery; each building footprint receives the maximum predicted height inside it, and test-time augmentation variance is used as an uncertainty measure. The resulting 3 m height grids are 30 times finer than prior global raster products, and the fusion adds more than one billion buildings missing from earlier footprint sets.

What would settle it

Take a set of African cities absent from training (for example Nairobi, Lagos, and Kinshasa), obtain independent airborne LiDAR or field-verified building heights, and compute the RMSE of GBA.LoD1 heights against them; if the African RMSE lies far outside the reported 1.5 to 8.9 m range or height completeness drops well below 97%, the global-generalization claim fails. A complementary check would test whether predicted heights in high-rise districts of Medellín, Wakayama, or comparable cities show systematic underestimation against LiDAR, which would imply global building volumes are biased low even if per-building RMSE looks acceptable.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central discovery is that a complete global LoD1 building inventory can be generated from openly available optical satellite images and open building footprints, without LiDAR or local surveys. The resulting GlobalBuildingAtlas consists of three products: GBA.Polygon (2.75 billion fused building footprints), GBA.Height (global 3 m building height maps), and GBA.LoD1 (2.68 billion extruded building boxes with heights). The paper reports that this is the first dataset to combine instance-level completeness with global coverage, beating prior products in footprint quality metrics (AP50poly, AR50poly, N-ratio, IoU) across most continents and achieving height RMSEs of 1.5 to 8.9 m. Using reference N-ratios to correct for under- and over-counting, it estimates roughly 2.71 billion buildings worldwide (range 2.64 to 2.97 billion), suggesting the UN's roughly 4 billion figure overestimates the global building stock.

Load-bearing premise

The height model was trained only on LiDAR-derived reference data from regions concentrated in North America, Europe, and Oceania, and the paper's claim of global height accuracy depends on that model transferring to the whole world, including Africa, where no training or validation data exist.

Editorial extensions

If this is right

  • Urban research can move from coarse 90 to 250 m raster height products to building-level 3D volumes with 3 m height maps.
  • Countries without LiDAR or building surveys gain a consistent baseline for planning, risk exposure, and infrastructure assessment, since all data derive from open satellite imagery and open footprints.
  • Volume-based SDG indicators outperform area-based ones: volume per capita correlates with GDP per capita at 0.85 versus 0.76 and raises pairwise ranking agreement from 79.6% to 83.5%.
  • The global building stock is closer to 2.7 billion buildings (range 2.64 to 2.97 billion) than the UN's roughly 4 billion estimate, changing denominators for per-capita and density calculations.
  • Because the pipeline uses only PlanetScope imagery, the dataset can be regenerated over time to track urban growth and compute land-consumption-to-population-growth ratios (SDG Indicator 11.3.1) directly.

Reading between the lines

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

  • If the systematic underestimation in high-rise districts seen in Medellín and Wakayama extends elsewhere, total building volumes in dense Asian and South American cities could be understated even where per-building RMSE looks acceptable.
  • The paper's transfer of confidence from South America to Africa is an extrapolation rather than a validation, so African building counts and volumes should be treated as provisional until independent height reference data exist.
  • The same satellite-only pipeline could be rerun on older PlanetScope mosaics to create a global building-volume time series, which would let SDG Indicator 11.3.1 be measured directly instead of through GDP correlations.
  • The N-ratio scaling used to estimate the global count assumes the reference cities' detection rates represent each continent; if small informal buildings are under-detected, the true count could be higher than 2.7 billion.
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 / 4 minor

Summary. The paper introduces GlobalBuildingAtlas (GBA), a publicly released dataset combining global building polygons, a 3-meter-resolution building height raster, and LoD1 building models. The pipeline relies exclusively on PlanetScope optical imagery for building mask extraction and monocular height estimation, then fuses the resulting polygons with existing footprint sources through a quality-guided strategy. The authors claim 2.75 billion building polygons worldwide, 2.68 billion of them with predicted heights, height RMSEs of 1.5–8.9 m across continents, and the first complete global LoD1 building dataset. Validation is performed against government LiDAR/LoD1 references in 28 cities across Asia, Europe, North America, South America, and Oceania, with no reference data in Africa.

Significance. If the claims hold, this would be an important community resource: the dataset and code are public, the 3 m height raster is far finer than existing global products, and the height validation in evaluated continents is anchored to independent government LiDAR references. The quality-guided polygon fusion is also a useful methodological contribution, and the population/volume and SDG analyses demonstrate plausible downstream value. However, the global completeness and global accuracy claims currently exceed what the validation evidence supports, especially for Africa and for the overall completeness metric.

major comments (4)
  1. [Abstract, §1, §5.2.3, Table 3] The claim that GBA.Polygon is 'the first complete set of global building polygons' is not supported by the paper's own validation in the only under-represented continent with test data. For South America, GBA.LoD1 achieves N-ratio 0.69 and APpoly50 0.5, meaning a substantial fraction of reference buildings is missing at the instance level, and no validation is provided for Africa. The 'complete' wording should be replaced by a claim about coverage relative to existing products, or complemented by validation in the regions where completeness is asserted.
  2. [§5.2.1, §5.2.3, §5.4] The height accuracy claim 'RMSEs ranging from 1.5 m to 8.9 m across different continents' is unsupported for Africa. Section 5.2.1 states that no reference data exist in Africa; Section 5.2.3 substitutes South American results 'by analogy'; and Section 5.4 concedes that the height model was neither trained nor validated on Africa and may be subject to domain shift. South America is the weakest evaluated continent (RMSE_BH 8.9 m versus a 5.5 m global average, and RMSE_BV 586.8 m3/100m2). Extrapolating from the worst-performing validated continent to an unvalidated one is not a substitute for validation. The authors should either provide African validation or explicitly restrict the accuracy claim to the evaluated continents and label Africa as an unvalidated extrapolation.
  3. [Abstract, §5.2.2, Table 3] The headline 'height completeness of more than 97%' is computed as 2.68/2.75 billion polygons that received any predicted height, not as the completeness metric defined in Section 5.2.2, which is the proportion of reference ground-truth buildings with a valid prediction above the 1 m threshold. The reference-matched completeness of GBA.LoD1 in Table 3 ranges from 0.76 (Asia) to 0.96 (North America). These two quantities are not comparable, and presenting the assignment fraction as 'height completeness' overstates the validated completeness of the LoD1 model. Both numbers should be reported separately, and the headline claim should be aligned with the validation metric.
  4. [§5.1, §5.2.1] Continent-level accuracy statements (global average RMSE_BH 5.5 m; per-continent RMSEs) are derived from a test set of only 28 cities, with no city list, no selection criteria, and no confidence intervals. Since Asia alone contains 1.22 billion predicted buildings, a 28-city sample cannot support continent-level accuracy claims without stratified sampling or uncertainty quantification. Please provide city-level results, describe how the 28 cities were chosen, and either qualify the continent-level claims or report their statistical uncertainty.
minor comments (4)
  1. [Abstract, §1] There are typos in the product name: 'GlobalBuildingAltas' in the abstract and 'GlobalBuildignAtlas' in Section 1.
  2. [Figure 4] The figure appears to repeat the Asia block twice and uses inconsistent legend labels ('GHSL-3D' in one block and 'GHS-BUILT-H' in the other); please correct the duplicated panel and standardize the labels.
  3. [§4.5.2, Appendix B] Assigning building height as the maximum height within each footprint may introduce a positive bias, especially since PSR-derived polygons are acknowledged in Appendix B to merge adjacent buildings; a brief discussion or sensitivity analysis using mean or median height would be helpful.
  4. [§7] The code is described as available under an MIT license with the Commons Clause, which is not an OSI-approved open-source license; if the paper claims 'open' code, the practical restrictions of the Commons Clause should be stated explicitly.

Circularity Check

1 steps flagged · score 4.0 of 10

Headline 'height completeness >97%' is an internally defined coverage fraction, not the paper's own reference-based completeness metric; the core dataset otherwise rests on external LiDAR/LoD1 validation.

  1. self definitional [Abstract; Sec. 5.2.2 Metrics; Table 3; Sec. 8 Conclusions]
    "GBA.LoD1 represents the first complete global LoD1 building models, including 2.68 billion building instances with predicted heights, i.e., with a height completeness of more than 97%, achieving RMSEs ranging from 1.5 m to 8.9 m across different continents."

    In Sec. 5.2.2 the paper defines 'Building height completeness (Comp.)' as the proportion of ground-truth reference buildings for which a valid (>1 m) prediction exists; Table 3 reports GBA.LoD1 Comp values of 0.76-0.96 across the evaluated continents, never above 0.96. The abstract's 'height completeness of more than 97%' is instead the ratio 2.68/2.75 billion, i.e., the share of the dataset's own polygons that were assigned any height value from GBA.Height. The headline number is therefore not the paper's externally anchored completeness metric but an internal coverage statistic, so the 'complete LoD1' claim is partly true by construction rather than by reference-based validation.

full rationale

The central derivation chain is largely self-contained: building heights are trained on government LiDAR-derived nDSM data (Sec. 4.4.1) and validated against separate government LoD1 reference cities (Sec. 5.2.1), and polygon quality is evaluated against those same external references rather than against the training labels. The self-citations to Zhu et al. (2024) for the OSM base-layer choice, to Chen et al. (2023) for HTC-DC Net, and to Zhang et al. (2025) for polygonization are methodological and externally falsifiable; they do not by themselves force the validation results. The Africa generalization argument is an extrapolation and a stated domain-shift limitation, not a circular reduction. The one genuine circularity is the presentation of the fraction of polygons that received any predicted height as 'height completeness' when the paper's own validated completeness metric is defined against ground-truth buildings and ranges from 0.76 to 0.96. Because the accuracy claims (RMSE 1.5-8.9 m) and the polygon completeness numbers are still anchored to external references, the circularity is partial rather than total, giving a score of 4.

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

No new physical entities are postulated. The load-bearing assumptions are about the representativeness of training data, the reliability of external validation data, and the correctness of filtering and fusion choices. The free parameters are thresholds and scaling factors that influence the reported quality and completeness metrics.

free parameters (5)
  • Minimum building height threshold = 1 m
    Used to define whether a predicted height is valid in the completeness metric (Section 5.2.2). Hand-chosen; directly affects reported completeness values.
  • WorldCover built-up mask dilation window = 250 m
    Used to filter false-positive building polygons (Section 4.3.5). This ad hoc window size influences global polygon recall and precision.
  • Cloud cover threshold = 10%
    Scenes with more than 10% cloud cover were discarded (Section 4.2). Affects which imagery is used and therefore spatial coverage and quality.
  • Sliding window stride / max predictions per pixel = 128 px; up to 4
    Uncertainty quantification uses up to 4 overlapping predictions per pixel (Section 4.4.3). The choice affects the reported variance-based uncertainty.
  • N-ratio global average = 1.03
    Global average N-ratio computed from validation cities is used to rescale the detected building count into an estimate of the true global count (Section 5.3). This correction is fitted to the paper's own test dataset.
assumptions (4)
  • domain assumption LiDAR nDSM data from North America, Europe, and Oceania are representative of global building height morphology
    The height model is trained only in these regions (Section 4.4.1) yet applied globally, including continents with no training data.
  • ad hoc to paper OSM is the highest-quality base footprint layer except in South America and Africa
    This choice is taken from the authors' prior work Zhu et al. (2024) rather than re-evaluated in this paper (Section 4.5.1).
  • ad hoc to paper WorldCover built-up mask correctly captures all built-up areas worldwide
    Polygons outside a 250m-dilated WorldCover built-up mask are discarded as false positives (Section 4.3.5); if the mask misses built-up areas, true buildings are lost.
  • domain assumption Government-published LoD1 reference datasets are accurate and complete
    Validation RMSEs are computed against these references (Section 5.2.1); errors in the references would propagate to reported accuracies.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GlobalBuildingAtlas: An Open Global and Complete Dataset of Building Polygons, Heights and LoD1 3D Models." pith.science (2026). https://pith.science/paper/RSIEPIFD

@misc{pith2026250604106,
  author       = {Pith},
  title        = {Pith review of: GlobalBuildingAtlas: An Open Global and Complete Dataset of Building Polygons, Heights and LoD1 3D Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RSIEPIFD}},
  note         = {Machine review of arXiv:2506.04106}
}
read the original abstract

We introduce GlobalBuildingAtlas, a publicly available dataset providing global and complete coverage of building polygons, heights and Level of Detail 1 (LoD1) 3D building models. This is the first open dataset to offer high quality, consistent, and complete building data in 2D and 3D form at the individual building level on a global scale. Towards this dataset, we developed machine learning-based pipelines to derive building polygons and heights (called GBA.Height) from global PlanetScope satellite data, respectively. Also a quality-based fusion strategy was employed to generate higher-quality polygons (called GBA.Polygon) based on existing open building polygons, including our own derived one. With more than 2.75 billion buildings worldwide, GBA.Polygon surpasses the most comprehensive database to date by more than 1 billion buildings. GBA.Height offers the most detailed and accurate global 3D building height maps to date, achieving a spatial resolution of 3x3 meters-30 times finer than previous global products (90 m), enabling a high-resolution and reliable analysis of building volumes at both local and global scales. Finally, we generated a global LoD1 building model (called GBA.LoD1) from the resulting GBA.Polygon and GBA.Height. GBA.LoD1 represents the first complete global LoD1 building models, including 2.68 billion building instances with predicted heights, i.e., with a height completeness of more than 97%, achieving RMSEs ranging from 1.5 m to 8.9 m across different continents. With its height accuracy, comprehensive global coverage and rich spatial details, GlobalBuildingAltas offers novel insights on the status quo of global buildings, which unlocks unprecedented geospatial analysis possibilities, as showcased by a better illustration of where people live and a more comprehensive monitoring of the progress on the 11th Sustainable Development Goal of the United Nations.

Figures

Figures reproduced from arXiv: 2506.04106 by the authors.

Figure 1
Figure 1. Distribution of city-scale regions of interest where training and test data were collected. The 3D and 2D training sets were used to train the building height estimation and building polygon generation pipelines, respectively. a spatial resolution of approximately 3 meters. Additionally, the PlanetScope constellation offers a high temporal revisit frequency, capturing imagery of the same location up to daily, which … view at source ↗
Figure 2
Figure 2. Workflow of the proposed pipeline. 4.2 Global Data Acquisition This section details the acquisition, organization and preprocessing of PSR data to prepare them for the subsequent deep learning models and LoD1 model generation. We divided the Earth’s surface into grid cells of 0.2 ◦ × 0.2 ◦ . Grids overlapping with built-up areas, as defined by the Global Urban Footprint (GUF) dataset (Esch et al. (2010, 2011, 2012, … view at source ↗
Figure 3
Figure 3. Overview of our resulting GlobalBuildingAtlas dataset, consisting of global building polygons (GBA.Polygon), building height map (GBA.Height) and LoD1 models (GBA.LoD1). The top section presents the dataset statistics by continent, including the number of buildings, total building area and volume, as well as height RMSE and volume RMSE across the test cities. The middle section displays the global distribution of bu… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Overview of the continental comparison results across large-scale building height products. In the resolution dimension, the products were first scored based on whether they provide vectorized building footprints. Raster-based products were then further scored accordin…
Figure 5
Figure 5. Figure 5: Visual comparison of existing building height products on the test cities Portland (North America), Medellín (South America), Bordeaux (Europe), Launceston (Oceania) and Wakayama (Asia). 5.4 Discussions Based on the preceding analysis of our dataset, its key strengths …
Figure 6
Figure 6. Figure 6: Regression analysis between population and building volume were conducted for the entire EU as well as for each of the 27 EU member states individually. The first graph shows the regression analysis of the EU as a whole, followed by the 27 member countries sorted in de…
Figure 7
Figure 7. Figure 7: Building volume per capita and the harmonized correlation coefficients for the 27 EU member states and the EU as a whole. The harmonized correlation coefficients were computed as an average of the Pearson (r) and Spearman (ρ) correlation coefficients. (Year 2021) in Af…
Figure 8
Figure 8. Figure 8: Regression analysis of population and building volume across all countries and territories globally with Pearson (r) and Spearman (ρ) correlation coefficients. Top 10 and bottom 10 countries or territories are displayed by building volume per capita. (Year 2019) we wer…
Figure 9
Figure 9. Figure 9: Correlations between GDP per capita and two built environment indicators: building volume per capita and building area per capita. Pearson (r) and Spearman (ρ) correlation coefficients are reported. (Year 2019) As [PITH_FULL_IMAGE:figures/full_fig_p025_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

23 extracted references · 17 canonical work pages

  1. [9]

    Esch, T., Brzoska, E., Dech, S., Leutner, B., Palacios-Lopez, D., Metz-Marconcini, A., Marconcini, M., Roth, A., and Zeidler, J.: World Settlement Footprint 3D - A first three-dimensional survey of the global building stock, Remote Sensing of Environment, 270, 112 877, https://doi.org/10.1016/j.rse.2021.112877,

  2. [11]

    GADM: Global Administrative Areas (GADM), https://gadm.org/index.html, accessed: 2025-04-23,

  3. [12]

    Ma, X., Zheng, G., Xu, C., et al.: A global product of 150-m urban building height based on spaceborne lidar, Scientific Data, 11, 1387, https://doi.org/10.1038/s41597-024-04237-5,

  4. [14]

    Peters, R., Dukai, B., Vitalis, S., van Liempt, J., and Stoter, J.: Automated 3D reconstruction of LoD2 and LoD1 models for all 10 million buildings of the Netherlands, https://doi.org/10.14358/PERS.21-00032R2,

  5. [15]

    Shi, Q., Zhu, J., Liu, Z., Guo, H., Gao, S., Liu, M., Liu, Z., and Liu, X.: The Last Puzzle of Global Building Footprints– Mapping 280 Million Buildings in East Asia Based on VHR Images, Journal of Remote Sensing, 4, 0138, https://doi.org/10.34133/remotesensing.0138,

  6. [17]

    The World Bank: Population, total - World Bank Data, https://data.worldbank.org/indicator/SP.POP.TOTL, accessed: 2025-04-16, 2019b

    The World Bank: GDP per capita (current US$), https://data.worldbank.org/indicator/NY .GDP.PCAP.CD, accessed: 2025-04-16, 2019a. The World Bank: Population, total - World Bank Data, https://data.worldbank.org/indicator/SP.POP.TOTL, accessed: 2025-04-16, 2019b. United Nations: The Sustainable Development Goals Report 2024, https://unstats.un.org/sdgs/repor...

  7. [18]

    United Nations: Global Indicator Framework for the Sustainable Development Goals and Targets of the 2030 Agenda for Sustainable Development, https://unstats.un.org/sdgs/indicators/Global-Indicator-Framework-after-2025-review-English.pdf, accessed: 2025-04-08,

  8. [19]

    rep., United Nations Statistics Division, https://unstats.un.org/sdgs/metadata/files/Metadata-11-03-01.pdf, accessed: 2025-04-16,

    United Nations Human Settlements Programme (UN-Habitat): Metadata for Indicator 11.3.1: Ratio of Land Consumption Rate to Population Growth Rate, Tech. rep., United Nations Statistics Division, https://unstats.un.org/sdgs/metadata/files/Metadata-11-03-01.pdf, accessed: 2025-04-16,

Show all 23 references
  1. [21]

    X.: Global Collinearity-aware Polygonizer for Polygonal Building Mapping in Remote Sensing, https: //arxiv.org/abs/2505.01385,

    Zhang, F., Shi, Y ., and Zhu, X. X.: Global Collinearity-aware Polygonizer for Polygonal Building Mapping in Remote Sensing, https: //arxiv.org/abs/2505.01385,

  2. [23]

    Zhu, X. X., Chen, S., Zhang, F., Shi, Y ., and Wang, Y .: GlobalBuildingAtlas: An Open Global and Complete Dataset of Building Polygons, Heights and LoD1 3D Models, https://doi.org/10.14459/2025mp1782307, accessed: 2025-06-02,

  3. [2010]

    Esch, T., Schenk, A., Ullmann, T., Thiel, M., Roth, A., and Dech, S.: Characterization of Land Cover Types in TerraSAR-X Images by Combined Analysis of Speckle Statistics and Intensity Information, IEEE Transactions on Geoscience and Remote Sensing, 49, 1911– 1925, https://doi...

  4. [2011]

    Esch, T., Taubenböck, H., Roth, A., Heldens, W., Felbier, A., Thiel, M., Schmidt, M., Müller, A., and Dech, S.: TanDEM-X Mission - New Perspectives for the Inventory and Monitoring of Global Settlement Patterns, Journal of Applied Remote Sensing, 6, 061 702, https://doi.org/10...

  5. [2012]

    Esch, T., Marconcini, M., Felbier, A., Roth, A., Heldens, W., Huber, M., Schwinger, M., Taubenböck, H., Müller, A., and Dech, S.: Urban Footprint Processor – Fully Automated Processing Chain Generating Settlement Masks from Global Data of the TanDEM-X Mission, IEEE Geoscience ...

  6. [2013]

    Esch, T., Zeidler, J., Palacios-Lopez, D., Marconcini, M., Roth, A., Mönks, M., Leutner, B., Brzoska, E., Metz-Marconcini, A., Bachofer, F., Loekken, S., and Dech, S.: Towards a Large-Scale 3D Modeling of the Built Environment—Joint Analysis of TanDEM-X, Sentinel-2 and Open St...

  7. [2015]

    Pesaresi, M. and Politis, P.: GHS-BUILT-H R2023A - GHS building height, derived from AW3D30, SRTM30, and Sentinel-2 composite (2018), https://doi.org/10.2905/85005901-3A49-48DD-9D19-6261354F56FE, 2023a. Pesaresi, M. and Politis, P.: GHS-BUILT-V R2023A - GHS built-up volume gri...

  8. [2017]

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., et al.: An image is worth 16x16 words: Transformers for image recognition at scale, arXiv preprint arXiv:2010.11929,

  9. [2018]

    Zanaga, D., Van De Kerchove, R., De Keersmaecker, W., Souverijns, N., Brockmann, C., Quast, R., Wevers, J., Grosu, A., Paccini, A., Vergnaud, S., Cartus, O., Santoro, M., Fritz, S., Georgieva, I., Lesiv, M., Carter, S., Herold, M., Li, L., Tsendbazar, N., Ramoino, F., and Arin...

  10. [2019]

    The GDAL Development Team: GDAL - Geospatial Data Abstraction Library, https://gdal.org, accessed: 2023-04-02,

  11. [2020]

    Esch, T., Thiel, M., Schenk, A., Roth, A., Müller, A., and Dech, S.: Delineation of Urban Footprints From TerraSAR-X Data by Analyzing Speckle Characteristics and Intensity Information, IEEE Transactions on Geoscience and Remote Sensing, 48, 905–916, https://doi.org/10.1109/TG...

  12. [2021]

    Che, Y ., Li, X., Liu, X., Wang, Y ., Liao, W., Zheng, X., Zhang, X., Xu, X., Shi, Q., Zhu, J., Zhang, H., Yuan, H., and Dai, Y .: 3D-GloBFP: the first global three-dimensional building footprint dataset, Earth System Science Data, 16, 5357–5374, https://doi.org/10.5194/essd-1...

  13. [2022]

    Eurostat: Population distribution - GEOSTAT, https://ec.europa.eu/eurostat/web/gisco/geodata/population-distribution/geostat, accessed: 2025-04-16,

  14. [2024]

    Chen, S., Shi, Y ., Xiong, Z., and Zhu, X. X.: HTC-DC Net: Monocular Height Estimation From Single Remote Sensing Images, IEEE Transactions on Geoscience and Remote Sensing, 61, 1–18, https://doi.org/10.1109/TGRS.2023.3321255,

  15. [2025]

    X., Li, Q., Shi, Y ., Wang, Y ., Stewart, A., and Prexl, J.: GlobalBuildingMap – Unveiling the Mystery of Global Buildings, https: //arxiv.org/abs/2404.13911,

    Zhu, X. X., Li, Q., Shi, Y ., Wang, Y ., Stewart, A., and Prexl, J.: GlobalBuildingMap – Unveiling the Mystery of Global Buildings, https: //arxiv.org/abs/2404.13911,

Pith tools

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