Datasets:

feifei140729
/

HighBuild-1M

@@ -58,3 +58,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text

 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+benchmark_v1/manifest_tiles_1024.csv filter=lfs diff=lfs merge=lfs -text
+benchmark_v1/manifest_webdataset_tiles_1024.csv filter=lfs diff=lfs merge=lfs -text
+benchmark_v1/manifest_patches_256.csv filter=lfs diff=lfs merge=lfs -text
+benchmark_v1/splits/random_64_16_20/patches_256/train.txt filter=lfs diff=lfs merge=lfs -text

LICENSES.md ADDED Viewed

	@@ -0,0 +1,363 @@

+# Imagery Source Licenses and Attribution
+This file documents the aerial/satellite imagery sources used or configured by
+this project. It is intended for a NeurIPS 2026 Evaluations & Datasets data
+release package.
+NeurIPS 2026 E&D requires hosted datasets to be accessible to reviewers and to
+include Croissant metadata with core and Responsible AI fields. For this dataset,
+the Croissant metadata should reference this file from `prov:wasDerivedFrom`,
+`rai:dataCollection`, and the dataset/license documentation fields.
+This is an engineering license inventory, not legal advice. Before public
+release, re-check the official terms on the access date used in the paper and
+keep a frozen copy of the source URLs/capabilities metadata.
+## Release Rule
+The imagery in this dataset is multi-source. Do not publish the whole image
+collection under a single permissive license such as CC BY 4.0 unless every
+included image source is compatible with that license.
+Recommended release policy:
+- Release only sources marked `OK` or `OK-CONDITIONAL` after the listed
+  conditions are satisfied.
+- Exclude or regenerate sources marked `EXCLUDE` before the public NeurIPS
+  dataset release.
+- Keep each city/source subset separable in the hosted dataset and in Croissant
+  `FileSet` metadata.
+- Include all attribution strings in the dataset README, paper appendix, and
+  Croissant provenance fields.
+## Current Processed Data Source Map
+The following mapping is inferred from `src/sources/*.py` and the current
+`data/processed/images/*` directories.
+| Processed directories | Source key | Imagery provider / service | Release status |
+|---|---:|---|---|
+| `America_NewYork_Filtered`, `Chicago_buildings_full`, `LosAngeles_Downtown_buildings_full`, `NewYork_Manhattan_buildings_full`, `SanFrancisco_FiDi_buildings_full`, `Seattle_Downtown_buildings_full` | `usa` | Esri ArcGIS Online World Imagery | `EXCLUDE` |
+| `France_Lyon_buildings_full`, `France_Marseille_buildings_full`, `France_Strasbourg_buildings_full`, `France_Toulouse_buildings_full`, `paris_buildings_with_height_Filtered` | `fra` | IGN / Geoportail orthophotos | `OK` |
+| `Japan_Osaka_buildings_full`, `Japan_Osaka_buildings_Filtered` | `jpn` | GSI seamless aerial photo tiles | `OK-CONDITIONAL` |
+| `Amsterdam_buildings_Filtered` | `nld` | PDOK `luchtfotorgb` / `Actueel_orthoHR` | `OK` |
+| `HKG_Core` | `hkg` | Hong Kong LandsD Imagery Map API | `OK-CONDITIONAL` |
+| `TWN_Buildings_ready`, `TWN_Buildings_ready_Filtered` | `twn` | Taiwan NLSC PHOTO2 WMTS | `EXCLUDE` unless permission is confirmed |
+| `berlin_all_buildings_height_Filtered` | `deu` | Brandenburg/Berlin DOP20c WMS | `OK` |
+| `Germany_Frankfurt_buildings_full` | `deu` | Hessen DOP20 WMS | `OK` |
+| `Germany_Munich_buildings_full` | `deu` | Bavaria DOP20 WMS | `OK` |
+| `Oceania_Sydney_buildings_full` | `syd` | NSW Imagery MapServer | `OK-CONDITIONAL` |
+| `Oceania_Melbourne_buildings_full` | `mel` | City of Melbourne 2020 true ortho imagery | `OK-CONDITIONAL` |
+| `Africa_CapeTown_buildings_Filtered` | `afr` | City of Cape Town Aerial Imagery 2024 MapServer | `EXCLUDE` from unrestricted public release |
+| `sao_paulo_exact_Filtered` | `bax` | GeoSampa Ortofotos 2020 RGB WMS | `OK-CONDITIONAL` |
+| `Canada_Vancouver_buildings_Filtered` | `van` | City of Vancouver Orthophotos 2022 | `OK` |
+| `Canada_Toronto_Filtered` | `tor` | City of Toronto 2023 orthophoto MapServer | `OK` |
+| `Denmark_Aarhus_Filtered`, `Denmark_Copenhagen_Filtered`, `Denmark_Odense_Filtered` | `dam` | GeoDanmark Ortofoto / Datafordeler WMTS | `OK` |
+Implemented but not present in the current processed data:
+| Source key | Imagery provider / service | Release status |
+|---:|---|---|
+| `nzl` | LINZ Basemaps aerial imagery | `OK` |
+| `chn` | Tianditu imagery tiles | `EXCLUDE` unless separately licensed |
+## Source License Details
+### Esri World Imagery (`usa`) - EXCLUDE
+- Code: `src/sources/xyz_usa.py`
+- Service URL: `https://server.arcgisonline.com/ArcGIS/rest/services/World_Imagery/MapServer`
+- Official terms found: Esri World Imagery is licensed under the Esri Master
+  License Agreement. Esri states the ordinary World Imagery layer is not
+  intended for exporting tiles for offline use; the export layer is for small
+  offline use in ArcGIS contexts.
+- Dataset release decision: Do not redistribute image chips generated from this
+  source in a public NeurIPS dataset.
+- Required action: Regenerate all USA image subsets from a redistributable
+  source such as USDA NAIP/USGS public imagery, or exclude the USA directories
+  from the public release.
+Affected current directories:
+- `America_NewYork_Filtered`
+- `Chicago_buildings_full`
+- `LosAngeles_Downtown_buildings_full`
+- `NewYork_Manhattan_buildings_full`
+- `SanFrancisco_FiDi_buildings_full`
+- `Seattle_Downtown_buildings_full`
+### IGN / Geoportail Orthophotos (`fra`) - OK
+- Code: `src/sources/xyz_fra.py`
+- Service URL: `https://data.geopf.fr/wmts`
+- Layer: `ORTHOIMAGERY.ORTHOPHOTOS`
+- License: Licence Ouverte / Open Licence version 2.0 (Etalab 2.0), for public
+  IGN open data including BD ORTHO / ortho-imagery.
+- Redistribution: allowed with attribution.
+- Attribution: `Contains orthophotography from IGN / Geoportail, Licence Ouverte / Etalab 2.0.`
+- Croissant `prov:wasDerivedFrom`: `https://data.geopf.fr/wmts`
+### GSI Seamless Photo (`jpn`) - OK-CONDITIONAL
+- Code: `src/sources/xyz_jpn.py`
+- Service URL: `https://cyberjapandata.gsi.go.jp/xyz/seamlessphoto/{z}/{x}/{y}.jpg`
+- Dataset: GSI Tiles, latest seamless aerial photo (`全国最新写真（シームレス）`).
+- License / terms: Government of Japan Standard Terms of Use 2.0 / GSI content
+  terms, with source citation. GSI notes that GSI tiles can include tiles from
+  third-party organizations or legally restricted survey products.
+- Redistribution: acceptable only after checking that the specific seamless
+  photo tiles used for Osaka do not require an additional permission beyond
+  source citation.
+- Attribution: `Source: Geospatial Information Authority of Japan (GSI) / GSI Tiles. Modified for dataset generation.`
+- Croissant `prov:wasDerivedFrom`: `https://maps.gsi.go.jp/development/`
+### PDOK Netherlands Orthophotos (`nld`) - OK
+- Code: `src/sources/xyz_nld.py`
+- Service URL: `https://service.pdok.nl/hwh/luchtfotorgb/wmts/v1_0/Actueel_orthoHR/`
+- Dataset/layer: `Actueel_orthoHR`
+- License: use the dataset-specific Nationaal Georegister/PDOK metadata; PDOK
+  commonly uses CC BY 4.0 for geographic works and instructs users to follow the
+  NGR metadata for the specific dataset.
+- Redistribution: allowed if the dataset metadata license is CC BY 4.0.
+- Attribution: `Contains PDOK / Dutch aerial imagery data, CC BY 4.0; see PDOK/NGR metadata.`
+- Croissant `prov:wasDerivedFrom`: `https://service.pdok.nl/hwh/luchtfotorgb/wmts/v1_0/`
+### Hong Kong LandsD Imagery Map API (`hkg`) - OK-CONDITIONAL
+- Code: `src/sources/xyz_hkg.py`
+- Service URL: `https://mapapi.geodata.gov.hk/gs/api/v1.0.0/xyz/imagery/WGS84/`
+- Provider: Lands Department, Government of the Hong Kong SAR.
+- Official terms found: CSDI/LandsD Map API terms allow use of data through the
+  API subject to terms, rate limits, IP rights notice, and attribution. The API
+  documentation requires Lands Department attribution and copyright notice.
+- Risk: LandsD has separate notices for downloadable aerial photographs that
+  restrict third-party dissemination. Because this project stores and republishes
+  image chips rather than only displaying live map tiles, confirm that the Map
+  API / CSDI terms cover stored derivative image redistribution.
+- Attribution: `Aerial Photograph from Lands Department, Government of the Hong Kong SAR.`
+- Release condition: Include the attribution and, if required by LandsD, the
+  Lands Department logo/copyright notice. Obtain written confirmation if possible.
+- Croissant `prov:wasDerivedFrom`: `https://tools.csdi.gov.hk/csdi-webpage/apidoc/ImageryMapAPI`
+### Taiwan NLSC PHOTO2 (`twn`) - EXCLUDE unless permission is confirmed
+- Code: `src/sources/xyz_twn.py`
+- Service URL: `https://wmts.nlsc.gov.tw/wmts/PHOTO2/default/GoogleMapsCompatible/`
+- Provider: National Land Surveying and Mapping Center (NLSC), Taiwan.
+- Official terms found: NLSC terms allow showing captured/extracted service
+  contents on the Internet, videos, print advertisements, and theses for legal
+  purposes, but state that bulk download is forbidden.
+- Risk: this dataset generation is a bulk tile extraction and public
+  redistribution workflow.
+- Dataset release decision: Exclude Taiwan imagery from the public release
+  unless NLSC confirms redistribution is permitted for this derived dataset.
+- Attribution if permission is obtained: `Source: National Land Surveying and Mapping Center (NLSC), Taiwan.`
+### Germany Berlin / Brandenburg DOP20c (`deu`) - OK
+- Code: `src/sources/xyz_deu.py`
+- Service URL: `https://isk.geobasis-bb.de/mapproxy/dop20c/service/wms`
+- Provider: Landesvermessung und Geobasisinformation Brandenburg, with Berlin
+  data attribution where applicable.
+- License: Datenlizenz Deutschland - Namensnennung - Version 2.0
+  (`dl-de/by-2-0`).
+- Redistribution: allowed with attribution.
+- Attribution: `© GeoBasis-DE/LGB, dl-de/by-2-0; © Geoportal Berlin, dl-de/by-2-0 (Daten geändert)`
+- Croissant `prov:wasDerivedFrom`: `https://isk.geobasis-bb.de/mapproxy/dop20c/service/wms`
+### Germany Hessen DOP20 (`deu`) - OK
+- Code: `src/sources/xyz_deu.py`
+- Service URL: `https://www.gds-srv.hessen.de/cgi-bin/lika-services/ogc-free-images.ows`
+- Layer: `he_dop20_rgb`
+- Provider: Hessische Verwaltung fuer Bodenmanagement und Geoinformation (HVBG).
+- License: Datenlizenz Deutschland - Zero - Version 2.0 (`dl-zero-de/2.0`).
+- Redistribution: allowed without attribution requirement under dl-zero, but
+  citation is recommended for provenance.
+- Attribution / citation: `Source: Geoportal Hessen / HVBG, dl-zero-de/2.0.`
+- Croissant `prov:wasDerivedFrom`: `https://www.gds-srv.hessen.de/cgi-bin/lika-services/ogc-free-images.ows`
+### Germany Bavaria DOP20 (`deu`) - OK
+- Code: `src/sources/xyz_deu.py`
+- Service URL: `https://geoservices.bayern.de/od/wms/dop/v1/dop20`
+- Layer: `by_dop20c`
+- Provider: Bayerische Vermessungsverwaltung.
+- License: Creative Commons Attribution 4.0 International (CC BY 4.0).
+- Attribution required by provider: `Bayerische Vermessungsverwaltung - www.geodaten.bayern.de`
+- Croissant `prov:wasDerivedFrom`: `https://geoservices.bayern.de/od/wms/dop/v1/dop20`
+### NSW Imagery (`syd`) - OK-CONDITIONAL
+- Code: `src/sources/xyz_syd.py`
+- Service URL: `https://maps.six.nsw.gov.au/arcgis/rest/services/public/NSW_Imagery/MapServer`
+- Provider/copyright in service metadata: Department of Customer Service, NSW.
+- License status: the MapServer metadata gives copyright but not a clear open
+  redistribution license in the codebase.
+- Release condition: confirm the applicable NSW open data license or obtain a
+  redistribution statement before including Sydney image chips in the public
+  release.
+- Attribution: `© Department of Customer Service, NSW; contains external imagery sources as listed in NSW_Imagery service metadata.`
+### City of Melbourne 2020 Aerial Imagery (`mel`) - OK-CONDITIONAL
+- Code: `src/sources/xyz_mel.py`
+- Service URL: `https://gisags.melbourne.vic.gov.au/server_wa/rest/services/AerialImageryWGS84/AerialImage2020_WGS84/MapServer`
+- Dataset: City of Melbourne 2020 Aerial Imagery true ortho.
+- License: source portal metadata lists CC BY / CC BY-compatible terms. Confirm
+  the exact version (`CC BY 4.0` versus another CC BY variant) before final
+  Croissant publication.
+- Redistribution: allowed if released under the source CC BY terms.
+- Attribution: `Contains City of Melbourne 2020 Aerial Imagery, CC BY.`
+- Croissant `prov:wasDerivedFrom`: `https://data.melbourne.vic.gov.au/explore/dataset/2020-aerial-imagery-true-ortho/`
+### City of Cape Town Aerial Imagery 2024 (`afr`) - EXCLUDE from unrestricted public release
+- Code: `src/sources/xyz_afr.py`
+- Service URL: `https://cityimg.capetown.gov.za/erdas-iws/esri/GeoSpatial%20Datasets/rest/services/Aerial%20Imagery_Aerial%20Imagery%202024/MapServer`
+- Provider/service text: City of Cape Town aerial imagery 2024. The service
+  states that the original imagery is property of CCT and that there are no
+  restrictions on the digital file for non-commercial purposes.
+- Risk: non-commercial-only permissions are not compatible with an unrestricted
+  public ML dataset license.
+- Dataset release decision: either exclude this subset from the unrestricted
+  public release, release it as a clearly separate non-commercial subset, or
+  obtain written permission for dataset redistribution.
+- Attribution: `Aerial imagery property of City of Cape Town (CCT), 2024.`
+### GeoSampa Sao Paulo Ortofotos 2020 (`bax`) - OK-CONDITIONAL
+- Code: `src/sources/xyz_bax.py`
+- Service URL: `https://raster.geosampa.prefeitura.sp.gov.br/geoserver/geoportal/wms`
+- Layer: `ORTO_RGB_2020`
+- Provider: Prefeitura Municipal de Sao Paulo / GeoSampa.
+- License: Creative Commons Attribution Share-Alike 4.0 (CC BY-SA 4.0), per
+  GeoSampa license notice.
+- Redistribution: allowed with attribution and share-alike terms.
+- Release condition: keep the Sao Paulo imagery subset under CC BY-SA 4.0 and
+  do not label it as merely CC BY.
+- Attribution: `Source: GeoSampa / Prefeitura Municipal de Sao Paulo, CC BY-SA 4.0.`
+- Croissant `prov:wasDerivedFrom`: `https://geosampa.prefeitura.sp.gov.br/`
+### City of Vancouver Orthophotos 2022 (`van`) - OK
+- Code: `src/sources/xyz_can.py`
+- Service URL: `https://tiles.arcgis.com/tiles/qrcTTRTwUoS8N47o/arcgis/rest/services/Orthophotos_2022/MapServer`
+- Dataset: City of Vancouver Orthophoto imagery 2022.
+- License: Open Government Licence - Vancouver.
+- Redistribution: allowed with attribution under the open data terms.
+- Attribution: `Contains information licensed under the Open Government Licence - Vancouver; source: City of Vancouver Orthophotos 2022.`
+- Croissant `prov:wasDerivedFrom`: `https://opendata.vancouver.ca/explore/dataset/orthophoto-imagery-2022/`
+### City of Toronto Orthophotos 2023 (`tor`) - OK
+- Code: `src/sources/xyz_can.py`
+- Service URL: `https://gis.toronto.ca/arcgis/rest/services/basemap/cot_ortho_2023_color_10cm/MapServer`
+- Dataset: City of Toronto 2023 colour orthophoto, 10 cm.
+- License: Open Government Licence - Toronto.
+- Redistribution: allowed with attribution.
+- Attribution: `Contains information licensed under the Open Government Licence - Toronto; source: City of Toronto orthophoto imagery.`
+- Croissant `prov:wasDerivedFrom`: `https://gis.toronto.ca/arcgis/rest/services/basemap/cot_ortho_2023_color_10cm/MapServer`
+### GeoDanmark Ortofoto / Datafordeler (`dam`) - OK
+- Code: `src/sources/xyz_dam.py`
+- Service URL: `https://wmts.datafordeler.dk/GeoDanmarkOrto/orto_foraar_webm/1.0.0/WMTS`
+- Dataset: GeoDanmark Ortofoto foraar Web Mercator WMTS.
+- License: Creative Commons Attribution 4.0 International (CC BY 4.0).
+- Redistribution: allowed with attribution.
+- Attribution required by GeoDanmark: `@geodanmark` with a link to the GeoDanmark data terms.
+- Operational note: the WMTS requires API-key/OAuth access. Do not publish the
+  `SDFI_API_KEY`; publish only the derived data and source metadata.
+- Croissant `prov:wasDerivedFrom`: `https://datafordeler.dk/dataoversigt/geodanmark-ortofoto/ortofoto-foraar-web-mercator-wmts/`
+### LINZ Basemaps Aerial (`nzl`) - OK
+- Code: `src/sources/xyz_nzl.py`
+- Current data: no processed directory in this release.
+- Service URL: `https://basemaps.linz.govt.nz/v1/tiles/aerial/WebMercatorQuad/`
+- License: Creative Commons Attribution 4.0 International (CC BY 4.0), with
+  LINZ Basemaps attribution requirements and contributor attribution.
+- Attribution: `Sourced from LINZ. CC BY 4.0. Contains LINZ Basemaps aerial imagery and contributors.`
+- Croissant `prov:wasDerivedFrom`: `https://www.linz.govt.nz/products-services/data/licensing-and-using-data/attributing-linz-basemaps-data`
+### Tianditu (`chn`) - EXCLUDE unless separately licensed
+- Code: `src/sources/xyz_chn.py`
+- Current data: no processed directory in this release.
+- Service URL: `https://t0.tianditu.gov.cn/DataServer?T=img_w...`
+- License status: not audited in this project.
+- Dataset release decision: do not include Tianditu-derived imagery in the
+  public release without a separate redistribution license.
+## Recommended Public Release Subsets
+If publishing now, the lowest-risk public release should include only:
+- France: `fra`
+- Netherlands: `nld`
+- Germany: `deu` for Berlin/Brandenburg, Hessen, Bavaria
+- Vancouver: `van`
+- Toronto: `tor`
+- Denmark: `dam`
+- Sao Paulo: `bax`, but keep CC BY-SA 4.0 terms separate
+- Japan: `jpn`, after tile-specific source citation checks
+- Hong Kong: `hkg`, after Map API stored-derivative redistribution check
+- Melbourne: `mel`, after exact CC BY version check
+Exclude or regenerate before unrestricted public release:
+- USA / Esri World Imagery (`usa`)
+- Taiwan NLSC PHOTO2 (`twn`) unless permission is confirmed
+- Cape Town 2024 (`afr`) unless released as non-commercial-only or permission is obtained
+- Sydney NSW imagery (`syd`) until an explicit redistributable license is confirmed
+- Tianditu (`chn`) if ever used
+## Suggested Dataset-Level License Text
+Use this text in the dataset README and hosting page instead of assigning a
+single false license:
+> This dataset is a multi-source derived dataset. Image chips are licensed under
+> the terms of their respective upstream imagery providers, as documented in
+> `data/IMAGERY_LICENSES.md`. Building geometry, height labels, masks, and
+> metadata have separate source provenance documented in the dataset card and
+> Croissant metadata. Users must comply with the license and attribution
+> requirements for each source subset.
+For a NeurIPS-hosted package, prefer splitting files by source, for example:
+- `images/fra/...`
+- `images/deu_berlin_brandenburg/...`
+- `images/deu_hessen/...`
+- `images/deu_bavaria/...`
+- `images/nld/...`
+- `images/can_vancouver/...`
+- `images/can_toronto/...`
+- `images/dam/...`
+- `images/bax/...`
+Then assign per-subset license/provenance in Croissant `FileSet` entries.
+## Sources Consulted
+- NeurIPS 2026 E&D Call: `https://neurips.cc/Conferences/2026/CallForEvaluationsDatasets`
+- NeurIPS 2026 E&D Hosting Guidelines: `https://neurips.cc/Conferences/2026/EvaluationsDatasetsHosting`
+- Esri World Imagery item / terms notice: `https://www.arcgis.com/home/item.html?id=10df2279f9684e4a9f6a7f08febac2a9`
+- IGN / Geoportail orthophoto open data notices: `https://data.geopf.fr/wmts`, `https://www.data.gouv.fr/`
+- GSI tiles and terms: `https://maps.gsi.go.jp/development/`, `https://www.gsi.go.jp/ENGLISH/page_e30286.html`
+- PDOK copyright and NGR metadata guidance: `https://www.pdok.nl/copyright`
+- Hong Kong CSDI / LandsD Map API docs and terms: `https://tools.csdi.gov.hk/csdi-webpage/apidoc/ImageryMapAPI`
+- Taiwan NLSC terms: `https://maps.nlsc.gov.tw/pro/use_clause_en.jsp`
+- Brandenburg/Berlin DOP20c metadata: `https://geobroker.geobasis-bb.de/`
+- Hessen DOP20 metadata: `https://www.geoportal.hessen.de/`
+- Bavaria DOP20 metadata: `https://geodatenonline.bayern.de/geodatenonline/seiten/wms_dop20cm`
+- NSW Imagery MapServer: `https://maps.six.nsw.gov.au/arcgis/rest/services/public/NSW_Imagery/MapServer`
+- City of Melbourne 2020 Aerial Imagery: `https://data.melbourne.vic.gov.au/explore/dataset/2020-aerial-imagery-true-ortho/`
+- City of Cape Town Aerial Imagery 2024 MapServer: `https://cityimg.capetown.gov.za/erdas-iws/esri/GeoSpatial%20Datasets/rest/services/Aerial%20Imagery_Aerial%20Imagery%202024/MapServer`
+- GeoSampa license notice: `https://prefeitura.sp.gov.br/web/licenciamento/w/licen%C3%A7a-para-uso-de-dados-do-geosampa`
+- City of Vancouver Orthophoto imagery 2022 and terms: `https://opendata.vancouver.ca/explore/dataset/orthophoto-imagery-2022/`
+- City of Toronto Open Government Licence: `https://open.toronto.ca/open-data-licence/`
+- GeoDanmark / Datafordeler Ortofoto terms: `https://www.geodanmark.dk/home/vejledninger/vilkaar-for-data-anvendelse/`
+- LINZ Basemaps attribution: `https://www.linz.govt.nz/products-services/data/licensing-and-using-data/attributing-linz-basemaps-data`

README.md ADDED Viewed

	@@ -0,0 +1,263 @@

+---
+pretty_name: Global Building Height Dataset
+license: other
+task_categories:
+- image-segmentation
+- object-detection
+tags:
+- remote-sensing
+- geospatial
+- building-height
+- aerial-imagery
+- earth-observation
+size_categories:
+- 10K<n<100K
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: data/webdataset/train/*.tar
+  - split: validation
+    path: data/webdataset/validation/*.tar
+  - split: test
+    path: data/webdataset/test/*.tar
+---
+# Global Building Height Dataset
+This dataset contains high-resolution aerial or satellite image tiles paired
+with raster building-height masks and COCO-style building annotations. It is
+intended for building height estimation, building-level remote sensing
+understanding, geospatial computer vision, and benchmark evaluation.
+The public release excludes Taiwan and Hong Kong subsets. The included release
+contains 61,215 matched 1024 x 1024 samples across 24 city folders, 10
+countries, and 6 continents.
+## WebDataset and Compression
+For the Hugging Face hosted release, the recommended distribution format is
+WebDataset TAR shards generated from this staging directory:
+```bash
+python3 scripts/build_hf_webdataset.py \
+  --input hf_upload_staging \
+  --output hf_upload_webdataset \
+  --clean \
+  --workers 8 \
+  --max-shard-gb 1.0
+```
+The conversion writes split-aware shards under:
+```text
+data/webdataset/train/*.tar
+data/webdataset/validation/*.tar
+data/webdataset/test/*.tar
+```
+The conversion also writes
+`benchmark_v1/manifest_webdataset_tiles_1024.csv`, which maps each original
+tile id to its split, shard path, WebDataset key, and member filenames.
+Each WebDataset sample contains three members with the same key:
+```text
+<Continent>_<Country>_<City>__<tile_id>.jpg
+<Continent>_<Country>_<City>__<tile_id>.tiff
+<Continent>_<Country>_<City>__<tile_id>.json
+```
+The TAR shards themselves are not gzip/zstd-compressed, which keeps them
+streamable. Before each mask is written into a shard, the mask TIFF is
+losslessly recompressed with internal TIFF ZSTD compression:
+```text
+COMPRESS=ZSTD
+ZSTD_LEVEL=9
+PREDICTOR=3
+TILED=YES
+BLOCKXSIZE=512
+BLOCKYSIZE=512
+BIGTIFF=IF_SAFER
+```
+This changes only the TIFF storage encoding. No resampling, quantization, dtype
+conversion, or lossy compression is applied to mask values. The RGB JPEG images
+and COCO-style JSON annotations are copied into the shards without
+recompression.
+In the local staging package, mask TIFF files account for about 240 GB of the
+270 GB total. A local random sample of 50 masks compressed from 4.2 MB per mask
+to about 27 KB on average with identical GDAL checksums. Based on that sample,
+the mask portion is expected to shrink to roughly 1.5 GB, and the final
+WebDataset release is expected to be dominated by the existing JPEG imagery and
+JSON annotations.
+## Dataset Structure
+The original unsharded layout represented each sample by three matched files with the same tile basename:
+```text
+data/images/<Continent>_<Country>_<City>/<tile_id>.jpg
+data/masks/<Continent>_<Country>_<City>/masks/<tile_id>.tiff
+data/annotations/coco_json/<Continent>_<Country>_<City>/<tile_id>.json
+```
+The release also includes benchmark metadata:
+```text
+benchmark_v1/manifest_tiles_1024.csv
+benchmark_v1/manifest_patches_256.csv
+benchmark_v1/city_coverage.csv
+benchmark_v1/split_report.md
+benchmark_v1/splits/random_64_16_20/tiles_1024/{train,val,test}.txt
+benchmark_v1/splits/random_64_16_20/patches_256/{train,val,test}.txt
+```
+Paths in `manifest_tiles_1024.csv` describe the original unsharded layout. For the hosted WebDataset release, use `manifest_webdataset_tiles_1024.csv` for shard and member paths.
+## Included Cities
+The folder naming convention is `<Continent>_<Country>_<City>`.
+| Folder | Tiles |
+|---|---:|
+| `Africa_SouthAfrica_CapeTown` | 5,473 |
+| `Asia_Japan_Osaka` | 1,554 |
+| `Europe_Denmark_Aarhus` | 95 |
+| `Europe_Denmark_Copenhagen` | 1,618 |
+| `Europe_Denmark_Odense` | 90 |
+| `Europe_France_Lyon` | 101 |
+| `Europe_France_Marseille` | 159 |
+| `Europe_France_Paris` | 6,294 |
+| `Europe_France_Strasbourg` | 94 |
+| `Europe_France_Toulouse` | 231 |
+| `Europe_Germany_Berlin` | 10,355 |
+| `Europe_Germany_Frankfurt` | 99 |
+| `Europe_Germany_Munich` | 108 |
+| `Europe_Netherlands_Amsterdam` | 1,836 |
+| `NorthAmerica_Canada_Toronto` | 8,471 |
+| `NorthAmerica_Canada_Vancouver` | 126 |
+| `NorthAmerica_USA_Chicago` | 120 |
+| `NorthAmerica_USA_LosAngeles` | 123 |
+| `NorthAmerica_USA_NewYork` | 11,172 |
+| `NorthAmerica_USA_SanFrancisco` | 87 |
+| `NorthAmerica_USA_Seattle` | 78 |
+| `Oceania_Australia_Melbourne` | 602 |
+| `Oceania_Australia_Sydney` | 138 |
+| `SouthAmerica_Brazil_SaoPaulo` | 12,191 |
+## Splits
+The benchmark split is `random_64_16_20`:
+| Split | 1024 tiles | 256 patches | Ratio |
+|---|---:|---:|---:|
+| train | 39,178 | 626,848 | 64% |
+| validation | 9,794 | 156,704 | 16% |
+| test | 12,243 | 195,888 | 20% |
+All 256 x 256 patches inherit the split of their parent 1024 x 1024 tile to
+avoid leakage between train, validation, and test sets.
+## Data Fields
+- `image`: RGB image tile in JPEG format.
+- `mask`: TIFF raster mask aligned to the image tile. Pixel values encode the
+  building-height target used by the dataset generation pipeline.
+- `annotation`: COCO-style JSON annotation file for buildings in the tile.
+- `manifest_tiles_1024.csv`: one row per complete image-mask-annotation tile.
+- `manifest_patches_256.csv`: one row per 256 x 256 patch derived from the
+  1024 tile grid.
+## Reading and Extracting WebDataset Shards
+Most users do not need to manually decompress the mask TIFF files. GDAL,
+rasterio, and other TIFF readers with ZSTD-enabled libtiff support decompress
+the internal TIFF compression transparently when the array is read.
+Load the hosted WebDataset with Hugging Face Datasets:
+```python
+from datasets import load_dataset
+data_files = {
+    "train": "data/webdataset/train/*.tar",
+    "validation": "data/webdataset/validation/*.tar",
+    "test": "data/webdataset/test/*.tar",
+}
+dataset = load_dataset("webdataset", data_files=data_files, streaming=True)
+```
+Extract one shard with standard tar tools:
+```bash
+mkdir -p extracted/train
+tar -xf data/webdataset/train/train-000000.tar -C extracted/train
+```
+Read an extracted mask directly with rasterio:
+```python
+import rasterio
+with rasterio.open("extracted/train/Africa_SouthAfrica_CapeTown__grid_03328_z18.tiff") as src:
+    mask = src.read(1)
+```
+If an uncompressed TIFF file is required for a legacy tool, convert the mask
+back to an uncompressed TIFF with GDAL:
+```bash
+gdal_translate \
+  extracted/train/Africa_SouthAfrica_CapeTown__grid_03328_z18.tiff \
+  extracted/train/Africa_SouthAfrica_CapeTown__grid_03328_z18.uncompressed.tiff \
+  -co COMPRESS=NONE
+```
+To inspect the TIFF compression and checksum:
+```bash
+gdalinfo -checksum extracted/train/Africa_SouthAfrica_CapeTown__grid_03328_z18.tiff
+```
+## License and Attribution
+This is a multi-source geospatial dataset. The imagery and building-height
+labels are derived from public or permissioned regional sources with different
+license and attribution requirements. Do not treat the whole dataset as a
+single permissive-license corpus.
+See `LICENSES.md` for the source-level license inventory, attribution strings,
+and release conditions.
+## Intended Uses
+- Benchmarking monocular or single-image building height estimation.
+- Training and evaluating geospatial computer vision models.
+- Studying cross-city, cross-country, and cross-continent generalization.
+- Remote-sensing research on urban morphology and built environments.
+## Out-of-Scope Uses
+This dataset should not be used as the sole basis for legal, safety-critical,
+real-estate, insurance, tax, emergency-response, or infrastructure decisions.
+It should not be used to infer sensitive attributes about individuals or
+households.
+## Limitations
+- Coverage is uneven across continents, countries, and cities.
+- Source imagery dates and building-height label dates may not match exactly.
+- Spatial resolution, acquisition conditions, sensor characteristics, and
+  building-height definitions differ across source regions.
+- Some dense urban regions dominate the tile count.
+- The benchmark split is random at the 1024 tile level and is not a strict
+  geographic holdout split.
+## Citation
+Citation information will be added after the associated paper or dataset
+technical report is finalized.

benchmark_v1/city_coverage.csv ADDED Viewed

	@@ -0,0 +1,25 @@

+public_dir,source_dir,continent,country,city,tiles_1024,patches_256
+Africa_SouthAfrica_CapeTown,Africa_CapeTown_buildings_Filtered,Africa,SouthAfrica,CapeTown,5473,87568
+Asia_Japan_Osaka,Japan_Osaka_buildings_full,Asia,Japan,Osaka,1554,24864
+Europe_Denmark_Aarhus,Denmark_Aarhus_Filtered,Europe,Denmark,Aarhus,95,1520
+Europe_Denmark_Copenhagen,Denmark_Copenhagen_Filtered,Europe,Denmark,Copenhagen,1618,25888
+Europe_Denmark_Odense,Denmark_Odense_Filtered,Europe,Denmark,Odense,90,1440
+Europe_France_Lyon,France_Lyon_buildings_full,Europe,France,Lyon,101,1616
+Europe_France_Marseille,France_Marseille_buildings_full,Europe,France,Marseille,159,2544
+Europe_France_Paris,paris_buildings_with_height_Filtered,Europe,France,Paris,6294,100704
+Europe_France_Strasbourg,France_Strasbourg_buildings_full,Europe,France,Strasbourg,94,1504
+Europe_France_Toulouse,France_Toulouse_buildings_full,Europe,France,Toulouse,231,3696
+Europe_Germany_Berlin,berlin_all_buildings_height_Filtered,Europe,Germany,Berlin,10355,165680
+Europe_Germany_Frankfurt,Germany_Frankfurt_buildings_full,Europe,Germany,Frankfurt,99,1584
+Europe_Germany_Munich,Germany_Munich_buildings_full,Europe,Germany,Munich,108,1728
+Europe_Netherlands_Amsterdam,Amsterdam_buildings_Filtered,Europe,Netherlands,Amsterdam,1836,29376
+NorthAmerica_Canada_Toronto,Canada_Toronto_Filtered,NorthAmerica,Canada,Toronto,8471,135536
+NorthAmerica_Canada_Vancouver,Canada_Vancouver_buildings_Filtered,NorthAmerica,Canada,Vancouver,126,2016
+NorthAmerica_USA_Chicago,Chicago_buildings_full,NorthAmerica,USA,Chicago,120,1920
+NorthAmerica_USA_LosAngeles,LosAngeles_Downtown_buildings_full,NorthAmerica,USA,LosAngeles,123,1968
+NorthAmerica_USA_NewYork,America_NewYork_Filtered,NorthAmerica,USA,NewYork,11172,178752
+NorthAmerica_USA_SanFrancisco,SanFrancisco_FiDi_buildings_full,NorthAmerica,USA,SanFrancisco,87,1392
+NorthAmerica_USA_Seattle,Seattle_Downtown_buildings_full,NorthAmerica,USA,Seattle,78,1248
+Oceania_Australia_Melbourne,Oceania_Melbourne_buildings_full,Oceania,Australia,Melbourne,602,9632
+Oceania_Australia_Sydney,Oceania_Sydney_buildings_full,Oceania,Australia,Sydney,138,2208
+SouthAmerica_Brazil_SaoPaulo,sao_paulo_exact_Filtered,SouthAmerica,Brazil,SaoPaulo,12191,195056

benchmark_v1/manifest_patches_256.csv ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:ef636163272ea5627f89247f54d967106ab491803ea200b1c965fc775db55925
+size 154439117

benchmark_v1/manifest_tiles_1024.csv ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:ab72322507e3a16716f60cb0f8fd23bd9fce443654724b19640eef2281709efc
+size 20756891

benchmark_v1/manifest_webdataset_tiles_1024.csv ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:7d016c7f0d15037a6c69e61ca090fa12550c89d97844f64a91502c707e03f5fb
+size 34621025

benchmark_v1/split_report.md ADDED Viewed

	@@ -0,0 +1,84 @@

+# Benchmark V1 Split Report
+## Overall
+- split_name: random_64_16_20
+- split_seed: 42
+- split_policy: global 80/20 trainval/test, then 80/20 train/validation within trainval
+- excluded_cities: Taiwan, HongKong
+- tiles_1024: 61215
+- patches_256: 979440
+- cities: 24
+- countries: 10
+- continents: 6
+- annotations: 5222816
+- unknown_height_annotations: 0
+- estimated_height_annotations: 18548
+## Split Counts
+- train: tiles=39178, patches_256=626848, ratio=64.00%
+- val: tiles=9794, patches_256=156704, ratio=16.00%
+- test: tiles=12243, patches_256=195888, ratio=20.00%
+## Tiles By Continent
+- Africa: 5473
+- Asia: 1554
+- Europe: 21080
+- NorthAmerica: 20177
+- Oceania: 740
+- SouthAmerica: 12191
+## Tiles By Country
+- Australia: 740
+- Brazil: 12191
+- Canada: 8597
+- Denmark: 1803
+- France: 6879
+- Germany: 10562
+- Japan: 1554
+- Netherlands: 1836
+- SouthAfrica: 5473
+- USA: 11580
+## Tiles By City
+- Africa_SouthAfrica_CapeTown: 5473
+- Asia_Japan_Osaka: 1554
+- Europe_Denmark_Aarhus: 95
+- Europe_Denmark_Copenhagen: 1618
+- Europe_Denmark_Odense: 90
+- Europe_France_Lyon: 101
+- Europe_France_Marseille: 159
+- Europe_France_Paris: 6294
+- Europe_France_Strasbourg: 94
+- Europe_France_Toulouse: 231
+- Europe_Germany_Berlin: 10355
+- Europe_Germany_Frankfurt: 99
+- Europe_Germany_Munich: 108
+- Europe_Netherlands_Amsterdam: 1836
+- NorthAmerica_Canada_Toronto: 8471
+- NorthAmerica_Canada_Vancouver: 126
+- NorthAmerica_USA_Chicago: 120
+- NorthAmerica_USA_LosAngeles: 123
+- NorthAmerica_USA_NewYork: 11172
+- NorthAmerica_USA_SanFrancisco: 87
+- NorthAmerica_USA_Seattle: 78
+- Oceania_Australia_Melbourne: 602
+- Oceania_Australia_Sydney: 138
+- SouthAmerica_Brazil_SaoPaulo: 12191
+## Split Countries
+- train: {'Australia': 464, 'Brazil': 7803, 'Canada': 5541, 'Denmark': 1134, 'France': 4401, 'Germany': 6760, 'Japan': 970, 'Netherlands': 1180, 'SouthAfrica': 3561, 'USA': 7364}
+- val: {'Australia': 122, 'Brazil': 1963, 'Canada': 1320, 'Denmark': 302, 'France': 1149, 'Germany': 1706, 'Japan': 248, 'Netherlands': 301, 'SouthAfrica': 873, 'USA': 1810}
+- test: {'Australia': 154, 'Brazil': 2425, 'Canada': 1736, 'Denmark': 367, 'France': 1329, 'Germany': 2096, 'Japan': 336, 'Netherlands': 355, 'SouthAfrica': 1039, 'USA': 2406}
+## Notes
+- Split files are generated at `splits/random_64_16_20/`.
+- All 256 patches inherit the split of their parent 1024 tile.
+- Paths in manifests are relative to the Hugging Face dataset repository root.
+- `city_coverage.csv` lists every included release folder and its source directory.

benchmark_v1/splits/random_64_16_20/patches_256/test.txt ADDED Viewed