isp-uv-es/openearthmap-taco

OpenEarthMap

This is a repackaging, not a new dataset. It is OpenEarthMap by University of Tokyo / RIKEN (Xia et al.), converted to TACO. Pixel values and labels are kept as released except where the description below says otherwise. All credit belongs to the original authors: if you use it, please cite them and follow their licence.

original dataset · paper · licence: CC-BY-NC-SA-4.0

Repackaged into TACO by the Image and Signal Processing Group (ISP), Universitat de València, within the ELLIOT project.

Citation

Please cite the original work:

@inproceedings{xia2023openearthmap,
  title     = {OpenEarthMap: A Benchmark Dataset for Global High-Resolution Land Cover Mapping},
  author    = {Xia, Junshi and Yokoya, Naoto and Adriano, Bruno and Broni-Bediako, Clifford},
  booktitle = {IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  pages     = {6243--6253},
  year      = {2023},
  doi       = {10.1109/WACV56688.2023.00619}
}

About the data

3500 aerial and satellite RGB tiles at 0.25-0.5 m over 97 regions on six continents, with a dense 9-class land-cover mask. Tiles are 1000x1000, except the xBD-derived ones at 1024x1024. Converted from the original GeoTIFFs.

3,500 samples · splits: train 3,000 · validation 500 · tasks: semantic-segmentation

Packaged as TACO v3.

Full description

Splits. The official split, train and val only: 3000 / 500. It is per-tile, not per-region: every region appears on both sides, so it measures within-region accuracy. Regroup on the region column for cross-region generalisation. The 1500 test-split tiles, whose masks were never released, are in the held-out half below.

xBD tiles. 813 tiles are xBD (xView2) frames that OpenEarthMap re-annotates and deliberately does not redistribute. Their RGB is joined back in from the xView2 train, tier3 and hold trees through the release's own xbd_files.csv; they carry source=xbd and are 1024x1024 rather than 1000x1000.

Resolution. The release mixes 0.25 m and 0.5 m imagery without publishing which is which, so no GSD is declared on the bands and each tile records its own pixel size.

Legend. The native legend is kept verbatim and additionally mapped onto ml:lc11, where building, road and developed space all collapse onto built_up.

Held-out half

openearthmap_test.zip holds the 1,500 samples whose targets the publisher withheld. Same inputs, no target slots: it is there to be predicted on and submitted, and it belongs in no training mixture.

import os
from huggingface_hub import snapshot_download
from taco.ml import Dataset

root = snapshot_download("isp-uv-es/openearthmap-taco", repo_type="dataset",
                         allow_patterns=["openearthmap_test.zip", "openearthmap_test.zip/*", "openearthmap_test.zip/.tacocat/*"])
ds = Dataset(os.path.join(root, "openearthmap_test.zip"))

Getting started

git clone --recursive https://github.com/OscarPellicer/taco
pip install -e "taco/python[ml]"     # builds the reader: C++23, CMake, Ninja, pkg-config, libcurl >= 7.83, OpenSSL >= 3

Read it straight from the Hub:

import os
from huggingface_hub import hf_hub_download, snapshot_download
from taco.ml import Dataset, plot_sample

path = hf_hub_download("isp-uv-es/openearthmap-taco", "openearthmap.zip", repo_type="dataset")
ds = Dataset(path)
plot_sample(ds[0])

or from a local copy:

ds = Dataset("openearthmap.zip")
sample = ds[0]                      # {slot name: SlotValue}, arrays decoded
sample["image"].array.shape

Metadata without decoding anything:

import taco
taco.read("openearthmap.zip")              # one Arrow table, levels joined

Samples

What a sample contains

role	slot	holds	modality	detail
input	image	raster	optical	3 band(s), render
target	land_cover	mask	label_raster	9 classes

Licence

CC-BY-NC-SA-4.0

Terms of use:

• Non-commercial use only (CC-BY-NC-SA-4.0).
• Adapted material must be shared under CC-BY-NC-SA-4.0.
• Labels carry the licence of their source imagery: tiles from SpaceNet remain CC-BY-SA-4.0 and tiles from xBD remain CC-BY-NC-SA-4.0.

Required credits:

• OpenEarthMap: Xia et al.
• Per-region source attribution: https://open-earth-map.org/attribution.html

Providers: University of Tokyo / RIKEN (Xia et al.)

Acknowledgements

TACO was designed by César Aybar and is specified at https://asterisk.coop/taco/spec/.

Built by Oscar Pellicer within the ELLIOT project at the Image and Signal Processing Group (ISP), Universitat de València.