isp-uv-es/meter-ml-taco

METER-ML

This is a repackaging, not a new dataset. It is METER-ML by Stanford ML Group / METER initiative (Zhu et al.), USDA NAIP, Copernicus / ESA (Sentinel-1, Sentinel-2), 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-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{zhu2022meterml,
  title     = {METER-ML: A Multi-Sensor Earth Observation Benchmark for Automated Methane Source Mapping},
  author    = {Zhu, Bryan and Lui, Nicholas and Irvin, Jeremy and Le, Jimmy and Tadwalkar, Sahil and Wang, Chenghao and Ouyang, Zutao and Liu, Frankie Y. and Ng, Andrew Y. and Jackson, Robert B.},
  booktitle = {Proceedings of the 2nd Workshop on Complex Data Challenges in Earth Observation (CDCEO)},
  year      = {2022}
}

About the data

Methane-emitting infrastructure identification over the United States: 86598 locations, each imaged on one 720 m x 720 m footprint by three products at once.

86,598 samples · splits: test 1,018 · train 85,065 · validation 515 · tasks: scene-classification

Packaged as TACO v3.

Full description

Imagery.

• NAIP aerial at 1 m, R/G/B/NIR, 720x720, with NIR carried in the PNG's alpha slot.
• The complete 13-band Sentinel-2 L1C cube on its three native grids: 10 m 72x72, 20 m 36x36, 60 m 12x12, reflectance x 1e4.
• A dual-polarisation Sentinel-1 sigma-0 patch, VV then VH, 10 m 72x72, 12-bit requantised with no published scale.

Labels. Multi-label over six facility types: CAFOs, landfills, coal mines, gas processing plants, oil refineries and petroleum terminals, wastewater treatment. The release's Negative is the all-zero row, and 410 sites carry more than one type. One label in the test split, Roundabout, is outside the release's own vocabulary and is carried as unlisted:Roundabout rather than folded into the negative, and declared ignore_classes.

Splits. The release's own lists: 85065 train, 515 val, 1018 test. Val and test are expert-reviewed while train comes from the source infrastructure databases, so they differ in label quality.

Coverage. One site of the release's 86,599 ships only two of its five arrays and is skipped, leaving 86598 samples.

Geometry. The release's own GeoJSON footprint polygons are 720 x cos(latitude) metres rather than 720 m square, a generation defect, so the footprint here comes from the paper and the array shapes, which agree exactly; the polygons' measured size is carried per sample. No CRS or transform is published for the imagery and none is invented. lon/lat are the release's site centres.

Getting started

git clone https://github.com/OscarPellicer/taco
pip install -e "taco/python[ml]"     # builds the native reader: needs CMake and Ninja

Read it straight from the Hub:

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

path = snapshot_download("isp-uv-es/meter-ml-taco", repo_type="dataset")
ds = Dataset(path)                  # the repository is the container: parts + .tacocat
plot_sample(ds[0])

or from a local copy:

ds = Dataset("meter-ml")
sample = ds[0]                      # {slot name: SlotValue}, arrays decoded
sample["naip"].array.shape

Metadata without decoding anything:

import taco
taco.read("meter-ml/.tacocat")              # one Arrow table, levels joined

Samples

What a sample contains

role	slot	holds	modality	detail
input	naip	raster	optical	4 band(s), requantised
input	s2_10m	raster	optical	4 band(s), unit 1, scaled
input	s2_20m	raster	multispectral	6 band(s), unit 1, scaled
input	s2_60m	raster	optical	3 band(s), unit 1, scaled
input	s1	raster	sar	2 band(s), requantised
target	facility	class_multihot		7 classes

Licence

CC-BY-4.0

Providers: Stanford ML Group / METER initiative (Zhu et al.), USDA NAIP, Copernicus / ESA (Sentinel-1, Sentinel-2)

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.