isp-uv-es/sen4agrinet-taco

Sen4AgriNet

This is a repackaging, not a new dataset. It is Sen4AgriNet by Orion Lab, National Observatory of Athens (Sen4AgriNet), Agricultura, Ramaderia, Pesca i Alimentacio (Catalonia LPIS), French Paying Agency (France LPIS), ESA Copernicus / 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-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:

@article{sykas2022sen4agrinet,
  title   = {A Sentinel-2 Multiyear, Multicountry Benchmark Dataset for Crop Classification and Segmentation With Deep Learning},
  author  = {Sykas, Dimitrios and Sdraka, Maria and Zografakis, Dimitrios and Papoutsis, Ioannis},
  journal = {IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing},
  volume  = {15},
  pages   = {3323--3339},
  year    = {2022},
  doi     = {10.1109/JSTARS.2022.3164771}
}

About the data

10013 Sentinel-2 patches over Catalonia (ES) and France (FR) for 2019-2020, each a ragged time series of cloud-filtered L2A acquisitions with a dense crop-type label and a land-register parcel-id raster, annotated from the two countries' own LPIS farmer declarations and harmonised onto one FAO ICC taxonomy of 168 crop and land-cover classes.

10,013 samples · splits: test 2,252 · train 6,690 · validation 1,071 · tasks: semantic-segmentation, instance-segmentation

Packaged as TACO v3.

Full description

Bands. All thirteen are kept at their native resolution, as three leaves: 366x366 for the four 10 m bands, 183x183 for the six 20 m, 61x61 for the three 60 m (366 = 2x183 = 6x61, which is why the authors chose it). Each leaf band-stacks its series with frames on the outer axis, T in n_frames.

Target. The label raster is remapped to the authors' own contiguous 1..168 with 0 as no-label (ignore_index), using the LINEAR_ENCODER of their reference repository, so indices are comparable with the published numbers. The legend itself is their CROP_ENCODING, which the HF payload does not ship.

Radiometry. L2A BOA at 1e-4. The README says L1C, but every acquisition id in the release says MSIL2A and the pixels agree with L2A.

Splits. The release ships none, so the partition here is constructed: tile-disjoint across both years, assigned per country so each side holds both. 6690 train, 1071 val and 2252 test patches.

Licence. CC-BY-SA-4.0: the release's dataset card says MIT and its configuration says CC-BY-SA-4.0; the stricter is used.

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:

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

path = snapshot_download("isp-uv-es/sen4agrinet-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("sen4agrinet")
sample = ds[0]                      # {slot name: SlotValue}, arrays decoded
sample["s2_10m"].array.shape

Metadata without decoding anything:

import taco
taco.read("sen4agrinet/.tacocat")              # one Arrow table, levels joined

Samples

What a sample contains

role	slot	holds	modality	detail
input	s2_10m	raster_series	optical	4 band(s), unit 1, scaled
input	s2_20m	raster_series	multispectral	6 band(s), unit 1, scaled
input	s2_60m	raster_series	optical	3 band(s), unit 1, scaled
target	crop	mask	label_raster	169 classes
target	parcel	instance_id	label_raster

Licence

CC-BY-SA-4.0

Terms of use:

• Adapted material must be shared under CC-BY-SA-4.0.

Required credits:

• Sen4AgriNet: Sykas et al. (Orion Lab, National Observatory of Athens)
• Catalonia LPIS: Departament d'Agricultura, Ramaderia, Pesca i Alimentació, Generalitat de Catalunya
• France LPIS: Agence de Services et de Paiement
• Contains modified Copernicus Sentinel data

Providers: Orion Lab, National Observatory of Athens (Sen4AgriNet), Agricultura, Ramaderia, Pesca i Alimentacio (Catalonia LPIS), French Paying Agency (France LPIS), ESA Copernicus / 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.