isp-uv-es/ssl4eo-s12-downstream-taco

SSL4EO-S12-downstream

This is a repackaging, not a new dataset. It is SSL4EO-S12-downstream by Embed2Scale consortium, ESA Copernicus, 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:

@article{vinge2025neucobench,
  title   = {NeuCo-Bench: A Novel Benchmark Framework for Neural Embeddings in Earth Observation},
  author  = {Vinge, Rikard and Wittmann, Isabelle and Schneider, Jannik and Marszalek, Michael and Gilch, Luis and Brunschwiler, Thomas and Albrecht, Conrad M.},
  journal = {arXiv preprint arXiv:2510.17914},
  year    = {2025}
}
@article{blumenstiel2025ssl4eos12,
  title   = {SSL4EO-S12 v1.1: A Multimodal, Multiseasonal Dataset for Pretraining, Updated},
  author  = {Blumenstiel, Benedikt and Ait Ali Braham, Nassim and Albrecht, Conrad M. and Maurogiovanni, Stefano and Fraccaro, Paolo},
  journal = {arXiv preprint arXiv:2503.00168},
  year    = {2025}
}

About the data

13260 co-located datacubes, each holding four acquisitions of one place in three modalities at once: Sentinel-1 GRD (VV, VH, dB), Sentinel-2 L1C (13 bands, TOA) and Sentinel-2 L2A (12 bands, BOA), all 264x264. L1C carries B10 and L2A does not, so they are two slots, never one cube.

13,260 samples · splits: test 8,111 · train 4,634 · validation 515 · tasks: regression, similarity-search

Packaged as TACO v3.

Full description

Targets. Eleven downstream task columns ride as optional values over sparse subsets (biomass_mean 2415, biomass_std 2415, clouds_reg 1140, crops 3355, heatisland_mean 1659, heatisland_std 1659, landcover_agriculture 4691, landcover_forest 4691, nodata 13260, random_cls 1140, random_reg 1140), each with a has_ flag, so the pixels are stored once rather than repeated per task. Eight are declared targets; nodata (the image's own zero fraction) and the random control tasks random_cls and random_reg are columns only.

Splits. The release's own CVPR EarthVision phase, globally consistent across all nine tasks that publish it with zero conflicting ids: 5149 dev (written train) and 8111 eval (written test). A val set is carved out of train, and the release's own word is kept in split_original.

Time and georeferencing. The release publishes no acquisition dates and no georeferencing: every cube's time coordinate is the index [0,1,2,3] and its x/y are pixel indices 0..263. frame_index is carried, and no date, CRS or transform is invented.

Radiometry. S1 is dB (physical); both S2 products are reflectance x 10000 (scaled, 1e-4).

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/ssl4eo-s12-downstream-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("ssl4eo-s12-downstream")
sample = ds[0]                      # {slot name: SlotValue}, arrays decoded
sample["s1"].array.shape

Metadata without decoding anything:

import taco
taco.read("ssl4eo-s12-downstream/.tacocat")              # one Arrow table, levels joined

Samples

What a sample contains

role	slot	holds	modality	detail
input	s1	raster_series	sar	2 band(s), unit dB, physical
input	s2l1c	raster_series	multispectral	13 band(s), unit 1, scaled
input	s2l2a	raster_series	multispectral	12 band(s), unit 1, scaled
target	biomass_mean	scalar		unit Mg ha-1
target	biomass_std	scalar		unit Mg ha-1
target	clouds_reg	scalar		unit 1
target	crops	scalar		unit 1
target	heatisland_mean	scalar		unit K
target	heatisland_std	scalar		unit delta_degC
target	landcover_agriculture	scalar		unit %
target	landcover_forest	scalar		unit %

Licence

CC-BY-4.0

Providers: Embed2Scale consortium, ESA Copernicus

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.