isp-uv-es/seefar-landsat-taco

SeeFar V0 - Landsat

This is a repackaging, not a new dataset. It is SeeFar V0 - Landsat by USGS / NASA (Landsat 8/9), Coastal Carbon (SeeFar assembly; Lowman 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: USGS public domain (Landsat Data Distribution Policy)

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{lowman2024seefar,
  title   = {SeeFar: Satellite Agnostic Multi-Resolution Dataset for Geospatial Foundation Models},
  author  = {Lowman, James and Zheng, Kelly Liu and Fraser, Roydon and The, Jesse Van Griensven and Valipour, Mojtaba},
  journal = {arXiv preprint arXiv:2406.06776},
  year    = {2024}
}

About the data

The landsat half of SeeFar V0: 33282 four-band (blue, green, red, NIR) tiles from Landsat 8/9 OLI at 30 m. No labels -- this is a multi-resolution self-supervised pretraining corpus.

33,282 samples · splits: test 4,991 · train 23,307 · validation 4,984 · tasks: similarity-search

Packaged as TACO v3.

Full description

Splits. The release's own: 23307 train, 4991 test, 4984 val. The release spells the last one validate; the partition is untouched and split_original keeps its word.

Scope. SeeFar ships four sources in one archive and they are not co-registered: the four id spaces are independent, the four-way filename intersection is 337 of 23,527 in train and 1 in test, and for 120 of those 337 the closest same-name tile in another source is 546 km away, with 0 of 360 pairs within 10 km. Each source is therefore its own collection and nothing may be joined across them; shared_stem records the filename coincidence.

Geometry. Tiles are in per-tile UTM zones, so resolution_m is measured geodesically per sample. The band table is the release's own per-tile wavelength tags, asserted on every file.

Radiometry. Collection-2 Level-2 surface reflectance, 0.0000275 * DN - 0.2, exactly. Three measurements establish that rather than assume it:

• the corpus maximum over a 150-tile sample is 65,454, which is the C2 L2 valid ceiling;
• the release's own wavelength tags reproduce OLI's B2-B5 to the nanometre;
• reduced, the 40 most-vegetated tiles average 0.024 blue, 0.047 green, 0.035 red and 0.316 NIR, which is what a green canopy does.

Exclusions. 329 members of this source are truncated in the release -- 33 distinct shapes, none 384x384, and not one holding a single non-zero pixel over a full scan of all 329 -- and are excluded by name rather than repaired. The counts above are after that exclusion.

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/seefar-landsat-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("seefar-landsat")
sample = ds[0]                      # {slot name: SlotValue}, arrays decoded
sample["image"].array.shape

Metadata without decoding anything:

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

Samples

What a sample contains

role	slot	holds	modality	detail
input	image	raster	optical	4 band(s), unit 1, scaled

Licence

USGS public domain (Landsat Data Distribution Policy)

Providers: USGS / NASA (Landsat 8/9), Coastal Carbon (SeeFar assembly; Lowman 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.