isp-uv-es/cloudsen12plus-l1c-high-taco

CloudSEN12+ (L1C, expert-labelled subset)

This is a repackaging, not a new dataset. It is CloudSEN12+ (L1C, expert-labelled subset) by Image & Signal Processing group, Universitat de Valencia (CloudSEN12+), 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: CC0-1.0

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{aybar2024cloudsen12plus,
  title   = {CloudSEN12+: The largest dataset of expert-labeled pixels for cloud and cloud shadow detection in Sentinel-2},
  author  = {Aybar, Cesar and Bautista, Lesly and Montero, David and Contreras, Julio and Ayala, Daryl and Prudencio, Fernando and Loja, Jhomira and Ysuhuaylas, Luis and Herrera, Fernando and Gonzales, Karen and others},
  journal = {Data in Brief},
  year    = {2024},
  doi     = {10.1016/j.dib.2024.110852}
}

About the data

Cloud and cloud-shadow segmentation on Sentinel-2 L1C top-of-atmosphere imagery: thirteen bands at 10 m with a dense four-class expert label (clear, thick cloud, thin cloud, cloud shadow).

10,847 samples · splits: test 1,060 · train 9,175 · validation 612 · tasks: semantic-segmentation

Packaged as TACO v3.

Full description

Scope. The 10,847 label_type == high samples of CloudSEN12+'s 50,247, at the release's own two native sizes: 10,000 at 512x512 and 847 at 2048x2048, over disjoint ROIs.

• The 10,000 scribble samples are not included. They use seven class values (0-6, plus a 99 sentinel over the 82% of pixels a scribble never touches), a legend the release does not document and which cannot be folded into the four-class one without renaming three classes.
• The 29,400 nolabel samples are not included. They carry only {0, 99}, so a 0 there is not the class Clear.
• Built from the L1C part of the release; its L2A and extra parts are not included.

Target. The native legend is kept verbatim and additionally reduced to ml:binary contaminated/clear.

Radiometry. DN = 10000 x TOA reflectance on every record. The 35 records from processing baseline 04.00 (s2_id _N0400_, dated 2022-01-25 or later) ship with the baseline's +1000 DN offset (RADIO_ADD_OFFSET = -1000); it is subtracted here, leaving 0 and 65535 alone and clipping at 1 (9,353 of 1.72e9 valid pixels were at or below 1000 DN). radio_add_offset gives each record's product offset, so the release's values are the stored ones minus that offset.

Splits. The official split, with the release's own string also kept in release_split. 0 of 2,847 ROI ids cross a split.

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/cloudsen12plus-l1c-high-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("cloudsen12plus-l1c-high")
sample = ds[0]                      # {slot name: SlotValue}, arrays decoded
sample["s2"].array.shape

Metadata without decoding anything:

import taco
taco.read("cloudsen12plus-l1c-high/.tacocat")              # one Arrow table, levels joined

Samples

What a sample contains

role	slot	holds	modality	detail
input	s2	raster	multispectral	13 band(s), unit 1, scaled
target	cloud	mask	label_raster	4 classes

Licence

CC0-1.0

Providers: Image & Signal Processing group, Universitat de Valencia (CloudSEN12+), 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.