isp-uv-es/worldfloods-v2-taco

WorldFloods v2

This is a repackaging, not a new dataset. It is WorldFloods v2 by Image and Signal Processing group, Universitat de Valencia (Portales-Julia et al.), Copernicus Emergency Management Service, EC JRC / Google Global Surface Water, 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-NC-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{portales2023worldfloods,
  title   = {Global flood extent segmentation in optical satellite images},
  author  = {Portal{\'e}s-Juli{\`a}, Enrique and Mateo-Garc{\'i}a, Gonzalo and Purcell, Cormac and G{\'o}mez-Chova, Luis},
  journal = {Scientific Reports},
  volume  = {13},
  number  = {1},
  pages   = {20316},
  year    = {2023},
  doi     = {10.1038/s41598-023-47595-7}
}

About the data

509 Sentinel-2 flood scenes at 10 m, each with a three-class land/water ground truth, a three-class clear/cloud ground truth and the JRC permanent-surface-water layer.

509 samples · splits: test 18 · train 475 · validation 16 · tasks: semantic-segmentation

Packaged as TACO v3.

Full description

Splits. The official split: 475 train, 16 val, 18 test.

Targets. The release's 2-band ground-truth raster is split into two single-band targets, since the two bands carry unrelated legends. The legends come from the ml4floods reference implementation, and the band order was verified against the JRC water layer and against SWIR reflectance (band 1 is clear/cloud, band 2 is land/water). In both, 0 means invalid and is declared as the ignore index. The JRC water layer is an input rather than a target: the question is which water is new.

Extent. The original GeoTIFFs at native scene extent: whole Copernicus EMS activation scenes, up to roughly 10,000x7,600 px, not tiles.

Bands. Only Sentinel-2's documented 13 bands are carried. The files hold two further undocumented channels, one identically zero, which are dropped.

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:

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

path = hf_hub_download("isp-uv-es/worldfloods-v2-taco", "worldfloods-v2.zip", repo_type="dataset")
ds = Dataset(path)
plot_sample(ds[0])

or from a local copy:

ds = Dataset("worldfloods-v2.zip")
sample = ds[0]                      # {slot name: SlotValue}, arrays decoded
sample["s2"].array.shape

Metadata without decoding anything:

import taco
taco.read("worldfloods-v2.zip")              # 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
input	permanent_water	mask	label_raster	4 classes
target	water	mask	label_raster	3 classes
target	cloud	mask	label_raster	3 classes

Licence

CC-BY-NC-4.0

Terms of use:

• Non-commercial use only (CC-BY-NC-4.0).

Required credits:

• WorldFloods: Portales-Julià et al. (ISP, Universitat de València)
• Contains modified Copernicus Sentinel data
• Flood maps: Copernicus Emergency Management Service
• Permanent water: EC JRC Global Surface Water

Providers: Image and Signal Processing group, Universitat de Valencia (Portales-Julia et al.), Copernicus Emergency Management Service, EC JRC / Google Global Surface Water

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.