isp-uv-es/marida-taco

MARIDA - Marine Debris Archive

This is a repackaging, not a new dataset. It is MARIDA - Marine Debris Archive by National Technical University of Athens (Kikaki, Kakogeorgiou, Karantzalos et al.), 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-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{kikaki2022marida,
  title   = {MARIDA: A benchmark for Marine Debris detection from Sentinel-2 remote sensing data},
  author  = {Kikaki, Katerina and Kakogeorgiou, Ioannis and Mikeli, Paraskevi and Raitsos, Dionysios E. and Karantzalos, Konstantinos},
  journal = {PLOS ONE},
  volume  = {17},
  number  = {1},
  pages   = {e0262247},
  year    = {2022}
}

About the data

15381 Sentinel-2 tiles of 96x96 pixels annotated with 15 sea-surface classes covering marine litter and the natural features it is confused with. Eleven L2A surface-reflectance bands resampled to one 10 m grid.

15,381 samples · splits: test 6,183 · train 6,246 · validation 2,952 · tasks: semantic-segmentation

Packaged as TACO v3.

Full description

Labels. Sparse polygons. Class 0 means unannotated and is the ignore index, and labelled_frac records the annotated share per sample.

Chipping and splits. This is the GFM-Bench re-chip, which cuts each of the release's 1,381 256x256 patches into nine tiles and re-splits them 6246/2952/6183. The row count is tiles, not release patches.

• That split is not scene-disjoint: 12 of the 63 Sentinel-2 scenes have tiles in more than one split, covering 5,904 of the 15381 tiles (38.4%). Train, val and test tiles routinely come from the same granule, date and atmosphere.
• The benchmark's own split is the one carried. scene_id is on every row, so a scene-disjoint resplit is one groupby away.

Georeferencing. The re-chip carries no geotransform, so the rasters are stored unreferenced rather than with an invented one.

Related data. MADOS is the same team's later reprocessing of the same waters: Rayleigh-corrected reflectance at three native grids, under a different legend.

Getting started

git clone https://github.com/OscarPellicer/taco
pip install -e "taco/python[ml]"     # builds the native reader: needs CMake and Ninja

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/marida-taco", "marida.zip", repo_type="dataset")
ds = Dataset(path)
plot_sample(ds[0])

or from a local copy:

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

Metadata without decoding anything:

import taco
taco.read("marida.zip")              # one Arrow table, levels joined

Samples

What a sample contains

role	slot	holds	modality	detail
input	optical	raster	multispectral	11 band(s), unit 1, physical
target	label	mask	label_raster	16 classes

Licence

CC-BY-4.0

Providers: National Technical University of Athens (Kikaki, Kakogeorgiou, Karantzalos et al.), 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.