isp-uv-es/magicbathynet-s2-pixel-class-taco

MagicBathyNet s2 (pixel class)

This is a repackaging, not a new dataset. It is MagicBathyNet s2 (pixel class) by TU Berlin RSiM / BIFOLD (Agrafiotis et al.), Dep. of Land and Surveys of Cyprus (LiDAR reference), 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:

@inproceedings{agrafiotis2024magicbathynet,
  title     = {{MagicBathyNet}: A Multimodal Remote Sensing Dataset for Bathymetry Prediction and Pixel-based Classification in Shallow Waters},
  author    = {Agrafiotis, Panagiotis and Janowski, {\L}ukasz and Skarlatos, Dimitrios and Demir, Beg{\"u}m},
  booktitle = {IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium},
  pages     = {249--253},
  year      = {2024},
  doi       = {10.1109/IGARSS53475.2024.10641355}
}

About the data

533 patches of 18x18 Sentinel-2 patches at a measured 9.97-9.98 m, each covering 180x180 m, over Agia Napa, Cyprus (35) and Puck Lagoon, Poland (498).

533 samples · splits: test 107 · train 364 · validation 62 · tasks: semantic-segmentation

Packaged as TACO v3.

Full description

Splits. The release's own list for this sensor and task: 426 train, 107 test, disjoint. validation (62) is carved out of the release's train, and split_original keeps the release's own value.

Target. A seabed class mask over five classes: poseidonia, rock, macroalgae, sand, ignored.

• The legend is not shipped with the data. The masks are unlabelled 3-band RGB; the colour-to-class mapping comes from the authors' notebooks, and a scan of every annotated patch reproduces exactly those colours and no others.
• The two sites index the same colour differently: sand is 3 at Agia Napa and 0 at Puck Lagoon. This container uses Agia Napa's ordering for both, and release_class_index carries the published per-site index. Puck Lagoon uses neither rock nor macroalgae.
• The annotation is sparse survey points, not a dense mask: 697 labelled pixels of 18,144,000 across Agia Napa's 35 aerial patches (0.0038%), 753 of 258,163,200 across Puck Lagoon's 498. The same ~700 points recur at every resolution, and labelled_frac records the share per patch.

Resolution. resolution_m is measured geodesically per patch and reproduces the published 10 m.

Radiometry. Sentinel-2 is scaled by 1e-4 to BOA reflectance, the published Sen2Cor quantification value.

Scope. One container per (sensor, task): the release publishes a split list per (sensor, task) and the annotated sets differ. Intersecting the three sensors' bathymetry sets keeps 2,002 patches of a 2,892-patch union, so a single multimodal container would drop 30.8% of the benchmark. site + patch_id joins the MagicBathyNet containers.

Licence. CC-BY-NC-4.0, per the Zenodo record and the archive README: redistributable with attribution, non-commercially.

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/magicbathynet-s2-pixel-class-taco", "magicbathynet-s2-pixel-class.zip", repo_type="dataset")
ds = Dataset(path)
plot_sample(ds[0])

or from a local copy:

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

Metadata without decoding anything:

import taco
taco.read("magicbathynet-s2-pixel-class.zip")              # one Arrow table, levels joined

Samples

What a sample contains

role	slot	holds	modality	detail
input	image	raster	optical	3 band(s), unit 1, scaled
target	seabed	mask	label_raster	5 classes

Licence

CC-BY-NC-4.0

Terms of use:

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

Required credits:

• MagicBathyNet: Agrafiotis, Janowski, Skarlatos and Demir
• Contains modified Copernicus Sentinel data

Providers: TU Berlin RSiM / BIFOLD (Agrafiotis et al.), Dep. of Land and Surveys of Cyprus (LiDAR reference)

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.