MADOS: Marine Debris and Oil Spill Sentinel-2 Segmentation Dataset (TACO)
- Provider
- Kikaki, Kakogeorgiou, Hoteit, Karantzalos (National Technical University of Athens); TACO repackaging by ISP, Universitat de Valencia
- License
- CC-BY-4.0 (source)
- Size
- 9 files
- Format
- PNG (in sampled files)
- Stored at
- Hugging Face (https://huggingface.co/datasets/isp-uv-es/mados-taco)
- Access
- Free and public. No account with the data provider is needed.
- DOI
- 10.1016/j.isprsjprs.2024.02.017
- Category
- Earth Observation & Geospatial
- Source
- https://huggingface.co/datasets/isp-uv-es/mados-taco
About MADOS: Marine Debris and Oil Spill Sentinel-2 Segmentation Dataset (TACO)
MADOS (Marine Debris and Oil Spill) is a Sentinel-2 benchmark for detecting marine pollutants and the sea-surface features they are confused with (ISPRS Journal of Photogrammetry and Remote Sensing, 2024). It contains 2,803 crops from 174 Sentinel-2 scenes, with sparse polygon annotations in 15 sea-surface classes covering marine debris (litter), oil spills and look-alikes such as sargassum and other floating algae, natural organic material, ships, foam, turbid and sediment-laden water, shallow water and waves; class 0 means unannotated. Imagery is eleven Rayleigh-corrected reflectance bands kept on their native 10 m, 20 m and 60 m grids, plus per-pixel annotation confidence and report masks. The official splits are 1,433 train, 642 validation and 728 test. This copy is a 2.5 GB TACO archive (mados.zip) readable with the tacoreader library or GDAL /vsizip/. The original is on Zenodo under CC BY 4.0; cite Kikaki et al. 2024.
How to load this dataset in Python
from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="isp-uv-es/mados-taco", filename="assets/logos.png", repo_type="dataset")
print(path)Data structure
File types in sampled files: PNG
Files
Example file: assets/logos.png
Analyze this dataset without downloading it
On Scigantic, MADOS: Marine Debris and Oil Spill Sentinel-2 Segmentation Dataset (TACO) opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.
This page is also available as Markdown and as JSON.
Related Earth Observation & Geospatial datasets
- Daylight Map Distribution of OpenStreetMap
- DEA Water Observations Statistics, Multi-Year Summary (Landsat, Australia, 1987 to 2025)
- Light Stable Representations (LSR): Drone Orthomosaic Tiles at Three Times of Day with Canopy Height, MPG Ranch, Montana
- Field Boundaries (Delineate Anything)
- ESA WorldCover 10 m 2020 Global Land Cover Map, Version 100 (v100)
- PoroTomo
- Africa Field Boundary Labels: Additional rasterized field labels and corresponding Planet images
- IDTReeS 2020: Tree Crown Delineation and Species Classification in NEON Airborne Imagery (TACO)