SPADA: Sentinel-2 Land Cover Segmentation with Sparse CORINE Scribbles and LUCAS Point Labels

Provider
LINKS Foundation
License
CC-BY-4.0 (source)
Size
158 files
Format
PNG (in sampled files)
Stored at
Hugging Face (https://huggingface.co/datasets/links-ads/spada-dataset)
Access
Free and public. No account with the data provider is needed.
Category
Earth Observation & Geospatial
Source
https://arxiv.org/abs/2306.16252

About this dataset

SPADA is the dataset from Land Cover Segmentation with Sparse Annotations from Sentinel-2 Imagery (IGARSS 2023), built to train land cover and fuel-type segmentation models from weak, sparse labels over European areas of interest. Inputs are Sentinel-2 L2A images from 2018 with 12 bands in reflectance x 10,000 (GeoTIFF), obtained from the Microsoft Planetary Computer. Two kinds of label are provided: scribble annotations derived from CORINE Land Cover 2018 polygons after a scribbling process that keeps only sparse strokes inside each class region, and point-wise labels from the 2018 LUCAS in-situ survey rasterised in 5x5 pixel blocks. The training split also provides overlapping 512x512 tiles, while validation and test are cut into non-overlapping 2048x2048 sections. The data are stored as segmented tar.gz parts under data/ that must be concatenated and extracted (about 165 GB); training code is in github.com/links-ads/igarss-spada. Released by LINKS Foundation under CC BY 4.0; cite the IGARSS 2023 SPADA paper.

How to load this dataset in Python

from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="links-ads/spada-dataset", filename="assets/sample.png", repo_type="dataset")
print(path)

Data structure

File types in sampled files: PNG

Files

Example file: assets/sample.png

Analyze this dataset without downloading it

On Scigantic, SPADA: Sentinel-2 Land Cover Segmentation with Sparse CORINE Scribbles and LUCAS Point Labels 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