SSL4EO-S12 (v1.0): Large-Scale Multimodal Multitemporal Sentinel-1/2 Pretraining Dataset
- Provider
- Wang et al. (Technical University of Munich, DLR)
- License
- CC-BY-4.0 (source)
- Size
- 37 files
- Stored at
- Hugging Face (https://huggingface.co/datasets/wangyi111/SSL4EO-S12)
- Access
- Free and public. No account with the data provider is needed.
- Category
- Earth Observation & Geospatial
- Source
- https://arxiv.org/abs/2211.07044
About this dataset
SSL4EO-S12 is the original release of the self-supervised pretraining dataset for Earth observation by Wang et al. (IEEE Geoscience and Remote Sensing Magazine, 2023). It samples about 250,000 locations around the 10,000 most populous cities worldwide and captures each in four seasons, with three Copernicus products per location and season: Sentinel-1 GRD (VV, VH), Sentinel-2 Level-1C top-of-atmosphere (13 bands) and Sentinel-2 Level-2A surface reflectance (12 bands), as 264x264 pixel patches at 10 m, about 3 million patches in total exported from Google Earth Engine. Folders are s1_grd/, s2_l1c/ and s2_l2a/, plus ssl4eo-s12-ml/ with an ML-ready packaging. It is the corpus behind many MoCo, DINO, MAE and data2vec remote sensing backbones and differs from the later SSL4EO-S12 v1.1 Zarr release by Embed2Scale, which re-exported and aligned the data. About 1.6 TB. The authors release the dataset under CC BY 4.0 (code Apache 2.0); cite Wang et al. 2023.
How to load this dataset in Python
from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="wangyi111/SSL4EO-S12", filename="s1_grd/s1_grd.tar.gz.partaa", repo_type="dataset")
print(path)Data structure
Files
Example file: s1_grd/s1_grd.tar.gz.partaa
Analyze this dataset without downloading it
On Scigantic, SSL4EO-S12 (v1.0): Large-Scale Multimodal Multitemporal Sentinel-1/2 Pretraining Dataset 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
- Land Cover Datasets
- Trazo Field Boundary Annotations (fiboa)
- LandCoverNet Australia: Annual Land Cover Labels on Sentinel-1, Sentinel-2 and Landsat-8 Time Series (2018)
- Digital Earth Africa GeoMAD
- Digital Earth Africa Coastlines: S3 inventory files
- Open-Meteo Weather API Database
- Digital Earth Africa Sentinel-2 Level-2A Surface Reflectance Collection 1
- KyFromAbove on AWS