CloudSEN12+ v1.1: Expert-Labelled Cloud and Cloud Shadow Detection for Sentinel-2
- Provider
- Cesar Aybar et al., Image and Signal Processing group, University of Valencia (TACO Foundation)
- License
- CC0-1.0 (source)
- Size
- 16 files
- Stored at
- Hugging Face (https://huggingface.co/datasets/tacofoundation/cloudsen12)
- Access
- Free and public. No account with the data provider is needed.
- Category
- Earth Observation & Geospatial
- Source
- https://www.sciencedirect.com/science/article/pii/S2352340924008163
About this dataset
CloudSEN12+ is the largest dataset of expert-labelled pixels for cloud and cloud shadow semantic segmentation in Sentinel-2 imagery, extending CloudSEN12 with twice as many reviewed labels and refined earlier annotations. Labels classify clear, thick cloud, thin cloud and cloud shadow, with high-quality, scribble and unlabelled patches at 509 x 509 and 2000 x 2000 pixels (padded to 512 and 2048). Version 1.1 provides three modes: cloudsen12-l1c (Level-1C), cloudsen12-l2a (Level-2A from Google Earth Engine) and cloudsen12-extra with Copernicus DEM elevation, ESA WorldCover, Sentinel-1 VV/VH backscatter and cloud masks from QA60, Sen2Cor, s2cloudless, Cloud Score+, UnetMobV2 and SEnSeIv2 for comparison. The listing (measured 2026-11-14) shows 15 .part.taco files of up to 20 GB each, about 248 GB in total, in the TACO format; load them with tacoreader (tacoreader.load("tacofoundation:cloudsen12-l1c")) and read rasters with rasterio. GeoTIFFs are also on ScienceDataBank. The publishers release it under CC0, placing it in the public domain; cite Aybar et al. 2024 (Data in Brief).
How to load this dataset in Python
from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="tacofoundation/cloudsen12", filename="cloudsen12-extra.0000.part.taco", repo_type="dataset")
print(path)Data structure
Files
Example file: cloudsen12-extra.0000.part.taco
Analyze this dataset without downloading it
On Scigantic, CloudSEN12+ v1.1: Expert-Labelled Cloud and Cloud Shadow Detection for Sentinel-2 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
- NREL NSRDB v3 typical direct years (TDY) in HDF5 format
- Data for Canada - Orthoimagery Data Packages
- Public Utility Data Liberation Project
- PMTiles terrain tiles from dem1a by Geospatial Information Authority of Japan
- High Resolution Canopy Height Maps by WRI and Meta
- Argentina National Census Data
- Finnish Meteorological Institute Weather Radar Data
- Digital Earth Pacific WOfS All-Time Summary: Long-Term Water Frequency 1984 to 2023 (Landsat, 30 m)