BigEarthNet v2.0 (reBEN) Sentinel-1/Sentinel-2 Multi-Label Benchmark, LMDB Format

Provider
Leonard Hackel (TU Berlin RSiM / BIFOLD); BigEarthNet v2.0 by Clasen, Hackel, Demir et al.
License
CDLA-Permissive-1.0 (source)
Size
5 files
Format
Parquet (in sampled files)
Stored at
Hugging Face (https://huggingface.co/datasets/hackelle/BigEarthNetV2-LMDB)
Access
Free and public. No account with the data provider is needed.
DOI
10.5281/zenodo.10891137
Category
Earth Observation & Geospatial
Source
https://zenodo.org/records/10891137

About this dataset

BigEarthNet v2.0, also called reBEN (refined BigEarthNet), is the large-scale multi-label remote sensing benchmark of co-registered Sentinel-2 multispectral and Sentinel-1 SAR patches over Europe, labelled with CORINE land cover classes and used for multi-label scene classification, image retrieval and pretraining. This repository is a community pre-conversion, by one of the reBEN authors, into a single LMDB database in the format produced by rico-HDL: each entry is a SafeTensors object keyed by the Sentinel-2 patch id (bands B01 to B12 including B8A) or the Sentinel-1 patch name (VV, VH). It is labelled an unofficial mirror; the authoritative files are on Zenodo (record 10891137) and bigearth.net. The listing (measured 2026-11-14) shows a BENv2.lmdb folder and metadata.parquet, a copy of the official reBEN metadata with labels, train/validation/test splits and geography for patches without seasonal snow and cloud shadow; the repository uses about 155 GB. Load with configILM dataloaders, or with lmdb plus safetensors and pandas. BigEarthNet v2.0 is released under CDLA-Permissive-1.0 on Zenodo; cite the reBEN paper.

How to load this dataset in Python

from datasets import load_dataset
ds = load_dataset("hackelle/BigEarthNetV2-LMDB")
print(ds)

Data structure

File types in sampled files: Parquet

Files

Example file: metadata.parquet

Analyze this dataset without downloading it

On Scigantic, BigEarthNet v2.0 (reBEN) Sentinel-1/Sentinel-2 Multi-Label Benchmark, LMDB Format 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