Dataset Overview

Aerial Imagery

• Size: 512 × 512 x 5
• Spatial Resolution: 0.2 m
• Channels: 5 (RGB, NIR, Elevation)

Sentinel‑2 Imagery

• Spatial Resolution: 10-20 m
• Spectral Bands: 10
• Snow/Cloud Mask: probability range 0-100
• Multi‑time Steps: format T × 10 × W × H (where T, W, H vary)

Labels (Mask)

• Size: 512 × 512
• Number of Classes: 13

Classes

Usage

Install Dataset4EO

git clone --branch streaming https://github.com/EarthNets/Dataset4EO.git

pip install -e .

Then download the dataset from the Huggingface repository.

import dataset4eo as eodata
import time

train_dataset = eodata.StreamingDataset(input_dir="optimized_flair2_test", num_channels=5, channels_to_select=[0,1,2], shuffle=True, drop_last=True)
sample = dataset[101]
print(sample.keys())
print(sample["image"])        
print(sample["simage"].shape)
print(sample["label"]) 

Citation

@article{garioud2023flair, title={FLAIR# 2: textural and temporal information for semantic segmentation from multi-source optical imagery}, author={Garioud, Anatol and De Wit, Apolline and Poup{e}e, Marc and Valette, Marion and Giordano, S{e}bastien and Wattrelos, Boris}, journal={arXiv preprint arXiv:2305.14467}, year={2023} }

Dataset License

This dataset is licensed under the "OPEN LICENCE 2.0/LICENCE OUVERTE", a license created by the French government to promote the dissemination of open data by public administration bodies.