README.md

metadata

license: cc-by-nc-sa-4.0

OSCD100

OSCD100 is a subset of the Onera Satellite Change Detection (OSCD) dataset, containing 100 pre-cropped image pairs at 256x256 resolution. It is designed for tutorials and demonstrations, not benchmarking.

Dataset Structure

The dataset maintains the same file structure and all 13 Sentinel-2 bands as the full OSCD dataset:

• Bands: B01, B02, B03, B04, B05, B06, B07, B08, B8A, B09, B10, B11, B12
• Splits: train (60 samples), val (20 samples), test (20 samples)
• Format: GeoTIFF for imagery, PNG for labels
• Resolution: 256x256 pixels per crop
• Temporal: Each sample contains two images (pre and post change)

Files

• Onera Satellite Change Detection dataset - Images.zip (220MB): All image pairs with 13 Sentinel-2 bands
• Onera Satellite Change Detection dataset - Train Labels.zip: Training labels
• Onera Satellite Change Detection dataset - Val Labels.zip: Validation labels
• Onera Satellite Change Detection dataset - Test Labels.zip: Test labels

Usage

from torchgeo.datasets import OSCD100

# Load dataset
dataset = OSCD100(root='data', split='train', download=True)

# Get a sample
sample = dataset[0]
print(sample['image'].shape)  # torch.Size([2, 13, 256, 256])
print(sample['mask'].shape)   # torch.Size([1, 256, 256])

Citation

If you use this dataset, please cite the original OSCD paper:

@inproceedings{daudt2018urban,
  title={Urban change detection for multispectral earth observation using convolutional neural networks},
  author={Daudt, Rodrigo Caye and Le Saux, Bertr and Boulch, Alexandre},
  booktitle={IGARSS 2018-2018 IEEE International Geoscience and Remote Sensing Symposium},
  pages={2115--2118},
  year={2018},
  organization={IEEE}
}

License

Same as the original OSCD dataset.