---
task_categories:
- image-segmentation
---
# DFC2020 
**DFC2020** dataset is a large-scale remote sensing dataset designed for land cover classification and semantic segmentation tasks. It contains both Sentinel-1 images and Sentinel-2 images. The original DFC2020 dataset contains only a validation set of 986 samples and a test set of 5,128 samples. To better utilize this dataset, we treat the original test set (5,128 samples) as our training and validation sets, and the original validation set (986 samples) as our test set. In addition, since the image resolution is 256 x 256 pixels, we follow [CROMA](https://arxiv.org/abs/2311.00566)’s method, further dividing each image of 256 x 256 pixels into 9 smaller patches of 96 x 96 pixels with the overlap of 16 pixels. As a result, our final training set contains 41,537 training samples, the final validation set contains 4,615 samples and the final test set consists of 8,874 samples. All images are 96 x 96 pixels. Moreover, we follow [DFC2020](https://github.com/lukasliebel/dfc2020_baseline), mapping the segmentation labels to DFC2020 classes.

## How to Use This Dataset
```python
from datasets import load_dataset

dataset = load_dataset("GFM-Bench/DFC2020")
```

Also, please see our [GFM-Bench](https://github.com/uiuctml/GFM-Bench) repository for more information about how to use the dataset! 🤗

## Dataset Metadata

The following metadata provides details about the Sentinel-2 imagery used in the dataset:
- **Number of Sentinel-1 Bands**: 2
- **Sentinel-1 Bands**: VV, VH
- **Number of Sentinel-2 Bands**: 13
- **Sentinel-2 Bands**: B01 (**Coastal aerosol**), B02 (**Blue**), B03 (**Green**), B04 (**Red**), B05 (**Vegetation red edge**), B06 (**Vegetation red edge**), B07 (**Vegetation red edge**), B08 (**NIR**), B8A (**Narrow NIR**), B09 (**Water vapour**), B10 (**SWIR – Cirrus**), B11 (**SWIR**), B12 (**SWIR**)
- **Image Resolution**: 96 x 96 pixels
- **Spatial Resolution**: 10 meters 
- **Number of Classes**: 8

## Dataset Splits
The **DFC2020** dataset consists following splits:
- **train**: 41,537 samples
- **val**: 4,615 samples
- **test**: 8,874 samples

## Dataset Features:
The **DFC2020** dataset consists of following features:
- **radar**: the Sentinel-1 image.
- **optical**: the Sentinel-2 image.
- **label**: the segmentation labels.
- **radar_channel_wv**: the central wavelength of each Sentinel-1 bands.
- **optical_channel_wv**: the central wavelength of each Sentinel-2 bands.
- **spatial_resolution**: the spatial resolution of images.
## Citation
If you use the DFC2020 dataset in your work, please cite the original paper:
```
@ARTICLE{9369830,
  author={Robinson, Caleb and Malkin, Kolya and Jojic, Nebojsa and Chen, Huijun and Qin, Rongjun and Xiao, Changlin and Schmitt, Michael and Ghamisi, Pedram and Hänsch, Ronny and Yokoya, Naoto},
  journal={IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing}, 
  title={Global Land-Cover Mapping With Weak Supervision: Outcome of the 2020 IEEE GRSS Data Fusion Contest}, 
  year={2021},
  volume={14},
  number={},
  pages={3185-3199},
  keywords={Earth;Data integration;Remote sensing;Satellites;Training;Tensors;Synthetic aperture radar;Convolutional neural networks (CNNs);deep learning;image analysis and data fusion;land-cover mapping;multimodal;random forests (RFs);weak supervision},
  doi={10.1109/JSTARS.2021.3063849}}
```