Dataset Card for SAR2Opt

A collection of 600x600 SAR/EO image pairs at a spatial resolution of 1m.

Dataset Details

Dataset Description

From the paper: We used TerraSAR-X to collect SAR images at the spatial resolution of 1 m of ten cities in Asia, North America, Oceania, and Europe from 2007 to 2013. The corresponding optical images collected from Google Earth Engine are coregistered to SAR images by manual selection of control points. We extracted image patches of size 600×600 pixels from coregistered SAR-to-optical image pairs to form the SAR2Opt dataset.

• Curated by: Yitao Zhao, Turgay Celik, Nanqing Liu, Heng-Chao Li
• License: MIT

Dataset Sources [optional]

• Repository/Demo: https://github.com/MarsZhaoYT/SAR2Opt-Heterogeneous-Dataset/tree/main
• Paper: https://ieeexplore.ieee.org/document/9779739

Dataset Structure

The dataset is stored in two different formats, raw and combined. In raw format, fold A contains the SAR images, and fold B contains the EO images. In combined format, the SAR and EO images are concatenated (as required by pix2pix). In both formats, the train/test splits are unchanged from the authors' original splits.

Citation [optional]

BibTeX:

@article{zhao2022comparative, title={A Comparative Analysis of GAN-based Methods for SAR-to-Optical Image Translation}, author={Zhao, Yitao and Celik, Turgay and Liu, Nanqing and Li, Heng-Chao}, journal={IEEE Geoscience and Remote Sensing Letters}, year={2022}, publisher={IEEE} }

Dataset Card Authors [optional]

Zach Button, University of Missouri - Kansas City

Dataset Card Contact

[email protected]