RS-ESRGAN: Super-Resolution of Sentinel-2 Imagery Using Generative Adversarial Networks
May 16, 2025 ยท View on GitHub
See our project website here.
We used BasicSR codes, which can be found here, and modified it to suit our purposes to work with remote-sensing data.
License
This code cannot be used for commercial purposes. Please contact the authors if interested in licensing this software.
Data
To train with WorldView-2 European Cities dataset, you must download the data and follow the directions in options/train folder.
WorldView-Sentinel pairs unfortunately cannot be shared due to license acquisition issues.
Training
- Train the model for Super-resolution with European-cities Dataset
python train_esrgan_EUCities.py -opt ./option/train/train_ESRGAN_ESRGAN_WV_5x_v1.json. Checkpoints and logs will be saved as defined in the json file.
Contact
For questions and suggestions use the issues section or send an e-mail to luis.fernando.salgueiro@upc.edu