Awesome

Single-image-Super-Resolution-of-Remote-Sensing-Images-with-Real-World-Degradation-Modeling

Paper: https://www.mdpi.com/2072-4292/14/12/2895

The code is based on natural image SR code by Xiaozhong Ji et al.

Requirements

• numpy
• scipy
• pytorch
• torchvision
• lpips
• argparse
• yaml
• opencv-python

Data preparation

• Prepare the AID dataset or other remote sensing image dataset.
• Use 'train.py' in './preprocess/KernelGAN/' to collect the kernel dataset. You may need to modify the path of input and output.
• Use 'collect_noise.py' in './preprocess/' to collect the noise patch dataset. You may need to modify the path of input and output in 'paths.yaml'.
• Generate the ideal or real-world training datasets with 'create_bicubic_dataset.py' or 'create_kernel_dataset.py' in './preprocess/'.

Training

• Train models with ideal or real-world datasets with 'train.py' in the root path. You may need to modify the path in './options/aid/train_bicubic.yml' or './options/aid/train_kernel_noise.yml'.

Test

• Train models with ideal or real-world datasets with 'test.py' in the root path. You may need to modify the path in './options/aid/test_aid.yml'.