This project is a PyTorch implementation of https://arxiv.org/abs/2007.03107, Additionally, a set of pre-processing tools is created to generate a multi-image Low-Resolution data from High-Resolution images.
pip install -r requirements.txt- First, you need datasets! You can Get it from NASA APPEARS or use this client . Remember to keep a split for validation.
- Dataset must be tiffs files.
- Once you have the datasets, complete the path in Train
- Check the configuration file to define the parameters of the degradation model
- Run!
Objective: The objective is to create
$N$ Low-Resolution images from one High-Resolution image. Then, the LR will be the input of the super-resolution neural network that will try to regenerate the HR image.
The dataset then:
- Takes an HR input
- Makes N pixel shifts
- Degrade the image using aDegradation Model
- reduce the resolution by a factor of X ( The pixel shifts are now subpixel shifts)
- Yield the
$N$ number of images.
