We have forked from BasicSR

For the original README document from BasicSR, refer to README.md.txt.

BasicSR

TBD

Our changes

We have updated model tunings for high-resolution satellite data. See Datasets.

These new changes are authored by Ethan D. Kyzivat and Ekaterina Lezine.

Image import and preprocessing workflow

0. Download images to Scenes folder
1. plot_hists_serial.py > saves histogram for each scene
2. Compute_mean_hist.py > averages these histograms
3. extract_subimgs_single.py > Divides satellite scenes into subsets
4. rand_shuf.sh > Randomly creates training and validation partitions
5. generate_mod_LR_bic_parallel.py > Upscales and downscales subsets via a number of methods

Dependencies and Installation

• Python 3 (Recommend to use Anaconda)
• PyTorch >= 1.0
• NVIDIA GPU + CUDA
• Python packages: pip install numpy opencv-python lmdb pyyaml
• TensorBoard:
  • PyTorch >= 1.1: pip install tb-nightly future
  • PyTorch == 1.0: pip install tensorboardX

Datasets

TBD

Get Started

Please see wiki for the basic usage, i.e., training and testing.