Conditional Stochastic Normalizing Flows for Blind Super-Resolution of Remote Sensing Images

by Hanlin Wu, Ning Ni, Shan Wang, and Libao Zhang, details are in paper.

Usage

Clone the repository:

git clone https://github.com/hanlinwu/BlindSRSNF.git

Requirements:

• pytorch==1.13.0
• pytorch-lightning==1.5.5
• numpy
• opencv-python
• easydict
• tqdm

Test with our pretrained models

1. Download the checkpoints from this url.
2. Unzip the downloaded file, and put the files on path: logs/

Train:

1. Download the training datsets from this url.

2. Unzip the downloaded dataset, and put the files on path: load/

3. Do training:

   For ansio degradation:

   python train.py --config configs/blindsrsnf_aniso.yaml
   

   For iso degradation:

   python train.py --config configs/blindsrsnf_iso.yaml
   

   For ansio degradation with the WorldStrat dataset:

   python train.py --config configs/blindsrsnf_iso.yaml
   

Test:

python test_diff.py --checkpoint logs/your_checkpoint_path

or

sh scripts/test_diff_aniso.sh logs/your_checkpoint_path