Learning Dynamic Scale Awareness and Global Implicit Functions for Continuous-Scale Super-Resolution of Remote Sensing Images

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

Usage

Clone the repository:

https://github.com/hanlinwu/SADN.git

Requirements:

• pytorch==1.10.0
• pytorch-lightning==1.5.5
• numpy
• opencv-python
• easydict
• tqdm

Pretrained checkpoints:

1. Model pretrained on the DIV2k dataset: download from this url

   Note: For natural image data test sets such as Set14, B100, etc., to obtain results consistent with those in the paper, please set the test_Y attribute in line 82 of test.py to True.

2. Model pretrained on the remote sensing dataset: download from this url

Train:

1. Download training datset from this url.
2. Change the train.datapath and valid.data_path in config/your_config_file.yaml
3. Do training:

   python train.py --config config/your_config_file.yaml
   

Test:

1. Download benchmark datasets Set14,B100,Urban100,Manga109, and put them on path: load/benchmark/datset_name

2. test without self-ensemble

   python test.py --checkpoint your_checkpoint_path --datasets Set14,B100,Urban100 --scales 2,3,4
   

3. test with self-ensemble

   python test.py --checkpoint your_checkpoint_path --datasets Set14,B100,Urban100 --scales 2,3,4 --self_ensemble