PAPS

This is the PyTorch implementation of paper "PAPS: Progressive Attention-based Pan-sharpening".

Build Your Datasets

Before running the codes, you should modify the input path in codes according to your datasets. The related codes of this part are all in 'process_data' folder.

You can generate your datasets as follows:

cd process_data
python real_processing.py
python to_patch.py

Train the Model

You can just modify the args in args.py and run

python train.py

The logs and the model will be saved in the 'log' and 'checkpoint' folders respectively.

Test the Model

You can just modify the args in testsolver.py and run

python test.py

Generally the results will be generated below the 'output' folder.

Main Architecture

Dependencies

Ubuntu 16.04.6 LTS
Python 3.7.0

main python packages:

gdal = 2.3.2
pytorch = 1.9.1     
torchvision = 0.10.1
opencv-python = 4.5.3.56
cudatoolkit = 11.1.74