satesr

A toolbox aimed at increasing the usage scenarios and utilization of free and open source satellite maps using super-resolution.

set up the environment:

python setup.py

Achieved functions (see demo.ipynb):

1. Download Bing Aerial maps with coordinates and a definable map radius.
2. Implementation of Real-ESRGAN model.
3. A SR model optimised for satellite imagery.

The aboved functions have been tested on Linux/Windows/MacOS. GPU acceleration is only avliable for CUDA enabled devices.

To do list:

1. Use generic models(e.g. YOlO, Unet) to test the SR performance(in progress)..
2. move the pytorch model to Tensorflow model for better GPU compability(in progress).
3. Further refining the model, the current model is trained using a single RTXA4000 with ~50 high-res photos with minimal train pipeline.
4. add other satellite image sources.
5. add other super-resolution models.

acknowledgment: sr model developed from https://github.com/xinntao/Real-ESRGAN aerial photos to train the model: https://arxiv.org/abs/1807.09532