This project is a collaboration between mjs-py and KaipoCraft to create a generative adversarial neural network that generates novel satellite imagery. This GAN would allow the creation of fantasy maps and a view into the common elements (visible from satellites) of the urban landscape found across the world's megacities.
Sample image from training data
- Find a database to train the model off of
- Code the GAN logic in Python using Jupyter Notebook
- Train the GAN
- Get generated output
- Transfer model to Google Colab to leverage Google Earth Engine
- Import higher resolution NASA Landsat database from Google Earth Engine
- Transform the imagery - reducing cloudiness, pinpount bounding boxes around world's megacities
- Get the data to transform into tensors for use with tensorflow
- Train a higher resolution version of the model
- Get generated output
Distributed under the MIT License. See LICENSE.txt for more information.