Dreaming of map tiles, or how to make mapping sparsely populated areas more fun
MIT License
Dreaming of map tiles.
Machine-learning experiments on aerial images and OSM data.
Download Bing aerial image tiles and metadata about each tile with:
python scripts/get_mapswipe_data.py 1 # 1 -> sets mapswipe project
This will fetch tiles for a subset of those for which mapswipe has a positive rating. It will also fetch random tiles which do no appear in the mapswipe dataset. These are assumed to be "boring".
Next reorganise the data a bit and train a CNN:
python scripts/build_data.py
python scripts/train_cnn.py
You can also train an extra trees classifier instead of a CNN:
python scripts/train_extratrees.py
This does not need to have the data prepared via build_data.py
.
You need python and keras
.