Awesome

Instance Segmentation of Geometrical Shapes in Aerial Images

This is my Master thesis project at ETH Zurich, which contains two parts:

• Buildings Segmentation (see buildings/Train.py),
• Roads Segmentation (see road_polygon_new/Train.py).

Data should be placed correctly (not in the folder data). See Config.py file for more details. If you want the data please send me an request email.

Arguments:

• --city: the city name.
• --net: backbone of the net.
• --load: resume training or not.
• --vis: whether visualize the prediction or not.

To train the model just simply python3 Train.py --XXX YYY is fine. To run inference just python3 Evaluate.py --XXX YYY.