Topographic map segmentation with u-nets.
WIP.
Cloning this repo recursively to get the submodule containing the unet implementation with
git clone --recursive git@github.com:luftj/unet-ms.git
Requires
- Python3 (tested with 3.9.1)
- GDAL (for generating synthetic training data)
$ python3 -m pip install -r requirements.txt
(will also install requirements of the submodule)
-
copy training data to Pytorch-UNet/data/imgs and Pytorch-UNet/data/masks respectively. We use 512x512px tiles.
-
train model as described here
-
segment a map image with
$ ./segment_image.sh [image path]
- fix path to checkpoints