Teaching resources for landscape/use/form image segmentation with U-Nets (a type of deep neural network).
There are two courses that can be accessed and run online through Google Colab
This repository contains versions of those courses that may be adapted for your own purposes, running on the computers you manage and have access to.
Step 1: get the code
open an anaconda command window
git clone --depth 1 https://github.com/MARDAScience/UNets4LandscapeClassification.git
Step 2: create the conda environment
cd UNets4LandscapeClassification
conda env create -f unet_imseg.yml
conda activate unet_imseg
Now, you have a few options
- Run through the CSDMS workflows (parts 1 and 2) as python scripts
cd scripts
python CSDMS_May2020_part1a_of_2.py
python CSDMS_May2020_part1b_of_2.py
python CSDMS_May2020_part2_of_2.py
- Run through the CSDMS workflows (parts 1 and 2) as jupyter notebooks
cd notebooks
jupyter notebook
(open the notebooks from your browser)
- Train a model on your own data
a. make a new folder inside data and organize your images and label images similarly, into train, test and validation folders
b. make a config file like those for the other data sets
c. copy CSDMS_May2020_part1a_of_2.pyand adapt it to your needs, modifying (at least) the paths to your data, and other specifics