This Repo conatins several Notebooks which were used to work on the Tree-segmentation Model
To reproduce the exact environment used for this project, run:
pixi install
This will create an environment which can then be used as a kernel for the jupyter Notebooks.
The most important Notebook is the one used for prediction it will load the deeptrees library and monkey patches it, so it can be used with 6 channel .tif Images
This Notebook should be setup in such a way that it loads the Model saved in data/pretrained_models and predicts the Tree crowns in the example-image , the results (shapefiles) can then be visualized in a software like QGIS.
construct_new_Net.ipynb - Notebook used for constructing a 7/9 Channel-Unet based on a pretrained 5(7) Channel UNet
stack_images.ipynb - Notebook used for stacking LiDAR information into the RGBI.tif
Train_net.ipynb - Notebook where the training of the Net actually happens, requires a Set of Images and corresponding shapefiles for training/validation. This was not included in this repo, as it would increase the Filesize massively