treespec 0.1.0

A deeplearning pipeline to classify tree species in terrestrial panorama pictures.

treespec

A deeplearning pipeline to classify tree species in terrestrial panorama pictures.

About

treespec is a PyTorch-lightning based deep learning pipeline equipped with tools helpfull for creating datasets from images, 3D point clouds and shapefile data containing inventory information.

Features:

• match tree inventories
• create masked datasets
• train standard torchvision models

Installation

Method 1: Docker Container

Download the current treespec container on Docker Hub (Docker required).

docker pull vogelingmar/treespec:latest

Run the container with docker run and mount folders you want to work with.

docker run -it --gpus all -v *local_path*:/workspace/data vogelingmar/treespec

Method 2: GitHub Repository

Clone the treespec repository from GitHub (Git required).

When first setting up treespec you have to have Python3 installed on your system. To create a virtual environment and install all the required dependecies to run the treespec pipeline follow these steps:

1. Navigate into your local treespec repo.

cd treespec/

2. Run the setup script.

bash setup.sh

Usage

1. Activate the virtual environment created by setup.sh.

. venv/bin/activate

2. Run the pytest tests to check if everything works.

pip install -e .[dev]; pytest test

3. Configure the settings of the scripts (src/treespec/scripts) in the src/conf/config.yaml file (see config.py/ config_parser.py for available options).

nano src/conf/config.yaml

4. Run any script (example: train.py).

python src/scripts/train.py

Now you should see the training progress in your terminal, along with some metrics. In the end you can see some statistics and the trained model is saved to src/io/models.

If you want to look further into the training statistics run this command and follow its instructions.

tensorboard --logdir=lightning_logs/

5. For further help you can build the documentation.

pip install -e .[docs]; cd docs; make html

You can now find the generated html files in docs/_build/html.