Installation

You can install the package either from the Python Package Index (PyPI) or directly from the source on GitHub. Please not that this package requires Python >= 3.12.

Install from PyPI

The easiest way to install the package is via pip:

pip install leaf-toolkit

This will install the latest released version of the package from PyPI.

Install from Source (GitHub)

If you prefer install the latest development version directly from GitHub, clone the repository manually and install in editable/development mode:

git clone https://github.com/RadekZenkl/leaf-toolkit.git
cd leaf-toolkit
pip install -e .

This allows you to make changes to the source code and have them immediately reflected without reinstalling.

Requirements

Make sure you have Python >= 3.12 and pip installed.

If you are going to use data_prep module make sure that your environment has a sufficient version of gcc. If you are using a conda environment execute: conda install -c conda-forge gcc. If you are running directly on the system running: sudo apt install build-essential (or equivalent) should help.

Installation Validation for Inferece and Evaluation using GPU

Run the following code block to check if everything is installed correctly.

from leaf import models, visualization, metrics

models.test()
visualization.test()
metrics.test()

This downloads a sample image into test/images/… and all the necessary model weights. Afterwards it predicts on the image. Please note that the test runs on a sample image which has resolution of 4096 x 4096 px. We adjusted the patchsize so <8Gb of VRAM is required. The inference results are saved in test/export/…. The visualization test then utilizes the inference results to produce visualizations in test/visualizations/…. Consequently, the test of metrics utilizes the inference results to produce a computed metrics file in test/canopy_test.csv.