TCD Pipeline Documentation

Researchers at Restor and ETH Zurich have developed this powerful and easy-to-use pipeline for detecting individual trees and tree canopy in aerial images. You can also explore our open-access dataset of labelled trees which may be useful for training your own models or benchmarking against other methods.

Trees detected from aerial imagery over the city of Zurich

Head over to the introduction page for general inforamtion about the project.

For more information about using our models and pipeline:

• How to install the pipeline
• Predicting tree cover in images
• Training your own models
• Exporting models for production deployment
• Sample model card
• Datasets and data formats
• Benchmarking
• Output caching
• Pipeline architecture
• API/developer reference

Quickstart

This quickstart assumes that you have Conda installed. Open a terminal and

First, clone the repository:

git clone github.com/restor-foundation/tcd
cd tcd

Then, install the conda environment and install the pipeline:

conda env create -f environment.yml
pip install -e .[test]

And run a test prediction on sample data in the repo:

tcd-predict semantic input=data/5c15321f63d9810007f8b06f_10_00000.tif output=test_prediction

Citation

@misc{oamtcdpreprint,
      title={OAM-TCD: A globally diverse dataset of high-resolution tree cover maps}, 
      author={Josh Veitch-Michaelis and Andrew Cottam and Daniella Schweizer and Eben N. Broadbent and David Dao and Ce Zhang and Angelica Almeyda Zambrano and Simeon Max},
      year={2024},
      eprint={2407.11743},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2407.11743}, 
}