Artus : a GeoAI package to produce maps thanks to deep learning

Artus is a python package to automatically produce maps thanks to deep learning models. With artus, you can train deep learning learning models (neural network) on raster images annotated with vector files. You can then use the trained model to predict spatial occurrences on new unlabeled rasters. Predictions can be exported to a GeoJson format and uploaded in your favourite GIS software.

To handle large raster file, artus provides a way to tile raster into smaller tiles according to different cutting grids.

Artus has already been implemented in three use cases using 3 differents inputs data : satellite images to detect gillnets vessels, orthomosaics to detect corals species and underwater images marked with a georeferenced point to detect marine species.

For example, the following map is generated by automatically detecting dead corals on images associated with a single GPS point:

You can also work directly on raster files (like the following underwater orthomosaics). Because this raster was heavy (more than 1Go) and could not fit in GPU memory, we tiled into smaller splits with artus and predict corals occurrences thanks to a trained deep learning model:

Installation

• Installation process for Python
  • 1. Prerequisites
  • 2. Installation
  • 3. Try artus!

Tutorials to learn

If you wan to use Artus and you don’t know where to start, notebooks have been prepared for you! Have a look to this repo : tutorials.

Tutorials

• 🛰 Train an AI model with geospatial data and use it to predict spatial occurrences
  • 🎬 Scenario
  • Input data needed
  • 1 Prepare data for training
    • 1. 1. Tile a heavy raster file
    • 1. 2. Convert annotations files into COCO format
    • 1. 3. Data splitting
  • 2 Train a deep learning model
    • 2. 1. Configure config file and train
    • 2. 2. Evaluate model
  • 3 Use the model you trained to predict new data!
    • 3. 1. Tile heavy raster
    • 3. 2. Deploy an unlabeled fiftytone dataset
    • 3. 3. Predict new annotations thanks to AI!
    • 3. 4. Export the results into a spatial format! 🎉
• 📷 Train an AI model with images associated to a GPS point and use it to make maps
  • 🎬 Scenario
  • Input data needed
  • 1 Data splitting
  • 2 Train a deep learning model
    • 2. 1. Configure config file and train
    • 2. 2. Evaluate model
  • 3 Use the model you trained to predict new data!
    • 3. 2. Deploy an unlabeled fiftytone dataset
    • 3. 3. Add image locations
    • 3. 4. Predict new annotations thanks to AI!
    • 3. 5. Export the results into a spatial format! 🎉

How to cite

Justine Talpaert Daudon. (2023). artus v0.X: A tool to automatically procude mamps with artificial intelligence.(v0.X). Zenodo. DOI:10.5281/zenodo.7852855

Financials partners

This project is being developed as part of the G2OI project, cofinanced by the European union, the Reunion region, and the French Republic.

API

• artus
  • artus package
    • Subpackages
    • Module contents

Indices and tables

• Index

• Module Index

• Search Page