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.
Artus's documentation is available on Read the docs
Here, we provide tutorials to get started with artus. Depending on the data you have and what you want to achieve, you may want to follow the full use cases or the step by step tutorials.
This project is being developed as part of the G2OI project, cofinanced by the European union, the Reunion region, and the French Republic.
If you want to get ready with artus, you should first install it. To help you, you can follow installation instructions!
In the case you have annotated data, whether it is georeferenced images or raster images, you may want to train a deep learning model that is able to detect specifically the classes you annotated manually. According to the type of data you have (raster images, georeferenced images) and the deep learning tasks you want to achieve, the path you will take may differ.
In full_use_cases/rasters_images.ipynb, you will find a complete notebook to train a deep learning model on your data type and use the AI model to predict new geospatial data.
USE CASE 2 : You have images with a GPS associated for all of them and you have a COCO annotation file
In full_use_cases/georeferenced_images.ipynb, you will find a complete notebook to train a deep learning model on your data type and use the AI model to predict new geospatial data.