rslearn is a tool for developing remote sensing datasets and models.

rslearn helps with:

1. Developing remote sensing datasets, starting with defining spatiotemporal windows (3D boxes in height/width/time) that are roughly equivalent to training examples.
2. Importing raster and vector data from various online or local data sources into the dataset.
3. Fine-tuning remote sensing foundation models on these datasets.
4. Applying models on new locations and times.

If you are new to rslearn, we suggest starting here:

1. First, read CoreConcepts, which summarizes key concepts in rslearn, including datasets, windows, layers, and data sources.
2. Second, read WorkflowOverview, which provides an overview of the typical workflow in rslearn, from defining windows to training models.
3. Finally, walk through the IntroExample, or find another example in Examples.md that can most readily be adapted for your project.

Other links:

• DatasetConfig documents the dataset configuration file.
• DataSources details the built-in data sources in rslearn, from which raster and vector data can be imported into rslearn dataset layers.
• ModelConfig documents the model configuration file.
• TasksAndModels details the training tasks and model components available in rslearn.

rslearn requires Python 3.11+ (Python 3.12 is recommended).

git clone https://github.com/allenai/rslearn.git
cd rslearn
pip install .[extra]

For linting and tests:

pip install .[dev]
pre-commit install
pre-commit run --all-files
pytest tests/unit tests/integration
# For online data source tests, you can store credentials in .env and they will be
# loaded by pytest-dotenv.
pytest tests/online

For questions and suggestions, please open an issue on GitHub.