Configuration Reference

rslearn is driven by two configuration files:

• Dataset configuration (config.json) defines your dataset: the layers, the data sources they are imported from, how rasters are composited, and how windows are stored. This is what you set up when building a dataset and importing raster and vector data into it.
• Model configuration defines model training: the model architecture and components, the training task, data transforms, and the training loop. This is what you set up when fine-tuning a model on a dataset.