Raster Vision

Raster Vision is a Python framework for building deep learning pipelines on raster imagery, such as satellite and aerial images. Its documentation describes the full path from raw data to a trained model: reading raster and vector data and labels, defining scenes and areas of interest, sampling training data, training a model, then running prediction and evaluation. The tutorials cover plotting samples with visualizers, working with pre chipped datasets, pulling imagery from a STAC API with labels from OpenStreetMap, working with time series of images, using the framework with Lightning, and predicting with an ONNX model. A pipeline system and command line interface run the work as configured commands, and the docs include guidance for using GPUs, running on AWS, setting up AWS Batch with CloudFormation, and bootstrapping new projects from a template. An architecture and customization section explains how to extend the framework through plugins and custom configs. The tool is aimed at data scientists and engineers working with geospatial imagery who want a repeatable, configurable pipeline rather than one off scripts. The documentation reflects version 0.31.1. The site publishes no pricing.