Aerial view of fields of crops

An open-source ML library for

deep learning on satellite and aerial imagery

Build intelligent geospatial applications with deep learning.

Raster Vision is an open source library and framework that bridges the divide between the world of GIS and deep learning-based computer vision. It provides a configurable computer vision pipeline that works on chip classification, semantic segmentation, and object detection.

Diagram illustrating the different applications of Raster Vision

How it Works

Introducing the Raster Vision pipeline

Example is showing a semantic segmentation task

Demonstration of how Raster Vision moves through images chip by chip Demonstration of how Raster Vision moves through images chip by chip

01 Data input

Input set of images and training data, optionally with Areas of Interest (AOIs) that describe where the images are labeled.

Start labeling with GroundWork
Diagram of Raster Vision's ability to analyze, predict, evaluate, chip, bundle, and train

02 Raster Vision pipeline

The computer vision pipeline configuration documentation is easy to read, reuse, and maintain.

Explore documentation
Computer vision rendering of how Raster Vision moves through images chip by chip Computer vision rendering of how Raster Vision moves through images chip by chip

03 Deployment

Model bundle is deployed in batch processes, live servers, and other custom workflows.

Scaling beyond
geospatial tech

Raster Vision is versatile and can seamlessly handle the idiosyncrasies of working with massive image datasets across a broad range of industries.

Build your custom pipeline
  • Forestry
  • Healthcare
  • Academia
  • Supply Chain
  • Energy Transition
  • Oceans & Water
  • Climate Adaptation
  • Non-profit
  • Agriculture
  • Artificial Intelligence

Built on popular technologies & open source libraries including

AWS (S3 and Batch), PyTorch, TorchVision, Albumentations, Rasterio, Shapely, Gdal and Numpy.

Why Use Raster Vision

Employ extraordinary data processing power to meet your goals.

Customizable

Utilize custom datasets, model architectures, and arbitrary loss functions.

Repeatable, configurable workflow

Create consistent and maintainable results.

Open source

Benefit from a constantly updated & supported framework released under the open source Apache 2.0 license.

Handles geospatial datasets

Handles a variety of geospatial file formats, map-based coordinates, incomplete labeling, and multiband (3+ bands) imagery.

Supports massive imagery

Works out-of-the-box with massive imagery commonly used in the geospatial field.

Low barrier to entry

Enjoy a high-level programmatic API with sensible defaults for configuring modeling pipelines. Doesn't require expertise in PyTorch or deep learning.

Fast AWS Batch setup process

Skip the manual setup and use CloudFormation templates.

Extensible architecture

Object-oriented architecture is extendable to new computer vision tasks and deep learning frameworks.

Easily use models in various deployment scenarios

Quick & repeatable configurations

Select a scenario

Based on open-source data.

An open-source computer vision and machine-learning framework developed by Azavea.

© Azavea Inc. All rights reserved.