Machine Learning Software Engineering Adeel Hassan

Introducing Raster Vision v0.21

08.25.2023

This release brings some exciting new functionality to Raster Vision making it easier than ever to apply machine learning to satellite imagery.

We highlight some major changes below. For a full list of changes, see the changelog.

Raster Vision can now consume imagery from STAC APIs

See tutorial: Imagery from STAC API + labels from OSM.

You can now pass an Xarray DataArray representing STAC items (e.g. from the results of a STAC API search query) to the newly-added XarraySource to get a RasterSource that is compatible with the rest of Raster Vision.

Code snippet from the tutorial, *Imagery from STAC API + labels from OSM*, showing how to convert an Xarray DataArray into a Raster Vision RasterSource.

A screenshot from Raster Vision — Code snippet from the tutorial, *Imagery from STAC API + labels from OSM*, showing how to convert an Xarray DataArray into a Raster Vision RasterSource.

Raster Vision can now work with temporal data

See tutorial: Working with time-series of images

Raster Vision can now read in time series of images via TemporalMultiRasterSource and XarraySource. These temporal RasterSources can be used as drop-in replacements for normal RasterSources.

Time-series of Sentinel-2 images over Paris, France as read from a temporal XarraySource. Image from the tutorial, *Working with time-series of images*.

A screenshot of 6 aerial images with dates above them ranging from June 2 through June 17, 2023. — Time-series of Sentinel-2 images over Paris, France as read from a temporal XarraySource. Image from the tutorial, *Working with time-series of images*.

Raster Vision can now export and use models in ONNX format

The model-bundles produced by Raster Vision now additionally include the model in the ONNX format, and Raster Vision will use an ONNX runtime to make predictions if the environment variable RASTERVISION_USE_ONNX=1 is set.

Raster Vision model bundles now include the models exported to ONNX format. Screenshot from the tutorial, *Training a model*.

Screenshot of Raster Vision demonstrating the ONNX format and ONNX runtime prediction. — Raster Vision model bundles now include the models exported to ONNX format. Screenshot from the tutorial, *Training a model*.

Useful links

For more information, and to get started using Raster Vision, view the resources below.

Website: https://rastervision.io/

GitHub repo: https://github.com/azavea/raster-vision

Documentation: https://docs.rastervision.io/en/0.21/index.html

Tutorials: https://docs.rastervision.io/en/0.21/usage/tutorials/index.html

Changelog: https://docs.rastervision.io/en/0.21/changelog.html#raster-vision-0-21