Deep learning can analyze enormous amounts of geospatial data, providing an exciting opportunity to automate analysis of large volumes of imagery. However, creating machine learning (ML)-compatible data requires computer vision expertise, accessing models requires knowledge of deep learning frameworks, and model outputs are ill-suited to deployment in a geospatial pipeline. CosmiQ Works has lifted these barriers with its suite of open source software tools for:

1. Generating ML-ready images and labels from geospatial formats,
2. Performing modeling using existing neural networks trained on geospatial data,
3. Generating deployable vector-formatted outputs, and
4. Analyzing model performance, all with a simple and well-documented command line interface.

Our workshop will begin with a tutorial on deep learning concepts. We will then take users through performing and scoring building footprint extraction using the CosmiQ software stack, open source models, and the freely available SpaceNet dataset and labels. We will also help users create their own modeling pipeline using the software stack’s API, enabling users to tailor future analyses to their own needs. The workshop is intended for Python users but does not require a background in deep learning.

The workshop contents can be found at https://github.com/cosmiq/solaris_tutorials/. Attendees will be provided access to an AWS EC2 instance to run code and will not need to pre-install workshop materials prior to attending.