This post doesn't give much technical detail but there's a lot more information in the docs: https://docs.rastervision.io/en/0.8/. Examples on open datasets: https://github.com/azavea/raster-vision-examples.

Awesome stuff. Not just for urban feature mapping and intelligence gathering. But all kinds of scientific and economic research: forestry, land use management, and poverty prediction

http://sustain.stanford.edu/predicting-poverty/

Very very cool! The thinking behind improving the data preparation phase to avoid a mess of scripts that reinvent the wheel is so on-point, especially for satellite data where preprocessing can get messy.

I'm looking forward to trying this out - bookmarked for my next satellite imagery project :)