This work builds on early research by @JCornebise and @deworrall92 who in 2017 built a machine learning model which automatically detected “human presence,” as well as “destroyed” and “partially destroyed” villages in Darfur, Sudan.
https://aiforsocialgood.github.io/2018/pdfs/track1/80_aisg_neurips2018.pdf
This research was possible thanks to thousands of digital volunteers who joined #AmnestyDecoders projects Decode Darfur and Decode the Difference, generating 13 million annotations of satellite images, covering 2.6 million satellite tiles.
Human rights researchers at @amnesty teamed up with engineers at @elementai to automatically detect and quantify the destruction of homes in #Darfur. The new model achieved a precision of 85% and a recall of 81%. More technical details in the blog!
To visualize the predictions, we built a web app that superimposes the probability of human presence/destruction on a map. Given the risks involved of such data falling in the wrong hands, this is only available to @amnesty researchers.
Massive thanks to all those who made this happen including @JCornebise, @AI_Micah, @deworrall92, @alkalait, @laure_delisle, Alex Kuefler, Denis Kochetkov, @BuffyPrice, @sherifea and to @SatAppsCatapult and @Maxar for access to imagery.
You can follow @milena_iul.
Tip: mention @twtextapp on a Twitter thread with the keyword “unroll” to get a link to it.

Latest Threads Unrolled: