Learning to Use Machine Learning - A learn along for folks who want to be using ML in their work.
In case you missed it, Dale Kunce tweeted this out yesterday:
The day of Machine Learning and OSM/Humanitarian mapping reckoning is
getting closer. Very excited for the possibilities these new methods
have for @hotosm @RedCross. Next frontier is making HOT and
@TheMissingMaps more valuable than just a training dataset for the
Toward that end, I have been watching and in some cases working with
various ML tool chains over the past 2 years and really, not having a
lot of luck with my level of skill and knowledge. I am a pretty
advanced sysadmin, comfortable on the command line, but understanding
the terminology and installations has been a bit beyond me.
So if anyone is like me and sees all of these great tool chains and
would like to learn how to use them with your peers learning along
with you and hopefully some experts as well, I created a dedicated
#mlearning-basic channel on the OSM-US slack (
OSM-US runs a lovely, informative, lively, international slack with
many channels and everyone is welcome!
The #mlearning-basic channel is for the absolute beginner basics, how
to install and use the existing and emerging tools chains and OSM/OAM
data to generate usable vector data from Machine Learning quickly.
You are all invited to join, but it is very basic. Hopefully some of
the ML experts from the projects below will be in there to hand hold
us newbies through actually making use of what we are seeing more and
more everyday. Excellent tool chains exist, world changing tool
chains, now we just need to get them into the hands of the people who
need and want to use them everyday :)
Everyone is welcome and encouraged to join, it is intended to be kind
of a "learn-a-long". Our first project, my first project, is building
on the Anthropocene Labs work and doing the same area using MapBoxes
RobotSat tool chain using Danial's and Maning's posts as a guide.
For reference please see this incredible work the community has shared
in the past months, much like humanitarian mapping in general, the
projects you see below will start changing the world over the next 12
months. Apologies if I missed any other OSM ML public projects, please
reply and let us all know!
Anthropocene Labs @anthropoco
#Humanitarian #drone imgs of #Rohingya refugee camps + pretrained
model finetuned w @hotosm data. Not perfect maps but fast, small data
need, works w diff imgs. Thx @UNmigration @OpenAerialMap @geonanayi
@WeRobotics 4 #opendata & ideas! #cloudnative #geospatial
This post follows Daniel’s guide for detecting buildings in drone
imagery in the Philippines. The goal of this exercise is for me to
understand the basics of the pipeline and find ways to use the tool in
identifying remote settlements from high resolution imagery (i.e
drones). I’m not aiming for pixel-perfect detection (i.e precise
geometry of the building). My main question is whether it can help
direct a human mapper focus on specific areas in the imagery to map in
Recently at Mapbox we open sourced RoboSat our end-to-end pipeline for
feature extraction from aerial and satellite imagery. In the following
I will show you how to run the full RoboSat pipeline on your own
imagery using drone imagery from the OpenAerialMap project in the area
of Tanzania as an example.
Skynet is our machine learning platform. It quickly scans vast
archives of satellite and drone imagery and delivers usable insights
to decisionmakers. Our partners use Skynet to reliably extract roads
and buildings from images that NASA, ESA, and private satellites and
drones record daily. The tool is remarkably versatile. We are
experimenting with using Skynet to detect electricity infrastructure,
locate schools, and evaluate crop performance.
Deep learning techniques, esp. Convolutional Neural Networks (CNNs),
are now widely studied for predictive analytics with remote sensing
images, which can be further applied in different domains for ground
object detection, population mapping, etc. These methods usually train
predicting models with the supervision of a large set of training