[Imports] Facebook's AI-Assisted Road Tracing for OSM

Kevin Bullock kevinbullock10 at gmail.com
Thu Mar 16 23:30:38 UTC 2017


Hi Everyone, it's Kevin representing DigitalGlobe. I do understand the
situation and I think some good points have been made in this thread. If
you've seen my SOTM presentations from last year, I have spoken at length
about the DG business model of licensing data. In short, we do this to
sustain our business and to operate, develop and launch more satellites!

I'm happy to report we are making progress towards an exciting option that
would allow DG to publish an imagery layer directly for OpenStreetMap
editing. This is being funded by a group of organizations, and would make
the +Vivid imagery layer available for registered OSM editors (the source
for Facebook). I am hoping this will resolve many of the concerns for
validation. The timeframe for this is likely 4-8 weeks. It will likely
warrant a new thread/forum somewhere instead of being embedded in the
Thailand thread and this one.

Thanks to everyone for their feedback and agreeing with the sentiment of
appreciation to Facebook for all the work they are doing to improve OSM and
sending informative communications.

Kevin Bullock

On Thu, Mar 16, 2017 at 5:11 PM, Andy Townsend <ajt1047 at gmail.com> wrote:

> On 16/03/2017 21:46, Kate Chapman wrote:
>
> 1. We can't demand more of Facebook than others doing imports. When a
> government releases data under an open license we don't demand to see the
> imagery they used to make it. I purchase the same imagery Facebook does for
> my job and our license also ensures that we have the rights to release the
> derived vectors under the license we choose. I think this is sufficient.
>
>
> Indeed, but it's worth bearing in mind a couple of points:
>
> 1) One is that unless people can compare the source DG imagery with the
> the derived roads, it's not possible to judge the process.  The first time
> that Facebook tried something like this in Egypt nad Thailand there were a
> very large number of false positives - many straight edges with a contrast
> change in the image (e.g. the edges of drainage channels in Egypt) were
> detected as "roads", and even with real roads there were many connectivity
> issues.  It may well be that the method by which roads are now detected is
> much improved (but it looks like image 3 may show some of the same
> connectivity detection problems as before), but without access to the DG
> imagery used, no-one can tell.
>
> 2) The discussion on the Thai forum has suggested that it'd be useful for
> OSM mappers to access the DG imagery for mapping purposes other than road
> derivation (see e.g. https://forum.openstreetmap.
> org/viewtopic.php?pid=632472#p632472 and https://forum.openstreetmap.
> org/viewtopic.php?pid=634119#p634119 ).
>
> Obviously there may be commercial hurdles to climb in order to do either
> of these, but it'd be difficult to comment on the accuracy of the the
> import without at least (1).  However in case people are unaware DG have
> already responded in the forum thread https://forum.openstreetmap.
> org/viewtopic.php?pid=634735#p634735 and said "we are looking at options
> to publish", so this may be in hand.
>
> Best Regards,
>
> Andy
>
>
> _______________________________________________
> Imports mailing list
> Imports at openstreetmap.org
> https://lists.openstreetmap.org/listinfo/imports
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/imports/attachments/20170316/99b46022/attachment-0001.html>


More information about the Imports mailing list