[Imports] Microsoft Buildings Import Inquiry
dr.kludge.gm at gmail.com
Tue Apr 30 15:27:41 UTC 2019
On Mon, Feb 18, 2019 at 9:16 AM Christoph Hormann <osm at imagico.de> wrote:
> On Monday 18 February 2019, Rory McCann wrote:
> > > If you ask the mapper why they chose a particular tag or created a
> > > particular geometry and they do not explain this with a particular
> > > observation in reality or in an image source but ...
> > This aligns with what I said IMO. "The mapper must approve it by OSM
> > standards". If you add some data, you need to have grounds for it *by
> > normal OSM mapping standards*. Which is local knowledge, aerial or
> > survey.
> Well- if for you it does that is great. But it is somewhat tricky to
> assume everyone has the same understanding what "normal OSM mapping
> standards" are. I therefore tried to put this on a somewhat less
> subjective basis.
> > The Microsoft building outline dataset is very simple, containing
> > only the outline of the building. No address, no height, no building
> > type. no name. It can be (and was) derived 100% from aerial imagery.
> > IMO if someone loads that up, and manually approves each building,
> > then that (IMO) is much less of an import.
> The dilemma here is that the reason why people do imports is to have
> less work than when manually mapping things. But if you want to use
> external secondary data sources without actually doing an import you
> also loose most of the efficiency advantage (which results from not
> verifying everything feature by feature but blindly relying on quality
> of the source data in some way).
> I am all for more mechanical support for mappers' work but fully machine
> generated data sets are simply not that useful here (outside the
> function of serving as a reference for finding missing features for
> example of course). There are many interesting technologies that could
> support human mappers in more efficiently recording their local
> geography in OSM. Replacing the mapper with bots and degrading the
> mapper for cleanup and verification work for bots is not an idea that
> has a lot of potential though (although having an 'openbotmap' project
> competing with OSM with such a paradigm would be interesting).
> Regarding the Microsoft building data by the way we had a discussion
> regarding a possible import half a year ago which might be interesting
> to read for anyone considering to use the data either in or outside of
> OSM in some way:
Thank you all for a great discussion. Many years ago there was a four day
mapping event called Operation Cowboy. As I recall I came in third behind
two German mappers. We all manages to add 50,000 plus nodes in those four
days. There were only around 1,000 nodes separating each of us. I remember
being fatigued at the end of the event. Another, 1,000 nodes to take
second place would have been impossible. With this perspective, the Bing
Outlines are a great gift.
The outlines start out as square polygons. I find some of the resulting
polygons interesting. For example, the computer aided drawing tool drew
the correct outline but it is slightly skewed. Another example is that
the building was drawn but other surrounding features of the imagery
somehow made the Computer Vision, CV software draw the building short of
the real outline. I am still working on the wiki page. Based on this
discussion and smaple data, I worked on small procedures that new mappers
or mappers not familiar with JOSM can see what would be required and what
tools in Josm are used to bring the Bing Building outlines to higher
quality. Hats off to Blake Girardot. I did not know about the JOSM merge
feature. That is a great tool to help a mapper focus on the import data in
one layer via the todo plugin. Then the merge layer function is invoked
followed by running the validator. Thus a mapper can catch any overlapping
new outlines that would overlay an existing outline.
As part of developing the import methodology in the wiki page, I have
created two study areas. One is listed in the wiki while the other has not
been added yet. The one not listed is a small town near the AZ Mexico
border, San Luis https://www.openstreetmap.org/#map=16/32.4977/-114.7886 .
The other is near the Scottsdale Airport where I craft mapped the airport
area. I walked across Thunderbird Road in this area between Scottsdale Road
and Tatum Boulevard. I have since added addresses from some of my Mapillary
sequences to these new features.
https://www.openstreetmap.org/#map=17/33.61108/-111.93858 . I certainly
was motivated to protect my work and others with this import. Again that
two layer method plus the merge feature are great tools to protect the map.
I just completed cutting up Arizona this last weekend. The data pointer is
in the wiki page in a table.
I have notified both the local PhxGeo meetup group  and they have
established tasking manager projects already for Phoenix only. 
Back to Operation Cowboy, Pascal says that these sample study areas are
worth around 31,000 nodes. I have not hit the 50,000+ mark yet.
However, I am just not fatigued by the effort. All the little procedures
that I focused on in the wiki page make this a great joy to add the
buildings. Starting out with a squared outline is a great great gift. All
the other JOSM features make the effort less taxing. You will note that I
accidentally deleted an existing outline. I reverted that one feature
because there was no need to add a new building outline from Bing. I
cannot explain how that happened.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Imports