[Osmf-talk] Daylight Map Distribution (was: Re: Mapping trees as buildings)

[Christopher Beddow]

Yes, I agree really: fundamentally better training of mappers to make good
decisions is the best goal. I spend time doing this and continue to plan to
do so, in webinars and in person. A lot of orgs and communities are also
excellent at this. And participating in community channels like Slack
allows discussion of nuanced issues, improve knowledge of everybody
involved, vs any isolated mapper who doesn't seek and use community
feedback.

There is also a user experience design topic to pursue here especially with
Rapid and the AI generated data: how to make sure it is presented in a way
that can be evaluated and not pushed to be simply accepted and merged into
the map (I don't think any company or org has incentive to make this
happen, nobody is for example profiting $1 per detected item that goes to
OSM--the incentive is to actually reject the bad data and keep the good
ones, aligned across any company or individual).

On OSM US slack or HOTOSM slack, please see the #rapid_feedback channel
where this UX topic comes up. No official plan or roadmap on it, but the
developers of Rapid are listening and interested (and really nice people,
please look for them online and at conferences).

My personal interest in a professional capacity is to figure out how to
utilize rejected data from AI to improve the model, and also remove it from
the dataset. Right now it hides temporarily when rejected. Lots of ideas
here!

On Fri, May 19, 2023, 13:20 Andy Townsend <ajt1047 at gmail.com> wrote:

> On 16/05/2023 07:34, Christopher Beddow**** wrote:
> > It seems the best approach to Emerson's concerns is to use Daylight
> > Map Distribution for any end user case because it removes errors and
> > graffiti from OpenStreetMap that the community does not miss.
>
> I think that that downplays the contribution of Facebook** (who produce
> "Daylight Map") here.  I've not seen an example where "errors and
> graffiti" is removed from the "Daylight" map but left in OSM.  The
> nearest example I've seen are objects with no useful OSM tags (perhaps
> something with a "name" tag and nothing else), but they're not otherwise
> problematic.  Instead, my experience is that when Facebook people spot
> errors in OSM they fix them in OSM, just like other OSM mappers do.
>
> Data consumers can choose to use Facebook's distribution from 2-6 weeks
> ago or OSM's from a couple of minutes ago - both will have "all the
> errors that have been spotted corrected" and both will have "some errors
> that haven't been spotted yet".  There will possibly be errors in
> Daylight that are fixed in OSM (and will therefore be fixed in the next
> Daylight distribution), but given the surprisingly small amount of OSM
> vandalism across the board any examples would be pretty hard to find***.
>
> However, none of this is relevant to finding "trees mapped as buildings"
> by a mapper who thought that tools such as "Rapid" were better than they
> actually are at detecting things.  The only solution there is better
> training of mappers using Rapid and similar tools, and if problems are
> still occurring, then what training there currently is clearly isn't
> good enough yet.
>
> Best Regards,
>
> Andy
>
> (from the DWG, but writing in an entirely personal capacity here)
>
> ** or whatever that company calls itself now - apologies for my lack of
> knowledge of how the various companies are arranged.
>
> *** comparison of "Daylight" and "OSM" is pretty easy with e.g. Osmium -
> extract the area you want from each, dump both into "opl" format, remove
> the stuff that Facebook remove from OSM, and diff.
>
> **** I'm guessing that this is the Christopher Beddow who a web search
> suggests is "Christopher Beddow - Geospatial Data Analyst - Meta", but
> presumably is writing in a personal capacity here.
>
>
>
>
> _______________________________________________
> osmf-talk mailing list
> osmf-talk at openstreetmap.org
> https://lists.openstreetmap.org/listinfo/osmf-talk
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/osmf-talk/attachments/20230519/e07f4103/attachment.htm>