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

[James]

I agree with Kate, we are NOT an academic institution, so why are we
requiring proof of the method when the results should be the thing that
matters. It would be like if we required to provide the mathematical method
every time we want to do long division:

 1200 | 10
         --------
-10      120
-----
   20
  -20
  -----
     00
    -  0
    ------
       0

Or you could use another method like:
https://www.youtube.com/watch?v=g0WEwGzyKwI

Method doesn't matter as both would give the same answer. So if we can
validate that the entire dataset has an acceptable(where review and
discussion/flagging errors come in) errors(won't be perfect as perfection
is impossible (acceptable margins: millimeter vs centimeter vs meter error
margin))

To validate would would need acceptable imagery resolution(Bing/Mapbox
might not have this) to determine road locations of roads, check if the
roads are in the right place(i.e. no rivers flagged as roads) and the
entire dataset as it may be 99.999999% correct so the sample size should be
100%


On Fri, Mar 17, 2017 at 12:19 PM, Kate Chapman <kate at maploser.com> wrote:

> Hi Rory,
>
> I'm not in disagreement that providing the code could improve it, though I
> don't think it is required to provide the code that made the data. If we
> are handling this as any other import than providing the entire data source
> should be sufficient. We don't require governments that provide open data
> to provide the tools that they used to make the data. In the three
> processing steps you mentioned I think the only one that would be required
> to make the source code available might be the custom iD if that is the
> point where the data is submitted to OSM.
>
> I'm not disagreeing at all that having additional imagery would be great,
> though I think as Kevin mentioned that is really a separate conversation
> (and seemingly in the works).
>
> -Kate
>
>
> On Fri, Mar 17, 2017 at 6:28 AM, Rory McCann <rory at technomancy.org> wrote:
>
>> Hi,
>>
>> Yes, we must treat Facebook the same as all others. We ask importers to
>> show all the data. FB have shown part of the data. If they want to import
>> lots of data for Thailand, then they could show all of that data for review.
>>
>> The Automated Edits & Imports guidelines do call for documentation of the
>> algorithms/source code. I think there are ~3 processing steps. The computer
>> vision code, the custom iD editor (w/ validator), and (maybe) some server
>> side processing/"internal processes". Many places bugs could creep in. With
>> many eyes, we can find bugs and possibly help make it better.
>>
>> From my reading, you should show the data or the code. Yes, we don't ask
>> for ArcGIS to be open source, but in that case one provides the data.
>>
>> :)
>>
>> Rory
>>
>> P.S.: I suggested sharing the imagery because OSM has 2 main global
>> imagery source (Mapbox & Bing). It would be great to have a third. :)
>>
>>
>>>      ____
>>>
>>>
>>>     _______________________________________________
>>>     Imports mailing list
>>>     Imports at openstreetmap.org
>>>     <mailto:Imports at openstreetmap.org>
>>>     https://lists.openstreetmap.org/listinfo/imports
>>>     <https://lists.openstreetmap.org/listinfo/imports>
>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> Imports mailing list
>>> Imports at openstreetmap.org
>>> https://lists.openstreetmap.org/listinfo/imports
>>>
>>>
>>
>>
>> _______________________________________________
>> Imports mailing list
>> Imports at openstreetmap.org
>> https://lists.openstreetmap.org/listinfo/imports
>>
>
>
> _______________________________________________
> Imports mailing list
> Imports at openstreetmap.org
> https://lists.openstreetmap.org/listinfo/imports
>
>


-- 
外に遊びに行こう！
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/imports/attachments/20170317/163dde30/attachment-0001.html>