[OSM-talk] Facebook mapping highways using AI in collaboration with OpenStreetMap

[Naveen Francis]

Thanks Simon for the caution.

Taking the of help AI doesn't mean that we do comprise on the quality.

All the methods for initial road tracing, GPS tracing, Satellite image
tracing or AI-assisted human mapping (which I have tried) has limitations.

On Indian roads, we try to document as much as possible so that we don't
make errors.  <https://wiki.openstreetmap.org/wiki/India:Roads>

On the rhetoric question:-
We are using OSM maps life savings projects. https://keralarescue.in/map/ .
(2018 Kerala floods maps)
So both quantity and quality are equally important.

Another good thing is we have already implemented your suggestion -
sensible comparison would be completeness measures per road categories

Categories wise completeness
NH : https://wiki.openstreetmap.org/wiki/India:National_Highways_(statewise)
SH:- https://wiki.openstreetmap.org/wiki/India:Roads#State_highways

All the best for fixing Tiger imports.

thanks,
naveenpf


On Sat, Jul 27, 2019 at 4:15 PM Simon Poole <simon at poole.ch> wrote:

>
> Am 26.07.2019 um 19:30 schrieb Naveen Francis:
>
> Including my ₹ 0.10 (Indian ten paisa)
>
> Echoes same thoughts of Brazilian Real.
>
> AI-assisted human mapping tools will be a good aid for the OSM community.
>
> "Map faster, Map better".
>
> 40,00,000 kms to be mapped in India.
> 15 years of OSM mapped 18,00,000 kms.
>
> The (rhetoric) question is, why is this the case?
>
> Because the community in India is still very small relative to the
> population size.
>
> So from where will the additional contributors come from that will turn
> the additional 4 million road geometries in to something really useful?
> There is a real danger of the desire for "completeness" instead of quality
> resulting in multiple TIGER 2.0s, and we are just now slowly working
> ourselves out of the hole we dug (full of good intentions) with the
> original.
>
> Note on the side: outside of raw total  road length, a much more sensible
> comparison would be completeness measures per road categories (which I
> suspect is likely to look far less dramatic) and which might give more
> realistic goals for the community.
>
> Simon
>
>
> thanks,
> naveenpf
>
>
> On Fri, Jul 26, 2019 at 4:42 AM Sérgio V. <svolk2 at hotmail.com> wrote:
>
>> Just adding my R$0,02 (Brazilian Real).
>> I guess soon the AI assisted Human mapping will happen, it may be a very
>> good help.
>> But I can't evaluate what's been publicized July 23, 2019 by
>>
>> https://ai.facebook.com/blog/mapping-roads-through-deep-learning-and-weakly-supervised-training
>> "To browse our machine learning road predictions or start mapping with
>> RapiD, please visit mapwith.ai."
>> So at "Map faster, Map better" https://mapwith.ai/#14/6.13864/6.7698 ,
>> I actually can't evaluate any result for roads at max zoom level 14, to
>> see if it's really better. I can just believe it can be.
>>
>> - - - - - - - - - - - - - - - -
>>
>> Sérgio - http://www.openstreetmap.org/user/smaprs
>> _______________________________________________
>> talk mailing list
>> talk at openstreetmap.org
>> https://lists.openstreetmap.org/listinfo/talk
>>
>
> _______________________________________________
> talk mailing listtalk at openstreetmap.orghttps://lists.openstreetmap.org/listinfo/talk
>
> _______________________________________________
> talk mailing list
> talk at openstreetmap.org
> https://lists.openstreetmap.org/listinfo/talk
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/talk/attachments/20190729/ed14d4a1/attachment.html>