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

[Rory McCann]

Hi,

It's great to see Facebook interested in OSM. FB is a large, wealthy and 
influential company, and could definitely use some of that strength to 
help the OSM community.

I looked at the sample data, the data quality is comparable to what is 
mapped in that area already. There are few tagging oddities, but nothing 
insurmountable. So that's already a plus! :)

For imports, we need to see the data, for automated edits, details of 
the algorithm. People can't give a blank cheque to an import/mechanical 
edit when they haven't seen the data/code which powers it.

You're clearly looking for feedback on a whole process. But there's some 
missing info, which I hope you can expand on.

Can you release the source code for the computer vision programme(s) 
which generates the assistance images? Since it's FB internal, there 
isn't the licencing issue like the DG imagery, right? This is 
*Open*StreetMap, so we shouldn't hide things because of the "trade 
secrets" or "financial sense". There's nothing stopping Facebook 
starting it's own mapping project to show off the AI generated maps, but 
you want to partake in *Open*StreetMap, then let's be open. (It is your 
corporate mission as well)

Can you release the fork of iD with the validator? Loading the sample 
data in JOSM shows it has missed some mistakes the JOSM validator 
catches. If you release the validator, we can check for other bugs. 
There's at least one mapping mistake in the sample data that I'm not 
sure /where/ it's coming from.

Can you tell us more about the "internal processes" which means you want 
to use iD rather than JOSM? Do you do any post-processing on the data 
from the mappers? What sort of things are you doing? Do people upload 
directly from your modifed iD editor? There's a least one strange thing 
in your sample data that /miiiiight/ be from some server side 
post-processing. This would be like making an OSM editor, which can be 
tricky to get right. Again, more eyes can help you fix this.

Can you tell us more about the tasking manager which you're using? 
What's the URL for it? The HOT Tasking Manager is public, so if I see 
some data in OSM from a HOT task, I am able to see who has validated 
that task/tile/data. (MapCraft is public too.) Will such a thing be 
released for the community to see who's validating things? The HOT TM 
uses one validator per tile (right?), you mention "multiple people" can 
you tell us the work flow you're using for validation? If you have a new 
validation/collaborative mapping tool, it might be beneficial for others 
to use as well! :)

Can you tell us about the mappers/validators? Is anyone being paid to 
map? Do they have any targets to meet? Do they have time limits to map 
things, or to validate things? Does your project have any goals, like 
"map X km of roads by $DATE"? I presume you're paying your engineers, 
that's not a problem.

All the mappers/validators have nearly the same profile description, do 
they have control/passwords over those account? Or is it all controlled 
by Facebook themselves? The mappers/validators don't link to their 
existing OSM accounts, do they all have experience with OSM? (There is 
where the [username]_import username trick is helpful, you can easily 
see the "main" OSM account).

Can you release more sample data? With more eyes looking at the 
resultant data, we can check for any more possible problems. This is 
only a small segment and I presume you'll be uploading much more, and I 
presume you've created much more already.

The wiki page states you gave a talk about this at SotM 2016, but I 
think it was the SotM US 2016 (the USA is not the world!). ( 
http://stateofthemap.us/2016/how-can-ai-help-us-make-maps/ ) I can't 
find a video, and have heard it wasn't recorded. Is this true? Is there 
any recording of it available? I'd like to learn more about this, to be 
able to give more informed advice. You do mention this talk on your 
import wiki page

Lest, you think I'm anti-import, I do a semi-regular mechanical edit for 
Ireland ( https://wiki.openstreetmap.org/wiki/Logainm_Import ) and you 
can see all the code here ( https://github.com/osmie/logainm-osm-import 
). I might appear negative, but I don't mean that at all. There are lots 
of ways to make this a great new tool in our toolchest.


Rory

P.S.: +1 to "release the DG imagery"

On 15.03.2017 03:39, osm wrote:
> Dear OSM Community,
>
> This email is to share the Facebook process and sample data for review
> by the OSM community. We welcome feedback.
>
> Our goal is to collaborate with the OpenStreetMap community to create
> quality road data, validated by trained mappers and local OSM
> communities. Our import documentation on the wiki
> <https://wiki.openstreetmap.org/wiki/AI-Assisted_Road_Tracing> explains
> the process in depth, but we're happy to answer any other questions you
> might have.
>
>
> To get community feedback we are sharing our sample data (attached) and
> outlining our process. We understand our process is not a traditional
> OSM import because we are creating the data ourselves and conflating it
> with existing OSM. As you will see on the wiki, the**process is similar
> to remote tracing using satellite imagery, where human mappers are
> assisted with road suggestions generated by machine learning algorithms.
> Using the tasking manager, multiple people go through each node and way
> making sure to tag each road appropriately, connect them to current OSM
> data, fix alignment, when necessary, add sensible changeset
> comments/notes and resolve conflicts before finally saving where it is
> merged with the current OSM data.
>
> We are working with just one country at a time and our edits will happen
> slowly by grouping areas into small regions so we complete one before
> moving to the next. We plan to spend the next few months in Thailand.
>
>
> We hope to continue to engage the local community, as we know this is an
> important part of being an OSM contributor. We are open to feedback and
> willing to improve processes and learn from feedback. We are happy to
> share our Tile Service as mentioned here
> <https://wiki.openstreetmap.org/wiki/AI-Assisted_Road_Tracing>. For now
> we can also share .osm files for specific areas of interest, while we
> work on creating a better way to share this through a Tasking Manager or
> something similar that is public facing. So far we have been chatting
> with a few members in the community, but as we learnt from the forum
> discussion we should have also posted publicly and will do that going
> forward.
>
> Here is an example of how we plan divide the tasks for the country. The
> colors indicates the current density of roads going from blue to red for
> high density areas. Of the 77 province boundaries we plan to start in
> the Southern Region in the province of Yala. Image 1
> <http://scontent.xx.fbcdn.net/v/t39.10482-6/16686312_1226149567492788_3453256145248976896_n.png?oh=77893d88f0e4a8dffbb8d0427c9ce026&oe=5936917E\>
>
>
>
>
> Using DG's Vivid+ imagery and have found we can get a 30% increase in
> road coverage for Thailand. Because imagery in OSM is much older in most
> areas we are able to pick newer roads with this imagery. Here are some
> examples of places we have been able to get road masks for in Thailand
> compared with OSM's current imagery. (Image 2
> <http://scontent.xx.fbcdn.net/v/t39.10482-6/16686458_1635190336509160_6301217537245839360_n.png?oh=a47d4813312469aca2068cd113d73693&oe=59304471>,
> Image 3
> <http://scontent.xx.fbcdn.net/v/t39.10482-6/16781470_1359746720759904_2477208192824639488_n.png?oh=eca7a3e9c7e5f644532c693865a9d292&oe=596D0EF6>
> Image 4
> <http://scontent.xx.fbcdn.net/v/t39.10482-6/16686846_1187920531335558_7954053478257524736_n.png?oh=db31bdd4af9acd3d6d309d1687a5ebc0&oe=5926F69E>,
> Image 5
> <http://scontent.xx.fbcdn.net/v/t39.10482-6/16686541_998489256917970_4428448510123180032_n.png?oh=f5e95744de1ab7c20aae37b755353abb&oe=5938D11B>)
>
>
>
>
>
>
>
>
>
>
>
>
>
> Please take a look at the attached .osm file for an area in Thailand
> that contains both our generated and vliadated roads and current OSM data.
>
> Best,
> Drishtie on behalf of the OSM at Facebook Team
>
>
>
>
>
> ------------------------------------------------------------------------
>
> *Frequently Asked Questions*
>
> * *
>
> *What does AI mean?*
> Artificial Intelligence (AI) is training a computer to do things that
> require intelligence when done by humans. In this case it involves
> creating training data for an area by tracing out satellite imagery the
> same way you would when editing and feeding it to a computer. The output
> of this is road masks. Here isan example prediction result
> <http://scontent.xx.fbcdn.net/v/t39.10482-6/16685038_1348901108507836_4612480449938194432_n.png?oh=be39e0c615c0d1dad3e1c55f7b7193ed&oe=59341B65>
> using deep learning for an area in Thailand.
>
>
>
> *Are there limitations with AI?*
> Yes, AI and algorithms still have a long way to go to figure things such
> as assigning tags according to OSM region specific guidelines, figure
> out the correct OSM relation, add bridges over waterbodies or figure out
> connections with crossing railways among many other things. Currently
> all of this is done by our mappers manually.
>
> *Are you adding roads directly to OSM without humans validating?*
> Every single road from the AI output will be added only after multiple
> rounds of human validation. After some processing of the road masks
> shown above we end up with a .osm file. We then treat this like remote
> tracing. Using iD editor and/or JOSM mappers make sure that merging with
> the current data has happened smoothly. They tag each road
> appropriately, connect them to current OSM edits, fix alignment, when
> necessary, add sensible changeset comments/notes and resolve conflicts
> before finally saving where it is merged with the current OSM data.
>
> *Why are you using iD Editor versus JOSM?*
> We know JOSM is efficient at what we are trying to do. In fact, we
> started primarily with JOSM. This said, to make our internal process
> more efficient, our engineers improved the iD editor so our version has
> comparable functionality as JOSM. We are working on making this public
> so we can share it with the community.
>
> *What Imagery are you using?*
> We are using DigitalGlobe's Vivid+ which is high-resolution (50cm /
> pixel, or zoom 18) color-corrected and cloud-free.
>
> *Can we share DG imagery?*
> While our license does not currently permit this. DG as noted in this
> forum <https://forum.openstreetmap.org/viewtopic.php?id=57387> is
> looking at options to publish imagery for OSM editing similar to the
> imagery that is being used by us.
>
> *Who are your editors?*
> Our team consists of 3 engineers and 14 editors, and usernames can be
> found on the Import wiki. All editors have also updated their profile to
> say they map for Facebook.
>
> *Why did you choose Thailand?*
> Facebook has a high number of users in this country and we would like to
> improve that map for this community. It is also our mission as a company
> is to make the world more open and connected and one way we can do is by
> filling in the missing gaps on the map. We also saw a strong OSM
> community that we could learn from and engage with while we refine our
> process for mapping. We are hoping for community feedback as we move
> forward so we can contribute high quality edits.
>
>
>
>
>
> _______________________________________________
> Imports mailing list
> Imports at openstreetmap.org
> https://lists.openstreetmap.org/listinfo/imports
>