Test Drive - AI-assisted road import by Facebook

Hello Drishtie,

I am missing a bit feedback from other mappers, especially from the imports list. Did you cross post the announcement of the test data set there?

Hello Stephan,

Other than https://lists.openstreetmap.org/pipermail/imports/2017-April/thread.html. I have not cross posted this.

Thanks,
Drishtie

DELETED ORIGINAL MESSAGE.

Sorry, meant to show someone how to use the forum, and hit submit instead of preview.

Hello Drishtie,

it’s a bit sad to see such little feedback from others. Based on the example you had given it is reasonable that you started importing.

I ask you to plan for a little pause after you have imported let’s say 25 tiles. This gives us the chance and time to review your work.
As there had been minor issues it sounds reasonable to have a chance to spot them early and correct tagging style if needed.

Don’t forget that your team is mapping way faster than we have the chance to review. So bringing in too many chunks can later cause problems as we did not have the chance to spot systematic problems not observed in the initial samples.

Can you give some details on the planned progress of the import? Which timeline and which areas are planned? Have you worked ahead internally already and have a huge backlog waiting to be submitted at once?

Thank you,

Stephan

In reviewing some of the areas being discussed I came up with several questions.

1.)

I don’t see the need to add this sort of way either but more importantly, the track goes right through a couple of buildings. I would never assume something like this was real without a ground check.

2.)

Then, checking this example from Tony’s post and putting aside the question of whether or not it passes between those buildings, I’m wondering why the short way (Way: 491189478) was tagged as a path, a key whose usage is already contentious, when it looks just like other ways in the neighborhood that were tagged residential.

How was the decision to tag that way and the short one south of it (Way: 491189476) as highway=path? I would have made both of them either residential ways or a service ways, even living_street, but never a path unless I could verify my choice on the ground. The Wiki definition of path is:

Although I haven’t been to the area, I’m quite sure that people are driving on that way, at least with motorcycles, which would make the highway=path tag inappropriate.

I’ll try to look at some of the other examples in the next few days and provide feedback.

Best,
Dave

Maybe related to the statement of Digital Globe to make imagery for us as well. Not sure about coverage in Thailand. The map linked on the website has areas with only low resolution landsat coverage.

Would be great to know from Facebook if that is the same imagery they are using.

http://blog.digitalglobe.com/news/digitalglobe-satellite-imagery-launch-for-openstreetmap/

Hello Stephan and Dave,

Thank you for your feedback. I was hoping for more local feedback as well.

We completely agree with you on being careful to add small amounts of tiles so we can be sure everything is ok before we move on. We have completed 5 tiles last week and can stay below 25 per week for now. Keep in mind we are also running OSMCHA and Osmose for every single tile in addition to JOSM Validator just to be extra careful.

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 area of Songkhla, same of the sample data we shared.

Here is a zoomed in version of the task we plan to complete over the next 3 weeks of May.

As for tagging, as mentioned before, we caution on the side of using a more generic tag and not adding extra tags to a road if we do not have on-the-ground verification. We hope that local contributors can help with situations that are ambiguous or difficult to map as remote mappers.

In the examples you mentioned Way: 491189478 and Way: 491189476 you can see in DG it does not go through any buildings. For Way: 491189476 because of the tree cover we could not see if it was a residential road, but can see there is definitely a way, so we decided to go with path until it can be verified on the ground that a car can actually pass. We hope that local contributors can help with situations that are ambiguous or difficult to map as remote mappers.



Lastly, the imagery you see now is not the same as what we are using. I have heard Thailand is on the roadmap to be added so you will see the same high resolution imagery we use by the end of the Month :slight_smile:

Thanks,
Drishtie

Isn’t this what highway=road is for? For cases where you don’t know what the classification is. “from the smallest footpath to the largest motorway” I suggest you use highway=road for these unsure things, highway=path means it’s a path. highway=road is a good way to highlight to other mappers “Hey, there’s something here you should check out”.

May hang off on any imports until this imagery is released?

Helo Rorym,

Thank-you for the feedback. We have only added 9 tiles of data and are moving slow to give the community time to check and hope to see new imagery up soon.

Also great advice on using highway=road for highways we are unsure about.

Best,
Drishtie

Thanks a lot for your efforts.
I have not scrutinized your work in detail, but only taken a quick look at the map if I find some strikingly strange items.
Way http://www.openstreetmap.org/way/497749494 looks wrong near its southern end. When you start from the minor road there, I think its first meters should be part of the parallel track on the other side of the irrigation canal, instead of completely parallel. Also: why do these parallel tracks end there? On the imagery available to us “common OSM users”, we can see that they extend farther south. On the other hand, I doubt that the track on the western side of the canal extends such far north.
Closer to the coast, several residential roads can be seen in the imagery which has not yet been mapped. Is that still waiting for import?
I hope I can help optimizing your AI project with these hints.
Have a lot of fun and success!

Hello Bernhard,

Thank you very much for looking through our edits. We appreciate the feedback. In the case you mentioned our the imagery seems to show both roads parallel to each other.

Our imagery

Current Imagery

Please note the DG Premium Layer should be updating soon so you will be seeing exactly what we see in a matter of days :slight_smile:

Thanks again for your help Bernhard!

Best,
Drishtie

Wow, those new images look great.
The strange track can be confirmed there.
Still, I find roads which are not yet mapped on the new images : e.g. east of the irrigation canal http://www.openstreetmap.org/way/454520276#map=17/7.33753/100.46728 , there is an asphalted road which even has a central line with a length of about 1 km. After a bridge, it continues farther south thru some villages, likely a concrete or older asphalt road there.
Can you improve the algorithm to catch such roads also? In my opinion, they are more important than agricultural tracks.
Thanks a lot for your efforts!

Hello DrishT,

in May it was expected that we might have access to the imagery Facebook is using by the end of the month.

Looks to me this is still not the case.

Let’s look at the following way:
way 528899013

It was added by VLD003 on 30th September 2017.

It was mentioned that the Digital Globe imagery being the one used by Facebook and that we will get access to it. Based on what is visible on these layers this is quite unlikely. Compare for yourself:

Digital Globe Premium:

Digital Globe Standard:

Esri:

Bing:

Mapbox:

As a side note: Google imagery for this area is even more recent and seems to come from Digital Globe as well:

So again the question: When will we have access to the imagery you are using to be able to verify what you do? I recently spotted quite many ways suddenly ending with imagery hinting they should connect to other roads.

Stephan

The quality of the running work is below of what was showed off before. As you can do better, please take measures to do so.

Example way 514898125:

There is absolutely no reason why it should end just in the middle of nowhere.

There are many more ways just ending, even with aerials giving clear hints on how they would connect.

This is why I wish we had the source code, so we could run it ourselves, and try to debug problems like this (maybe?)…

Hi Stephan, Drishtie and others on this thread. I work for DigitalGlobe and can confirm we had a gap in our imagery for this location. In the next few weeks, you will see an update in the DigitalGlobe Premium layer. Unfortunately we had some imagery tiles missing from our ingest and are replacing them now. Thanks, Kevin

Hello Rorym and Stephan,

Welcome back Stephan. Missed you on my trip to Thailand and the last few weeks of uploads. As always its great to have your feedback.

Thanks for flagging the issue with the imagery. Look like DG is taking care of that.

As for the edits. We love your tool (http://www.osm-tools.org/) to find the dangling roads. We definitely aim to map as many roads as we can see but sometimes we miss some. In the cases where roads end it’s usually because we map roads that lead to a building. Secondly we check for errors using JOSM, OSMCha, Keep Right and OSMOSE but these dangling roads don’t show up as an error. We hope the community on the ground can help us fill in some of the gaps.

Best,
Drishtie

I am glad that Facebook’s team takes so much effort to get those roads mapped. A few minor glitches like unconnected roads may happen, and also here in Germany the map is far from perfect (and Germany is among the best mapped countries).
Well, perhaps my “use case” is different from other persons’ use cases: I travel on bicycle, normally away from major roads (I don’t like the noisy traffic). Often, those areas are not mapped at all. Seeing some “unconnected” roads on the map means that the way I am traveling on may perhaps lead to that way and does not end in nowhere before (perhaps somewhen later on…). And some “agglomerations” of minor roads may indicate that a village is nearby etc. Such information does help.
Could you perhaps improve your algorithm to differentiate between paved (apshalt/concrete) and unpaved (dirt road, gravel, …) roads? Just imagine the rain starts during my bicycle tour.

Hello Bernhard,

Thank you for the feedback. The tagging for paved and unpaved roads is actually not done by an algorithm but our team that visually looks at the imagery. In cases we can clearly see is the road is paved or unpaved we are happy to include that tag. When we can not be certain from satellite imagery we refrain from adding the tag incase it does not match what is on the ground.

Thanks again! Really appreciate your feedback. Happy Cycling :slight_smile:

Best,
Drishtie

Riding my bike thru the province west of Phetburi, I was faced with an annoyingly bad quality of the map.
Many roads were unpaved, some of them actually tracks. Some “residentials” did not have any house along it.
The worst case I encountered can be seen on this photo (click for a larger version):

According to the map, there were “unclassified” roads along both sides of the canal, which I already changed (you may need to look into the history of the ways):
http://www.openstreetmap.org/way/514899120/history
http://www.openstreetmap.org/way/514899118/history

I suggest that the tagging policy be revised.

  • If a minor road may be unpaved, add the surface=unpaved tag in case of doubt. Finding out that it has a concrete or asphalt cover while expecting an unpaved road is not as bad as the other way round. Note that we assume a residential/service/unclassified etc. to be paved, a track not to be paved if no other information is given.
  • Public roads may end in an agricultural track (and that track may then connect to another public road). In case of doubt, cut the way into pieces with a highway=track inbetween.
  • If you are not sure if it is an agricultural/forestry track or a public road, prefer highway=track.
  • The traffic on “residential” is mainly caused by the people living in that road (or other people visiting them). “unclassified” roads lead to such “residential” roads. So, many of the current “residential” should actually be “unclassified”.
  • If there is no house along the road, it is not a residential.

The first point is most important in my opinion.