Facebookstreetmap.com

Hi Stereo,

My point re the screen shot is not the alignment, but the tag ‘track’. In my reading of the wiki, this is clearly a minor road. It links two villages, it’s 9.6 meters wide, seems to have a good surface, and has traffic. It is not a farm or forest access road.

Hi Dave,

My problem with FB is the industrial level they are adding roads and tracks of dubious quality. On the stats page, the top FB employees list roughly 120,000 created highways in the last two months. In the past, before good imagery, an amateur contributor would be just fine drawing less than perfect roads and I was happy to fine tune them when I find them. But these are professional mappers with lots of assets like imagery and AI etc. adding thousands of roads that will take a lifetime to improve. They just seem to place quantity far ahead of quality. If quality were even one of their goals, they would fix obvious problems that they just map around.

This may be mostly just me, but I draw roads with smooth curves. They just look better. FB and many users in the past made stop sign shaped curves that detract from the appearance. It’s much easier to draw then well than fix them later by adding nodes

There are lots of judgement calls in this. When I draw a road that disappears in some trees, then comes out the other side, it is not unreasonable to connect the sections. The use of the tag ‘track’ is just not usually a judgement call, and clarifying its use has failed after many attempts. I find this very frustrating.

If FB could be persuaded to do a better job, I would continue. If we were all in agreement on these issues, they might listen to us.

Cheers, Tom

Regarding your frustration, Tom, let’s try to decide what is a track, what is a service road and what is an unclassified road. If we do it again, maybe the confusion will lessen. Of course, given we always seek consensus, we might not resolve the issue to everyone’s satisfaction. But we must try.

Please, try to add at least some comments in their changesets pointing out specifically which way-id we have a specific issue with.

I just did here:
https://www.openstreetmap.org/changeset/60446779#map=13/15.1773/103.7269
https://www.openstreetmap.org/changeset/60444110

I don’t see the first way as a track, and certainly not unpaved. I would also have connected the other way to it.

way 604736514 is a bit more tricky. It could be a probably low quality road connecting the two villages, justifying as unclassified. Or maybe it is just used for agriculture to reach the fields, then a track. On the ground would be easier to judge than from an aerial.

@Tom, Russ, Dave: I’ll be around beginning of October. Do you want to meet in person to discuss this? I’ll invite you for a drink…

edit1: added another changeset comment link

I’m available for a meeting and always enjoy a chat session. Let me know what date and time work best for you.

Hi Tom,

My name is Jeff, I’m a QA on the Facebook team, and I want to try to address your concerns.

I understand your frustration that we can’t fix everything on the map, however our focus for this import is solely on roads. It would infinitely balloon the scope of the project if our mappers were required to add or fix everything on the map within their task bounds. OSM is a collaborative effort, we’re just trying to do one small part of it.

We do take geometry very seriously and strive to provide quality road data. However, there does not seem to be a great consensus on how detailed a feature should be digitized. We have received feedback both saying our work is too detailed or not detailed enough which makes it difficult to nail down the appropriate target.

For the example track you provided, I suspect Bing and DigitalGlobe Premium have slightly different alignment offsets. Its a very common issue between imageries unfortunately and without nearby GPS tracks it can be impossible to know which is more correct. Please feel free to share more examples, preferably with way ids, so that we can more clearly understand your concerns.

With regards to tracks, I think Stephan found a great example of the difficulty of classifying a lot of these unpaved roads with https://www.openstreetmap.org/way/604736514. This road does link two villages but it appears unpaved and may primarily be just for access to the fields in between. There is also a well paved route already connecting the two areas that is marked as unclassified and only slightly longer. Early on, we would have possibly marked this road as unclassified but after a user did some groundtruthing of a similar example, they found that the quality was much worse than satellite implied. Since then our approach has been to err on the side of caution to avoid routing mistakes.

I fully encourage everyone to add changeset comments as Stephan suggested. It really helps us to understand and address exactly what issues you have with our work and to avoid them in the future.

Hi Folks,

I have just landed in the central plains for retirement. I have mapped extensively around here, starting a few years back when everything was a blank canvas. I generally also use OSM for navigation with my Garmin.

Although I think I sit on the other side with regards to tracks (I prefer small dirt roads to be tagged as tracks), I was horrified to discover widespread appearance of small farm tracks and driveways in our neighbourhood and beyond tagged as residential roads. My thought was that the map is now polluted: you can’t rely on it for navigation any more, because I will find myself stranded in the rice paddies.

I concur with what was said above: err on the side of caution and if in doubt label something as unpaved. I think it will take me weeks to clean up our neighbourhood alone.

I think we should at least consider to mass-tag anything with import=yes as unpaved. It will take a lot less effort to reverse the few roads that are paved than the other way round.

Cheers,
Peter.

@Beddist

I have seen your username often in my work in Thailand over the last 5 years. Congratulations on your retirement. I’m wishing you much happiness in the future. So, you are retiring to the “central plains” but where exactly is that?

Best regards,

Dave

Hi Dave,

Staying with my in-laws at the moment here: https://osm.org/go/4TsPJRCA Hopefully will start building a few clicks away soon.

Kind regards,
Peter.

I have now found evidence that user VLD006 1 year ago has deleted roads and a bridge that I added a year earlier. Check out https://www.openstreetmap.org/way/436489101/history.

Here are my GPS tracks:

https://photos.app.goo.gl/MsbywneiTbehHKvf9

New roads (at the time) are shown in red. You can see in the bottom right that it crosses the canal. That and the adjacent piece are now missing.

Vandalism? Incompetence? Time to mass-revert?

Hi Beddhist,

I apologize for this data error that one of our editors caused. He is not with us anymore so I can only make assumptions on his reasoning. Looking at DG Premium the connection is less obvious and the bridge did not exist yet so i assume he thought it was a mistaken connection. However, in DG Standard or the GPS traces you added the connections can be confirmed as legitimate and the user should have checked both.

Regardless, changing community data like that is not our policy and we try to have as light a hand as possible with it. We assume the community members have the most context and up to date knowledge of the situation so we avoid altering the pre existing data to the best of our ability.

We are putting a ton of data onto the map, the Thailand road network has more than doubled in number since we started, and our editors are only human. We will inevitably make mistakes, but we are committed to cleaning up anything we get wrong the first time. Our team is passionate about the project and they want to make the best map possible, just like everyone here.

I have corrected the geometry of the road in question. Please feel free to post here or comment on any other changesets you might find that are not up to standard. Thanks for bringing it to our attention.

-Jeff

Hi Jeff,

Thank you for the feedback. There are several points that need addressing here:

In the areas where I have looked most of that appears to be private driveways and farm tracks mapped as residential or unclassified. At the very least anything shorter than about 400 m should be either reverted or marked as track. I was a passenger on these roads yesterday and observing my GPS about every other road we drove past didn’t exist. There is no way I can flag or correct all of them. Heck, even as a passenger I cannot set waypoints fast enough.

Roads should never be deleted by armchair mappers, no matter what you see in aerial images. Anything that has a source tag should only be deleted or modified after verification on the ground. I think that is common sense.

We need to remember that what we do in this project has a real-life impact on the people who want to use these maps. It can make the difference between getting to your destination efficiently, or having your day spoilt. OSM in the Thai countryside used to be far more accurate than Google Maps or even ESRI’s GPS maps you have to pay for. I am no longer confident that this is still the case.

I have no idea what your project has done to OSM in other countries. For Thailand I would say: revert all changes and start over, with functioning QA. Then perhaps we could persuade some of the old hands who have left in disgust to come back.

For me this is crunch time. I was pissed off when a while back OSM legal beagles had them delete a large chunk of data, just to change the licence. I got over that and moved on. This is far more serious. The overall quality of the map data is in question. I’m sitting on 2 years’ worth of tracks from Captain Slash that I was meaning to map. I’m holding back on this for now and if it doesn’t get resolved satisfactorily I, too, will vote with my keyboard.

Kind regards,
Peter.

I’m trying to make some videos about my concerns. Very boring and crude, sorry.

https://www.youtube.com/watch?v=SJCf9LUaXfk&t=9s

https://www.youtube.com/watch?v=B39tOGKI_Dc

I’m trying to make some videos about my concerns. Very boring and crude, sorry.

https://www.youtube.com/watch?v=SJCf9LUaXfk&t=9s

https://www.youtube.com/watch?v=B39tOGKI_Dc

Hi Tom,

Funny you picked road 2243 in your 2nd vid: it’s one of my early ones, probably laid down with the old GPS V on a blank canvas, so the accuracy may not have been too great and there was no hi-res imagery available at the time. We can, however, take the original trace of that road as a GPS track, probably as accurate as the other one that is there, which looks like made with a bicycle.

Now let’s move 5 junctions North and we come to this village under a cloud in Bing and DG Premium:

Switch to DG Standard and an unholy mess appears:

The main road badly needs correcting here. It seems clear that the FB user has used the DGS image for drawing, where clouds have obscured the Premium images. As you move S and E the discrepancy between the two increases. Which is correct? Without more traces it’s almost impossible to know. But the point is, the smaller roads cross here straight over the main road, not with a zig-zag as shown and with an additional 3 nodes the whole junction can be improved a lot.

Further, Bing Maps clearly shows the road heading E as Rural Road 4012, so can mark it as such. I don’t think we can say for sure whether any of these roads are sealed or not without going to Google Maps, which we are not allowed to do.

Now, let’s move further N again, just past the crossing of roads 2256 & 2243:

What happened here? Clearly, either the main roads are in the wrong place, or the two minor ones. Switching to DGP it becomes clear:

The mapper has used the DGP image to draw these two minor roads and ignored the image offset. There is clearly a break in the image on the main road, half way to the junction with the minor road. But it doesn’t extend due W, it curves SW, missing the minor road. So which one is correct? We don’t have a source tag for Rd 2357, but the GPS traces and main roads line up better with the DGS images, so I think we should use them. Both these roads need to be moved, the junctions lined up and the T junction with road 2256 needs to be cleaned up. Then, of course, we need to continue along the main roads in all directions and rinse and repeat.

This isn’t rocket science, but it does require to engage eyes and brain. If an organisation wants to mass-edit, then there is a learning curve for all involved and some initial hand-holding and reviewing of edits.

Via a mapper’s profile I have found the wiki page for the FB project: https://wiki.openstreetmap.org/wiki/AI-Assisted_Road_Tracing
That reads all very well, but where is the community engagement? Where is the QA?

You have added 150,000 km of roads. Most of that will have to be corrected. Where do we find the manpower to do that?

Hi Beddhist,

The road in your first two frames is my next video project. If you want to fix it, please let me know. I also noticed the offset, it’s pretty confusing.

I hope you understand that I am never complaining about OSM mappers who worked with about nothing in the past. I do think FB should be able to do better given all their resources now. But they don’t.

BTW, I also prefer to bicycle.

Best Wishes, Tom

No, I’m not touching that one, so that the folks from FB and the data working group will have something to look at. So I suggest you don’t edit it, either, for now.

And yes, we are on the same page here. I was merely trying to explain what is what, i.e. there is the GPS track and the road is also created from a somewhat crude GPS track log, as opposed to roads drawn from imagery.

I have just gone through “our” village and looked at about 12 FB residential roads. I had to edit every single one, as most of them are either private driveways and/or unpaved. On the other hand, in the open fields they have drawn many tracks that seem pretty good. It’s the roads in built-up areas that we seem to have the most problems with. That’s why I’m suggesting that everything shorter than 2-400 m be reverted, as well as any deletions they have done.

I have emailed the DWG, but no response so far.

Kind regards,
Peter.

I have found and uploaded my GPS track for road 2243.

What I predicted would happen today happened to me, twice: I got facebookstreetmapped. A residential road that in reality was a track so small, my wife would not let me ride my bike through. The other end was private property, it seems. Had they used DGS instead of premium it was crystal clear what this was.

I have seen zero response from the DWG, I could not find a single email in the import mailing list; it looks as if we are yelling into emptiness.

I am now suspending any further activity in OSM in the faint hope that this may get resolved. Perhaps this will be the second time I will port all my data to another project.

Good night.

Tom, Dave, Russ and others in Chiang Mai I might forgot to mention. Shall we meet somewhere mid to end of the coming week? We could discuss above what the worst problems are and how to address this with Facebook’s editing team.

Also problems with regards to quantity. I am certain we find exceptions for every rule thinkable, but if there is a systematic problem we should address it.

Good idea, Stephan. Too bad it’s too far away for me.

I have received a reply from Frederik Ramm of DWG:

They are well aware of the issue, but are unable to act, unless

He suggests we get together and work out exactly what we see as problems and how we want FB to address them.

There are now several forum topics dealing with this problem. May I suggest one is chosen and stickied?

From where I am sitting I can’t tell what FB are now doing. Are they continuing as before, have they stopped? Everything I have dealt with so far is about a year old already.

The problems I have seen so far:

  1. Ways mis-aligned, probably due to lack of GPS tracks and/or reliance on single imagery.
  2. Lots of private driveways mapped as residential. (Should we even map such driveways at all?)
  3. The tracks vs. roads and paved vs. unpaved issue. We need to sort this out as a group once and for all.
  4. Massive amount of data already imported that needs reviewing/fixing/deleting.

The ball is now in our court.

Kind regards,
Peter.