has released more of their computer vision road detection results for nearly all countries for """"import"""" into

And the data is bad.


Facebook is detecting walls as roads, because walls are lines too

Facebook detects some of these lines in a field as roads, others not. Because why not

These detected roads are just scribbles over a town.

I have no idea why thinks there is a street network in the middle of this lake

It took me a few minutes to find this crap in Facebook's data.

Tried to raise this with FB got the typical smarmy consdescing tone policy shite you expect from corporate america.

If you don't like what Facebook is doing, consider joing the the Foundation, to help set _some_ policy direction (though OSM is happily quite anarchistic).

I beat the FB candidate for the board by 1 vote a few weeks ago, so every little helps!


@BalooUriza There isn't one account, since regular OSMers use their own account and the RapiD software.

We _can_ block the editor 🤔

Facebook's initial sneaky imports were reverted. it's the only think they listen to

@rory I got the 'renew now to stay eligibile to vote' and the 'roughly a month minus a week 'til the election' email, but didn't see the one from the new voting platform.
It might be nice if the secretary also sent out an email saying 'voting open, check your inbox'.

@InsertUser There is the osmf-talk mailing list ( lists.openstreetmap.org/listin ) which new OSMF members aren't on by default, cause many aren't used to how discussion/talk/chat mailing lists like that work. If you subscribe to that, you'll get notices like that.

@InsertUser shame that you didn't get the ballot email..... You should have!

Can you contact the Membership Working Group ( wiki.osmfoundation.org/wiki/Me ) mwg@osmfoundation.org to check that your OSMF membership is all OK?

@rory after the results were announced I did manage to find the OpaVote emails lost in the newsletters and "infomail". Emails from the OSMF proper get hilighted so I actually see them. Using a novel domain as the only notice that voting is open means that they got buried for me. Of course, I can now add OpaVote noreply to my contacts, but who knows if we'll even be using them next time? A two line email from the OSMF itself after voting was open would have really helped.

@rory I don’t understand what you want to revert. Is this data in OSM, or detected and presented for human check in Rapid before editing?

@rory I don’t get it. What bad data is in osm that needs to be reverted?

@rory yes. “Revert bad data” isn’t specific. I’m honestly trying to understand the problem that you have encountered. Is it in rapid detections, or in osm changesets? Is it systematic in a specific place? Or widespread? Links to specific issues would be helpful.

@mikel “What bad data is there is OSM?” did sound like you thought all OSM data is perfect, to be fair. 😂
Maybe some day.... 🙂

@rory Sounds like what #Facebook wants with this project is for humans to train their algorithm. Not really to contribute to OSM.

@rory waves are just roads for surfers. it makes sense when you think about it

@rory just remember to stop QAing their shit for free after we've demonstrated how bad it is

@riking there is a long history if Facebook pushing pushing pushing, seeing as far as they can go, and only stopping with push back

@rory oops! You weren't supposed to get the Atlantis data

This is how you detect archeological sites >_> keep your eye out for Celtic burial mounds and bring your spade.


What is a wall, but a road from which you may fall?

@rory Devils Advocate here.

1. Leave the brand "Facebook" out of it; it's too loaded with negative issues that don't touch OSM at all. Just call it "AI" or "Neural net".
2. Any neural net needs training. Instead of calling "Revert", why not revert-in-a-way-that-trains-the-net ?

Humans have a very important role in mapping: verification! Training! Humans' role in drawing basic shapes is diminishing: it's something a computer is so much better at, once trained.

@berkes @rory
> why not revert-in-a-way-that-trains-the-net ?

Because Facebook don't provide that. It's all closed source.

> Leave the brand "Facebook" out of it; it's too loaded with negative issues that don't touch OSM at all.

When FB want to promote it, they keep mentioning "Facebook". It's not fair that they get marketing benefits, and don't like the negative PR

> Just call it "AI" or "Neural net".

I don't like calling it AI. That's marketing

@rory They don't *yet*.

I don't know how you conclude that it's *all closed source*. RapID is Open Source: ISC, same as upstream ID editor. github.com/facebookincubator/R

You *could* consider adding issues there to put this feature up. I think rallying up some people might give this some traction actually.
The folks at Teh Evil Facebook working on this are really nice persons. Very willing to cooperate with OSM communities; as seen on the last two SOTMs.

@berkes none of the machine learning algorithm software is open source

@berkes there might be friendly people working at Facebook. There were doctors who worked for tobacco companies. Doesn't mean the company isn't harmful

@rory We are discussing "Harmful for OSM", though, aren't we?

What would you suggest? Making OSM closed source, so we can keep out the baddies?

@rory what I'm saying, is that OSM, being open source, is by defenition welcome to all.

Including evil Facebooks or military using it to bomb people.

Facebook participating is a feature of open source. And IMO they are doing this part well.

If you want to bar Facebook from participating in OSM on principle of Facebook being evil, you effectively want to make OSM closed source.

@berkes (a) Facebook isn't a person. (b) just because I don't want lots of bad data added to OSM in a semi automated system doesn't mean I want "closed source" (c) Facebook is totally free to _use_ OSM data. I never said otherwise. Data contributions are a different thing. (d) And there is a growing concern that the current FSF & OSI definitions don't meet the needs of some people.

@rory it sounds like you’re referencing the conversation in Slack here. (linked below + screenshots) As you know, there’s nothing to revert in OSM since all of FB’s road detection results are offered for human review by any user of RapiD. None are automatically added.

You seem determined to mischaracterize FB’s work in new venues when you receive good faith responses about the way this data is created and used from Danil, Ben, and others working on the project.


@migurski Ummm.. Can we remain professional and not get personal?

Please don't cast aspirations against me. Don't some spaces (not this) have a "assume good faith" rule?

I don't know who works at Facebook on this, what they do at work is different from who they are as people. It would be wrong for me to blame them for all the bad things Facebook does, likewise its wrong for their good character to excuse Facebook.

Let's stick to the issues

@migurski plus, c'mon. "good faith" and "Facebook"? 😂 Let's be realistic & base things on past behaviour

Sign in to participate in the conversation
En OSM Town | Mapstodon for OpenStreetMap

The social network of the future: No ads, no corporate surveillance, ethical design, and decentralization! Own your data with Mastodon!