Hacker News Reader

Comment by cheschire | Hacker News Reader

I really WANT to contribute to OSM but I can’t help but feel like I’m just making up for the foundations inability to implement automated imports from city data. Why are we still drawing house by house when many cities offer this data freely?

I’m sure it’s a complicated and well reasoned decision, I just don’t get the same feeling contributing to OSM that I do contributing to wiki.

Maybe more mainstream integrations like android auto will help remedy this by generating more awareness.

View on HN

I haven't paid close attention recently, but my understanding is that government datasets do get imported to OSM—but they're not perfect and still require some amount of clean-up. So it's never as simple as downloading a file from the local government and uploading it to OSM. Drawing house by house may not be the most efficient way to improve OSM in the long term, but it has lower skill requirements than sanitizing and validating a third-party dataset.

Have a look at the Dutch buildings and addresses in OSM. 99% of buildings in the Netherlands are in OSM with a very accurate outline, and 99.9% of valid addresses. This data is made available by the government under a compatible licence, and mappers use a plugin in JOSM (OSM's standalone editor) to import this data and perform updates. If we mappers find errors in this source data (and we do), then we report it back via an easy to use system the government set up for citizens reporting errors in their geo-data, and eventually the fixed up data gets back to OSM.

We (the Dutch mappers) use several government data sources, from WMTS overlays containing large scale physical objects (kerbs, water, artificial structures, street outlines), to satellite imagery, and half the time we can simply because someone in our community just went and asked for a compatible licence. It's not impossible.

If your government is unable to do this, ask your representatives. Governments benefit hugely from an active community of people who notice errors when they are working on an area in OSM, and the data is already something that exists for the benefit of the citizens whose tax dollars paid for it.

I'm a novice contributor, primarily using StreetComplete and the online editor. From some of what I've read, they've been burned by some imports in the past and are gunshy with mass imports.

> Why are we still drawing house by house when many cities offer this data freely?

It's not perfect, but there are some AI assistants which make this less tedious.

https://mapwith.ai/#13/-18.221/35.1573/0/55b (note: I believe Facebook is running this AI assistant.)

Mapwith.ai is a great tool that lets the server draw the shapes of the buildings and roads, though you still have to fill out the data by hand. The tool is maintained by Facebook, but the front-end[1] is a FOSS fork of OSM's RapiD web editor, and OSM is the type of project that strongly benefits from corporate contributions.[2]

[1] https://github.com/facebookincubator/RapiD

[2] https://joemorrison.substack.com/p/openstreetmap-is-having-a...

Unsuspecting people using AI generated datasets have ended up freezing to death.

  https://en.wikipedia.org/wiki/R504_Kolyma_Highway#Present
  https://www.youtube.com/watch?v=J46vglp6g98

Mapwith.ai works by tracing satellite imagery. No data is added into OSM until a person manually confirms that the road and building outlines match the satellite imagery, adjusting the outlines if necessary.

While the Google Maps incident you linked to was tragic, using AI to accelerate road tracing before the outlines are confirmed by a person shouldn't be any worse than having the person trace the outlines without assistance.

It makes little difference if the person is on another remote continent. We used IR imagery to trace forests in Africa, but incorrectly routing someone on an abandoned road in -50°C temperatures because "this route is 30 minutes faster" is just irresponsible. Roads through the African jungle can change every few years. That's why one should prefer local mapping resources to AI while navigating.

Yes, local mapping is always preferred over what OSM calls armchair/remote mapping.*

Satellite imagery, Mapwith.ai, and other tools like KartaView/Mapillary street imagery are still helpful for mapping areas that you have personally visited. They are optional data sources that you can use and adjust to your needs and observations.

I can understand using more caution when mapping hazardous areas and areas that you're unfamiliar with, but there are plenty of other use cases for these tools to accelerate mapping while maintaining accuracy. An available tool doesn't have to be and shouldn't be used in every situation.

* https://wiki.openstreetmap.org/wiki/Armchair_mapping

As others have said, imports from government data do happen. Here's an example from a random village in the Czech Republic: https://www.openstreetmap.org/way/103631362

It's tagged as coming from RÚIAN, our basic geodata registry. I'm not sure how these imports are organised, if they are fully automatic or semi-manual, OSM is unfortunately opaque and divided into small subcommunities like that.

Many cities here do have more specific data available, but every dataset has a different licence (often none at all) and a different format, so there's not much incentive to do an import, except for large-ish cities, but those are usually mapped pretty well already.

Sadly the data from my city does not include everything (even their own GIS system is missing heaps of stuff or even has wrong or ambigious data) so importing from the city still leaves quite a lot of gaps that has to be filled in other ways. One example is my home city. My city's GIS data has no idea of houses just sections. So you may have a property that has been divided up into several sub sections with their own houses. In many cases the city data only has data for the parent property and assigns all the relevant data e.g. house numbers to one location. This results in one large property having multiple street numbers assigned to one fixed location (usually in the middle of the whole lot). I've had to fix this (by splitting up the street numbers so they are each correctly mapped to each house) quite a few times in OSM because of bad city data. This issue is also present in the city's own GIS map so that's how I know it's the city at fault—it's actually not always possible to determine the correct house number using their own data. I can see why may actually be more efficient to just skip the city data and get better data via other means.

We did just that but the datasets we introduced are comercially licensed and they're also from 2010. We keep the osm_id for OSM data so we can distinguish between the two datasets when we update. It took a lot of work to merge those two datasets.

It's actually really hard to align different data sets, their schemas, and their quirks, licensing, restrictions, completeness, conflicts, etc. And with governments and local governments pretty much doing their own thing everywhere, it's a unique set of challenges for each city pretty much. And half of the problem is that they are not being required to actively work on this problem. The work related to that is hard work and somebody needs to pay for that.

OSM makes it possible for people to contribute their data but it doesn't necessarily solve all the issues. A lot of the data in it actually comes from various people importing data. All the low hanging fruit here has long been picked.

One of the reasons for not importing "avaliable" data is, if there's a license incompatibility between data source's license and OSM's Open Database License (ODbL).

https://wiki.openstreetmap.org/wiki/Import/ODbL_Compatibilit...

I saw a talk by a guy working with OSM and he explained that they have lots of issues with licensing. Government data is often not compatible with their wacky licensing requirements. They then need to plead with bureaucracy to relicense their data. That naturally doesn't go well

OSM isn't exactly free to use bc it requires attribution (so you can't actually just screenshot a map and use it). And you sometimes can't tack on that kind of requirement to government data.

And from what I gathered they're so deep with their current licensing model that they can't modify it.

But I'm a bit hazy on the details. Hope someone more authoritative could chime in

They do import data. I can see it all over Europe. But sometimes the imported data is misaligned, or lacks information.

Unlike Wikipedia, OSM does not have a notability requirement. You do not have to draw houses, or stuff that is on offical maps, you can map anything physical (and static). From individual trees to CCTV cameras.

E.g 1. There was a great map posted here a while ago that just mapped power lines.

E.g 2. And a worthy map that I saw here was a wheelchair map showing where the curbs were good and venues for disabled people etc.

If such datasource becomes available in future we will just create osmV2 from it and be done… I’m working on import for my home country(Slovenia) for buildings and addresses, and I like to think our GIS is well organised and maintained, and yet I find things like demolished buildings still in GIS or new buildings not added… Relocated buildinga/addresses…

Addresses for Slovenia are particularly bad in OSM. Try OpenAddresses as they use data from an official source (Geodetska uprava)

https://batch.openaddresses.io/data

https://github.com/openaddresses/openaddresses/blob/master/s...

At least in Massachusetts, much of the data are being imported from state GIS resources. House data, street data, etc.

Most of the mapping which needs to be done are enhancing the park data, etc.

OSM and OsmAnd are different things. OSM data is used by more than just OsmAnd.

OSM includes a lot more than what the public datasets provide, particularly if you're leaving paved ground. No other map has this much information about trails, fences, benches and signs. In some countries, it's the only map that includes petrol stations and banks.

I mean you could contact your local city and ask if you could license their data as Open Data Commons Open Database License. Actually I once did this for a Wikipedia article once to copy text from the local town's website.

The Dutch land registry dumps their stuff into OSM every now and then. I assumed this process was easy, but maybe it isn't?

They don't. Dutch mappers import building outlines in areas they are updating (or get requested to do so by other mappers) using government data. It's not an automated process that pushes data into OSM; there is a quality control step performed by the mapper importing the data.

Reply on news.ycombinator.com