News for hackers. Stuff that matters.
Re:DeepOSM: Detect roads and features in imagery with neural... by MichailP on Tuesday May 31, 2016 @02:45PM (#11808979)
Offtopic, but does someone know a good reference and explanation for OSM tags? A lot of meaning is derived from tags, since the underlaying data structure is just nodes, ways and relations. And since the data is crowd sourced use of tags could be inconsistent, data could overlap and repeat itself. How someone makes a complex project like [1] using data like this?

[1] http://project-osrm.org/

Re:DeepOSM: Detect roads and features in imagery with neural... by skyebook on Tuesday May 31, 2016 @03:11PM (#11809224)
Taginfo[1] provides statistics on tags in use and there's a huge amount of documentation for tags (differences between tags, how to use them in certain situations, etc) in the wiki.

As an example, the page for highway tags[2] is quite rich.

[1] https://taginfo.openstreetmap.org [2] http://wiki.openstreetmap.org/wiki/Key:highway

Re:DeepOSM: Detect roads and features in imagery with neural... by Perixoog on Tuesday May 31, 2016 @03:55PM (#11809646)
Broadly the people who contribute, contribute because they want the data to be useful. The people who don't care about the data being useful aren't contributing.

Which leads to things like: https://wiki.openstreetmap.org/wiki/Quality_assurance

Re:DeepOSM: Detect roads and features in imagery with neural... by danpat on Tuesday May 31, 2016 @10:04PM (#11811653)
I'm paid to work on OSRM.

Obviously the data is imperfect, and there are lots of edge cases (https://github.com/Project-OSRM/osrm-backend/issues/2145). But overall, it works pretty well. There are regional disagreements about the highway hierarchy (primary/secondary/tertiary/etc), but as long as the structure is locally consistent, you generally get the routing results you expect (routes tend to major arteries, etc).

ETA calculation is a bit hit-and-miss, but the only way to correct this is to use real-world measurements.

The great thing is that if you find something weird, you can just fix it. This happens overwhelmingly more often that people deliberately introducing errors.