Working Group Minutes/EWG 2014-05-19

From OpenStreetMap Foundation


IRC nick Real name
RichardF Richard Fairhurst
zere Matt Amos


  • AWS / "big resources"
    • More ideas:
      • "Temporal tiling", or something OWL-like which is based on brute force rather than efficient indexing.
      • Use Mechanical Turk to identify areas of misclassified or poor spatial accuracy data (i.e: probably TIGER). gravitystorm popped up later to remind us of the "Rapid Assessment Tool": [1].


16:30:43 <zere> welcome! previous minutes (there's a lot of them this week)
16:31:57 <zere> here's the list:
16:32:14 <zere> if anyone sees any problems with any of them, please let me know.
16:33:24 <zere> the only topic i have for the agenda is a continuation of the "big resources" discussion
16:33:31 <zere> #topic big resources
16:33:37 <zere> ideas from the previous meeting:
16:33:44 <zere> mkl raised the idea of automatic imagery analysis, trying to compute if "something" is present on a tile which can then be compared to OSM to target areas where there's something in the real world, but nothing in OSM.
16:33:50 <zere> Another suggestion is multi-lingual maps (would require much more rendering power than currently available).
16:34:01 <zere> Rendering the full history over time as a series of tiled videos rather than still images, so you can pan anywhere in the world and view any period of the map evolution there.
16:34:06 <zere> Global routing analysis to spot problems with the data.
16:34:12 <zere> GPX/routing analysis along the lines of ris' "That Shouldn't Be Possible".
16:34:18 <zere> A system that stores all object's history in S3 as static files.
16:34:22 <zere> Offline Map tiles/Geocoding/Routing to the Amazon Kindle Fire or something similar.
16:35:20 <zere> anyone have any additional ideas?
16:37:33 <RichardF> Mikel's idea is interesting.
16:43:48 <zere> it is, but i'm not aware if anyone has written anything similar to do it already, or if it would be a "from scratch" effort.
16:48:20 <RichardF> I tend towards the idea that our big challenges are (a) getting more mappers, (b) making our mappers more productive. I'm finding it hard to see how to apply Amazon's resources to that. but I may be being unimaginative!
16:49:00 <RichardF> though - do you think there's any mileage in applying Amazon to "fixing OWL"?
16:53:26 <zere> do you mean the code? or possibly trying to turn the code into something similar that would run on some sort of hadoop / massive cluster?
16:56:00 <RichardF> perhaps the latter? basically - if OWL (or equivalent) is stalled, then maybe we can get it going again by applying massive amounts of Amazon CPU power to it, rather than trying to build something lean and efficient that runs on OSMF hardware.
16:57:23 <zere> yeah, it's possible that a "temporal" tiling scheme could make it amenable to brute force solution.
16:58:17 <zere> i can't quite wrap my brain around how, though. ;-)
16:58:42 <zere> but i'm sure if someone wanted to work on it then it could be a possibility.
16:59:09 <RichardF> well, yes, if Amazon could supply us with a few thousand robo-developers, that would be better :)
17:00:55 <zere> robo-developers, who also understood OSM, its data model and all the other tacit knowledge we take for granted :-)
17:01:22 <zere> we haven't had any ideas using mechanical turk...
17:03:07 <zere> it's probably not of significant worth to throw someone unprepared into maproulette...
17:03:33 <RichardF> weelll......
17:04:01 <RichardF> for a while I've been wondering about a HotOrNot UI for the vast number of untouched, misclassified TIGER ways
17:04:37 <RichardF> i.e. highway=residential which are farm tracks at best, highway=track which don't really exist, and so on
17:05:15 <zere> you mean, show OSM alongside/over aerial imagery and ask if they align?
17:05:29 <RichardF> ask if the tagging is right, yes
17:05:48 <RichardF> i.e. "yes, this is really a road", "no, this is a farm track", "no, it doesn't exist", "I dunno, show me the next"
17:06:20 <RichardF> the great advantage over MR is that the editing is so simple you don't actually need to invoke an external editor (and therefore the Turker doesn't need to learn how to edit)
17:07:23 <zere> but presumably, you assume the Turker can tell what a farm track looks like from the air, and/or that the farm track or road is sufficiently close to ground truth that it can be identified
17:07:31 <RichardF> most of these are in what Steve memorably described as South B-mf--k, Ohio, so our chances of getting a survey done in a sensible amount of time are pretty low.
17:07:56 <RichardF> yes, indeed. but IMX you can solve 75% of the misclassified ways from imagery.
17:07:57 <zere> not much chance of someone correctly identifying a road if the TIGER geom is a few 100 yards away and not on-screen
17:09:13 <RichardF> so, typical example (this was the first place I zoomed into):
17:09:14 <RichardF>
17:09:30 <RichardF> those are demonstrably not residential roads
17:09:43 <zere> certainly would be interesting to run... are you sure a previous version of maproulette didn't do something similar? i just feel like soemone has written this before...
17:09:44 <RichardF> the geometry is averagely broken for TIGER
17:10:13 <RichardF> I don't _think_ I've seen it for MR. I know I've mentioned it to Serge in the past and he hasn't said "oh we did that already"
17:13:42 <zere> cool, i'll add it to the list.
17:26:23 <zere> looks like that topic has run out of steam.
17:26:33 <zere> anyone want to discuss anything else?
17:26:36 <zere> #topic AoB
17:31:56 <zere> ok. thanks to everyone for coming. hope to see you again next week :-)