@imagico maybe I missed or misunderstood something on talk mailing list, but when discussion about Crimea started, Frederik wrote:

https://lists.openstreetmap.org/pipermail/talk/2018-October/081570.html

“the Crimea issue is currently being discussed in DWG.” “This policy is not likely to change any time soon.” And then wrote about tagging.

To me this does not sound like request for comments at all. Not even a possibility to have a say on the matter.

And then decision was made, NOT ANNOUNCED (I only found out about decision via Ukraine OSM twitter and later via this post) and that’s it.

This DWG decision makes a huge reputational damage to OpenStreetMap as a whole. Very sad.

The problem is that osmers got used to low quality imagery like the one from mikrosoft. Imagery should be fixed, not workarounds found. In some countries official precise imagery is available for vectorisation, so gps traces are only required for forest (invisible) routes, where gps would not get 30cm accuracy because of forest obstruction to gps signal. Also on the side of steep hills and mountains you would get consistant incorrect position of gps.

Thanks for sharing! Is there a difference in trained model prediction results from epochs 20-50? As train IoU look similar.

Thank you. ResNet decreases training time by 1/3!

Thank you for the answers. I ran through 8000 training and 1000 validation images (no augmentation) and IoU is still below 0.8. Of course images are quite different: buildings in forest, in rural areas, industrial buildings, urban buildings. And I haven’t done any hard-negative training. Training this without GPU takes almost a week :-) So yes, it would be useful to have access to trained models at least to compare the results, for others it could be much easier to take a model, run predictions and add missing data to OSM without the need to train and tune the model.

Thank you for sharing! How many training images do you think are necessary to have a decently trained model? Are there any plans to release some trained models (f.e. buildings, roads) to the public?

lonvia: brilliant! Putting way twice would also solve a problem of actual route length calculation.

So how do you expect non linear and non circular routes being mapped/sorted? Say the route of the shape Q - you start on the end of tail, get into a curcular part, then go back the tail route section to get back to start/end point? Or what has to be done with side trips? Because when its simple linear/circular, you can do as well as josm button “sort relation members” (except showing the start point).

Normalising with countries population would also be interesting thus giving proportion of mappers in the country.

Ok. So now it looks clear enough. People with little or no OSM experience make some fake* “analysis” just to give some “reference” pages just some days before the conference for garage hacker Zverik to say that “some people say craps.me is not SOO bad”. Good! Brilliant! Very “trump style”! :-D

From long term OSM mappers attitude: results are still the same. 50% of craps.me edits are revertable rubbish. And craps.me can be promoted to “the worst thing to happen to OSM ever”.

Bye

[*] edits from which region? what period? who analysed? what rules? :-D

Check the image in this blog post: https://blog.openmap.lt/2016/03/25/klaidos-baltijos-juros-regione/

You can clearly see the contours of Lithuania. That is because people in Lithuania take care and TAKE ACTION fixing errors instead of bragging or enjoying statistics of “increased contribution” which only increase the number of errors much more than increase the amount of useful data (like CRAPS.ME does).

That is there is already enough of QA done with results ignored. Fix that before doing other arbitrary “observations”.

And one more point. Subjective “checks” are pointless. Please add an objective and clear description (rules) what makes a change “bad”. HOW do you decide if a change is good or bad (and how do you decide it is “good”). Is it the error rule like one in keepright.at or osmose, is it a simple rule like “user=Komяpa” (actual rule which was used to find crap) or something else? Otherwise your findings are pointless.

Do not make simple problem larger. After ten years of practice:

• If the change brakes something seriously - you revert or fix it straight away and notify the user.
• If it is a small bug, you fix it and notify a user.
• If it is something with little impact - just notify a user (and take appropriate action after a day or so if the user does not react).

That is: QUALITY of the map is the priority and then you want a mapper to understand what he/she did wrong and why.

P.S. With one exception. If the change comes from CRAPS.ME user - revert the change and don’t even try contacting the user because its pointless. CRAPS.ME users do not realise they are changing OSM maps, they think they are changing “CRAPS.ME” “offline” maps, so they ignore messages from OSM as “irrelevant”.

Mapbox reacting to 78 changesets in 3 months!!!! only means they do not understand which changesets are good and which ones are damaging ones.

ЮрийИ is one of few who have made numerous edits, but he is not the only one with problem edits. From those in the first page of change stream with clearly rubbish: Totmorgan, Tetukas007. (I’m not counting notes, those are ok anyway, because somebody will review them anyway). In previous two months I have deleted/reverted more than 100 changesets from mapsme users so it is clearly more than those 4% stated in this blog post.

Very small number of changesets for good analysis. Maybe that is the reason why results are so different than noted by experienced mappers. It could also be that you simply did not have experience to understand where the problems actually are. You did not present a list of “ok” changesets.

My results from >200 changesets in Lithuania:

• 50% total rubbish - non existing points, adding russian names, adding addresses to names, adding artworks or castles for “my home” or “a place to meet”
• 40% more or less clear idea of what was being mapped but it was mapped incorrectly: position hundreds of meters away, duplicate objects etc.
• 10% more or less ok edits (tags have to be fixed, missing information added, but otherwise ok)

With only 10% of more or less good edits and 0% of edits which could go through without wasting any time of experienced mappers there is a big question if it is better to waste time fixing or it is more practical to simply revert mediocre maps.me changes.

I have nothing against new mappers and new ideas. But people should understand, that we do not start mapping each day from zero. So at least most widespread tags should not be rearanged without a very good reason and wide discussion. There are a lot of data consumers who depend on consistency of tagging. And in the case of water tagging change gives no benefit whatsoever so it should not had come through in the first place.

Zverik (at least his username) is registered in 2010-05, and in 2011-03 (after 10 MONTHS of participation) he makes a proposal to change one of the most widespread tags… I remember that proposal, but at that time I skipped it because it looked pointless and ungrounded (looks terrible now).

landuse=reservoir is included in JOSM presets. And most (if not all) power mappers use JOSM.

taginfo says that 367 133 objects are tagged with landuse=reservoir, 145 633 as landuse=basin That is half a MILLION of objects :-)

And it is a known problem when anybody (even people who do more talking than mapping) can have their own “voting” and change tag status…

This idea of reinventing the wheel in water tagging is totally pointless and onthologically useless. If 10 people voted “in favour” of abandoning landuse, it does not mean that we should all now change our tagging habbits. In some countries we have an agreement to mark specific objects with landuse tags and that agreement predates Zveriks proposal. This new water tagging gives absolutely no benefit I do not see why should we change our conventions and retag all objects. So we continue to tag with landuse and change newbies natural=water to landuse as well.