My name is Salim from TomTom. I know that you have valid concerns from previous bad experiences with company-driven organised editing, and so I am coming to you with a very careful proposal.
Our goal from this project is to improve the map quality in Thailand, and we have set up three challenges for data improvements that I am hoping you will review. In creating these, we followed the guidelines you have provided. And, we have checked the quality of our leads carefully. The challenges proposed are the following:
In general it is highly appreciated to get more support in improving the data quality of OpenStreetMap in Thailand.
I am not certain why you had chosen specifically the above three challenges.
I think there might be easier ones to get started with. Have you reviewed the Osmose findings for Thailand? My server updates the data nightly.
Not connected highways is a major problem and I appreciate that you chose it as one of the major items to fix.
I had a quick look at some of the findings reported. You are giving remote mappers a hard time fixing as you have not restricted the errors to specific highway classes. So you have also many ways reported which are leading towards some destination and ending there.
I recommend to make this task easier for mappers by restricting the findings to not connected highways of specific types. As a start, exclude track/path/service kind of ways.
Also “nearly connects” are a good candidate of ways where mapper just missed to click right in the editor to connect highways.
area tags might be tricky to get right without sort of local knowledge/good aerials, but limited amount listed, so should be doable.
With the turn restriction challenge I am a bit skeptical on how well these could be solved without local knowledge.
I would agree with Stephan, I would also like all the organized mappers to coordinate a task list for Thailand so there’s no duplicated work, and the community can keep track of what’s happening. This can be done through a wiki entry.
But please be careful, some of those roads end because they’re actually not physically connected and there’s a thin wall acting as a barrier, or a gate, even tho it looks connected from satellite imagery.
Thank you Mishari and stephankn for your recommendation and observations.
Our engineering team came back with thorough analysis by using this interface. We are able to filter existing issues in OSMOSE by country and issue type, and as a result, our team were able to check nearly every crossing case mentioned by community users, there are approx. 3760 cases to be solved and are able to extract this data and create MapRoulette challenges for the community and/or organized editing, would you like us to do this?
Regarding projects and MapRoulette challenges tracking, would it be helpful for me to place an additional line about the MapRoulette challenge that we already published, on the WikiProject Thailand/Organised Editing Teams page?
Thank you for the information, I’ve looked through some of the almost junction items and it’s really difficult to ascertain if the two roads link up or not. In many of these cases there might be a wall or some other barrier. This needs to be fixed using a survey or street level imagery.
We have to distinguish “not connected” and “almost junction” checks.
Another incarnation of “drawing bugs” are the “crossing highways”. If such a crossing highway is also listed as “almost junction”, that means it is crossing a road and ends shortly after the road. Then it is sort of guaranteed a bug and it can be connected to the road. User was in this case overshooting the target.
In the other case, it depends.
If the one road is a residential, then it might well be that it is a dead end road in a gated community and it simply ends in front of a wall. Without higher resolution imagery this is nearly impossible to fix from remote, as already mentioned by Mishari.
So I would exclude residential “nearly connects” from remote tasks. Whether they are remote-fixable depends a lot on the available imagery and assessing the situation
I am not certain what happened to the almost junctions. 2022 there was a massive drop. Unless a large portion of data was deleted, this likely results from a change in the evaluation code on server side.
I am not certain, what the benefit is to have it exported into Maproulette instead of processing directly on Osmose frontend. Or even getting one of the osmose extracts and loading that into josm/iD.
Thank you @Mishari, as part of our checks, come across these violations and map issues and indeed it needs either hyper local knowledge and/or street view and/or ground truth surveying/checks, however in case of lack of source material it can still be marked as ‘Too Hard’ to solve.
Our engineering team conducted “consistency checks” in Thailand, this run was successful and below you can find results as of a couple of days ago(numbers indicated violations amount):
INVALID LANES TAG: 6
NOT CONNECTED HIGHWAY: 1472 (still in discussion and analysis with the Thailand community on another thread)
TAG AREA=YES ON OBJECT WITHOUT FEATURE TYPE: 142072
ROAD NAME GAP: 305
THIS MULTIPOLYGON IS A SIMPLE ONE: 173
POLYGON IS NOT CLOSED: 19
POLYGON HAS SELF INTERSECTION: 22
SPIKY BUILDINGS: 267
INVALID TURN RESTRICTION: 494
INVALID MINIROUNDABOUT: 87
TOTAL: 144917
We can share more challenges towards the community if they want, probably polygon related ones – i.e. Spiky buildings, and if you have violation or issues that you want support in, we can get into direct conversation to facilitate and coordinate that
Thank you, Salim
Thank you, I think we can slowly look through it for edge cases and see if there are any concerns about fixing these issues in a semi-automatic manner?
@stephankn: TomTom has used public instances of OSM QC tools such as Atlas and Osmose in addition to internal validation to detect potential map data issues.
In some cases it might overlap with some Osmose outputs, in other cases it might differ from the output as we do have our internal validations. if you still have any other question, please do send and will be delighted to discuss.
I have been looking into the NOT CONNECTED highway campaign and closed around 60 items.
Most were impossible to guess with imagery alone, others were remote agricultural tracks/paths where no standard router user would go.
Only found a few rare missing residential connections that were useful fixes.
What I saw instead is some beginner mappers joining the campaign and connecting roads that were impossible to guess from imagery, leading potentially to further low-quality entries.
So my question is how do these campaigns really help your business if road connections do not actually exist and route your potential customers nowhere?
I do a lot of ground surveys and improvements in northern Thailand, and what would be really helpful for the local community and any third-party end-users like yours would be ground survey information like GPS traces, streetview images to confirm geometries and tags.
Would you be willing to share some of your data to improve Thailand’s OSM data and consequently your own products?
PS: could you please remove the tasks that include the highway=services tag? Some of these stations are missing area=yes and show up wrongly as a highway item to be fixed. thanks.
Your feedback re: the not-connected highway campaign is really helpful and it has been echoed by other OSM communities. We are grateful. As a result of this advice, our team is making some big changes in our approach. Here is a quick summary:
a. we have completely paused all projects like this one, globally, so that we can better understand local concerns, as well as OSM logic. We made our MapRoulette challenges undiscoverable. That said, if you see one that is open, please let us know.
b. we will be making improvements to our data analysis, editing and quality control processes for our quality checks. When we do resume, we will share a detailed description of our methods with you and get your feedback before we take any action toward editing.
c. we will improve our instructions that we give to mappers, so that beginning mappers have good guidance.
I welcome your request for ground survey information such as GPS traces, imagery and other data, and I will take it to the team. Once I know what is available, I’ll let you know. It might take us 1-2 weeks as it is peak time for summer holiday and some of our team will be attending State of the Map in Florence, but we will get back to you. We would love to help in any way we can with source material.
Thanks for your patience with my reply. Talk soon,
I met Salim at SOTM in Florence. You can send me a direct message if there is a specific topic I should clarify in person.
TomTom is looking for our feedback on what kind of analysis would help us as a community in Thailand. Or other ideas on how they could help?
Please try to provide some feedback while SOTM is still running.
