OSMPH/Tabang-AI

Mapping roads in Siquijor, PH with RapiD

Opened this issue · 13 comments



Siquijor Island, OpenStreetMap Cyclemap Style

Using AI-based detected roads, the goal of this project is to improve road coverage for the whole island of Siquijor.

We will use this ticket to coordinate mapping efforts. If you have questions, please comment on this ticket.

To contribute:

Changeset comment (OSMCha):

Mapping roads in Siquijor, PH. https://github.com/OSMPH/Tabang-AI/issues/10 #osmph #tabangAI

Partner organization: Local government of Lazi, Siquijor through @kinleah

Daghang salamat!

cc @kinleah @govvin @seav

Some users are reporting they are not seeing the tiles and detection. This likely due to adblocker.
"One workaround would be to configure the ad blocker to allow the path /maps/ml_roads on facebook.com, which is the endpoint from which we serve both the satellite imagery and the ML roads data for RapiD." Reference convo in SlackUS

From @ralleon
Screenshot 2019-09-11 16 39 17

Some users are reporting they are not seeing the tiles and detection. This likely due to adblocker.
"One workaround would be to configure the ad blocker to allow the path /maps/ml_roads on facebook.com, which is the endpoint from which we serve both the satellite imagery and the ML roads data for RapiD." Reference convo in SlackUS

From @ralleon
Screenshot 2019-09-11 16 39 17

One particular Extension on Chrome Browser observed to cause the problem is "Video Downloader PLUS". Just remove or disable this Extension.

For this particular Siquijor project, I can't see any AI-assist purple lines in RapiD on my Chrome or Firefox browsers, even with Ad Blocker & Video Downloader removed.

Disabling Video Downloader used to work in the last experiment in another project. But now...
image

Chrome: Ad Blocker & Video Downloader removed. All remaining extensions Disabled.
image

Same particular task from @maning
image

For this particular Siquijor project, I can't see any AI-assist purple lines in RapiD on my Chrome or Firefox browsers, even with Ad Blocker & Video Downloader removed.

Disabling Video Downloader used to work in the last experiment in another project. But now...
image

Chrome: Ad Blocker & Video Downloader removed. All remaining extensions Disabled.
image

Same particular task from @maning
image

Could you post the task id/url?

Panning around the previously validated areas (without any assistance from AI), I saw a good number of what looked like false positive & false negative road edits and validations.

Example of validated task (note the curved road in the center):
image

Note the blue arrows: those were either 'rural' footways, alleys or driveways, yet were not mapped; but was reviewed as ok by the validator. Or they may have been presumed as not part of objectives of the task.
image

Another example of validated task: (a case of 'possible but not probable')
An unclassified road was mapped under tree cover. Opinion: both the mapper and the validator failed to notice the presence of tree lines (shown by multiple purple arrows) which may indicate a fence or a private property line. I can't find substantial gap on the tree line (or on the green plants after the tree line). So it's not logical to gamble and assume that an automobile can pass there. This indicates that our validator was in a hurry (or had no idea about what to look for).
Approx location: 9.1313207, 123.6301104
image

Another validated task: Validator was not very particular on improving both road geometries and alignments (because it's not part of the goal of task?).
Opinion: A traced path that has consistent offset, yet following the shape of the road on imagery is normal (especially from old edits).
But for a validator not being able to notice and correct those 10 meters off-center portions shows: (1) he/she was in a hurry at the expense of quality validation, (2) needs to improve sensitivity to details/context, (2) learn how to use the map scale bar. 10 meters off is huge! you can fall off a cliff with that. :-)
Again, in a hurry, the validator missed what looked like a junction near school that leads to an alley/unpaved road. (Not sure if this alley was tagged by AI - not showing in my browser).
Tip: majority of public school compounds (by observation) have some sort of wide frontage (to an access road). So it's fair to guess that that alley is the supposed missing frontage (considering the weird angle of school ground and the general orientation of its buildings).
Approx location: 9.1495019, 123.6351008
image

Last weekend, I ran JOSM validators for all the road data in Siquijor. There are several issues in the data (changeset):

  • overlapping roads - this is likely a road detection on roads that were already mapped.
  • wrong road class - For example, primary where it should be a lower class road. In some cases, there is inconsistent tag on long segments (+--track--+--secondary--+--track--+). This is common on edges of the task.
  • unconnected road to the nearest junction - node is very close to the junction but its is not connected.

Most of these issues can be flagged by the built-in validator. We should remind new mappers to not ignore the validator messages before uploading.

Screenshot 2019-09-23 10 12 37

Screenshot 2019-09-22 17 08 13

Partial status as of 2019-09-23:

  • 56% - completed mapping
  • 21% - validated
  • 80% - of completed grid have at least 1 road
  • 20% - of completed grid is time "wasted" because no road exists.
  • 11% of roads added were from AI detections.

overlapping roads - this is likely a road detection on roads that were already mapped.

Correction, iD/RapiD do not have an overlapping road validator yet.

All pending road mapping tasks for SIG is complete. Efforts should now be directed to validation.

When validating tasks using RapiD, set the following to the issues panel:

  • check - Everything
  • where - Everywhere

This will catch all possible errors in the area your are validating.

Screen Shot 2019-10-02 at 11 41 22

Mapping and validation completed. 💥
Screen Shot 2019-10-23 at 11 45 36

Next actions

  • Compile summary stats and issues encountered @maning