Mapillary ORs

[amahon]

Mapillary creates and exposes a number of key Object Recognitions - traffic signs, semantic segmentation, and human generated tags. Today, iD editor provides support for viewing Traffic Signs as an overlay layer. Let's find a way to provide support to explore, discover, and leverage some of Mapillary's other OR data.

Semantic Segmentation

A couple of months ago, Mapillary released functionality to browse Semantic Segmentation data. Semantic Segmentation data represents imagery where we have identified the presence of a particular object. More details can be found on the Mapillary blog.

On Mapillary.com, it's possible to view Semantic Segmentation data on the map by filtering images to see only those containing a particular class, and to view the data overlaid onto each map image.

Viewing Semantic Segmentation on Map:

Viewing Semantic Segmentation on Map Image:

Semantic Segmentation Data can be accessed through Mapillary' Detection s API. Documentation can be found here. [TODO: add link]

It would be great to bring functionality to iD that allows OSM editors to rapidly identify images that contain objects of interest to then use those images to create map data.

Human Generated Tags

On Wednesday of this week, Mapillary will release a tool that allows humans to add tags to imagery. Think of this as similar to what the Semantic Segmentation, above, does, but with tags added by a human instead of a computer. To start, users will be limited to adding tags that come from the Mapillary Vistas corpus or from the OSM Map Features corpus.

A preview of the tagging tool can be found at https://elva-feat-tagging-11.mapillary.com/app/

Viewing Tags on Map, unfiltered:

Tag Filtering:

Tag displayed on Map Image:

Once launched, it will be possible to view the most commonly used tags for a geographic area, filter images for those containing particular tags, and view tags on each image.

Tag Data is not yet available via an API but work is in progress on an API that should mirror the Semantic Segmentation API described above. [TODO: add documentation for Tag API]

Similarly to Semantic Segmentation data, it wold be great to bring functionality to iD that allows OSM editors to discover imagery that contains objects of interest and then use that imagery to improve the map.

ORs

Since both Semantic Segmentation and Tag data follow a common model with the key differentiation being that one is computer generated, and one is human generated, it might be helpful to think of them similarly when envisioning additions to iD. With that in mind, let's think about what the functionality could accomplish.

• Allow a User to enable OR view.
• Allow a User to select between Semantic Segmentation, or Tagging view. This should limit the User's view to only ORs from one of the two types.
• Allow a User to view/filter/autocomplete a list of available OR classes for their currently active OR Type (Semantic Segmentation or Tagging).
• Allow a User to view the most common ORs, respective of currently active OR type, found in their viewport.
• Allow a User to view only images containing a particular class of OR.
• Allow a User to view the ORs overlaid onto map image containing ORs. This may introduce some complexity -- on Mapillary.com Semantic Segmentations are displayed a bit differently than Tags.

[amahon]

@bhousel - let me know what you think and where I can help out!

[bhousel]

Looks awesome @amahon 👍 It would be great to add some of these capabilities to iD.

My initial thought is to create a Mapillary info panel (see #4121 for this feature) that could expose some of this functionality for advanced users without complicating the UI for casual mappers.

[amahon]

@bhousel - I think that the info pane approach is legit!

A quick note regarding APIs -- the API for SS detections is likely going to change as the API for Tags is created. I'll keep you posted as things evolve - hoping to see some progress starting next week!